CN120602448A - Instant messaging message dynamic shielding method, device, computer equipment and storage medium - Google Patents

Abstract

The present invention discloses a method, apparatus, computer equipment and storage medium for dynamic shielding of instant messaging messages. The method includes: obtaining the content of complaint instant messaging messages submitted by users; extracting the content of complaint instant messaging messages to obtain key information; classifying the content of complaint instant messaging messages according to the key information to obtain classification results; judging whether the current user has a historical complaint-related portrait based on the classification results; if the current user has a historical complaint-related portrait, querying the historical complaint-related portrait and identifying the nature of the complaint in the current complaint instant messaging message content; and implementing a corresponding shielding strategy based on the nature of the complaint. By implementing the method of the present invention, rapid processing of complaint instant messaging messages is achieved without the need for manual searching of mobile phone numbers and selecting shielding types one by one, thus shortening the complaint processing time and improving processing efficiency. This technical solution can be applied to the fields of finance and medical health.

Description

Instant messaging message dynamic shielding method, device, computer equipment and storage medium

Technical Field

The invention relates to the technical field of instant messaging message shielding, in particular to an instant messaging message dynamic shielding method, an instant messaging message dynamic shielding device, computer equipment and a storage medium.

Background

In the fields of finance, medical health and the like, instant messaging messages have become an indispensable information exchange tool in daily life of people. However, with the popularization of the use of instant messaging messages, the problems of instant messaging message harassment, spam instant messaging messages, malicious complaint instant messaging messages and the like are increasingly prominent, and great trouble and inconvenience are brought to users. In order to cope with this problem, instant messaging message screening technology has been developed, and it is intended to intercept and filter unnecessary instant messaging messages by means of technology, thereby improving the use experience of users.

Currently, in practical application of instant messaging message screening technology, most enterprises or institutions still rely on traditional manual processing methods, especially in the scenario of processing user complaints about instant messaging messages. Specifically, after receiving a complaint instant communication message submitted by a user, a processor needs to manually search for a related mobile phone number in the system, and select a corresponding shielding type one by one according to complaint content, such as high-frequency complaint, malicious complaint and the like. The operation flow is complex, consumes long time and seriously affects the complaint treatment efficiency.

The inefficiency of the manual processing mode is mainly characterized in that firstly, the processing time is increased by manually searching mobile phone numbers, particularly when a large number of complaints are processed, the step becomes an efficiency bottleneck, secondly, the screening type is selected one by one to require a processor to carry out careful analysis and judgment on each complaint instant communication message, so that the workload is increased, misjudgment or omission is easy to be caused by human factors, and finally, the manual processing mode is difficult to realize real-time or near real-time screening operation, so that a user still can continue to be harassd or maliciously invaded by the instant communication message after complaints.

In addition, with the increasing number of instant messaging messages and the increasing diversity of complaint types, the traditional manual processing mode has difficulty in meeting the requirements of modern enterprises on efficient and accurate complaint processing. Therefore, developing an automatic and intelligent dynamic instant messaging message shielding method to replace the traditional manual processing mode and improve the efficiency and accuracy of complaint processing has become a technical problem to be solved in the technical field of instant messaging message shielding.

Disclosure of Invention

The invention aims to overcome the defects of the prior art and provide an instant messaging message dynamic shielding method, an instant messaging message dynamic shielding device, computer equipment and a storage medium.

In order to achieve the above purpose, the present invention adopts the following technical scheme:

The instant messaging message dynamic shielding method comprises the following steps:

Acquiring complaint instant communication message content submitted by a user;

Extracting the content of complaint instant messaging information to obtain key information;

Classifying complaint instant messaging contents according to the key information to obtain classification results;

judging whether the current user has a history complaint related portrait or not based on the classification result;

If the current user has a history complaint related portrait, inquiring the history complaint related portrait, and identifying the complaint property of the current complaint instant messaging content;

and implementing a corresponding shielding strategy according to the complaint property.

The invention also provides a dynamic shielding device for the instant messaging messages, which comprises:

The acquiring unit is used for acquiring the complaint instant communication message content submitted by the user;

the extraction unit is used for extracting the content of the complaint instant communication message to obtain key information;

The classifying unit is used for classifying the complaint instant messaging content according to the key information so as to obtain a classifying result;

The judging unit is used for judging whether the current user has a history complaint related portrait or not based on the classification result;

the inquiring and identifying unit is used for inquiring the history complaint related portraits and identifying the complaint properties of the contents of the instant messaging messages of the current complaints if the history complaint related portraits exist in the current user;

and the implementation unit is used for implementing the corresponding shielding strategy according to the complaint property.

The invention also provides a computer device which comprises a memory and a processor, wherein the memory stores a computer program, and the processor realizes the method when executing the computer program.

The present invention also provides a storage medium storing a computer program which, when executed by a processor, implements the above method.

Compared with the prior art, the method has the advantages that the quick processing of the complaint instant communication messages is realized through the steps of automatically acquiring, extracting, classifying, judging the user history complaint images and the like, the manual searching of mobile phone numbers and the one-by-one selection of shielding types are not needed, the complaint processing time is greatly shortened, the processing efficiency is improved, in addition, the complaint instant communication messages are classified by utilizing the key information, and the characteristics of the complaint instant communication messages can be judged more accurately by combining with the user history complaint related images for recognition, so that a more accurate shielding strategy is implemented, and the possibility of misjudgment and omission is reduced.

The invention is further described below with reference to the drawings and specific embodiments.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

Fig. 1 is a schematic diagram of an application scenario of an instant messaging message dynamic screening method according to an embodiment of the present invention;

Fig. 2 is a flow chart of a dynamic screening method for instant messaging messages according to an embodiment of the present invention;

fig. 3 is a schematic block diagram of an instant messaging dynamic screening apparatus according to an embodiment of the present invention;

fig. 4 is a schematic block diagram of a computer device according to an embodiment of the present invention.

Detailed Description

The following description of the embodiments of the present invention will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are some, but not all embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

It should be understood that the terms "comprises" and "comprising," when used in this specification and the appended claims, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof.

It should be further understood that the term "and/or" as used in the present specification and the appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes such combinations.

Referring to fig. 1 and fig. 2, fig. 1 is a schematic view of an application scenario of an instant messaging message dynamic screening method according to an embodiment of the present invention. Fig. 2 is a schematic flow chart of an instant messaging message dynamic screening method according to an embodiment of the present invention. The dynamic screening method of the instant messaging messages is applied to a server, the server and the terminal conduct data interaction, the steps of automatically acquiring, extracting key information, classifying, judging historical complaint images of users and the like are achieved, quick processing of the complaint instant messaging messages is achieved, manual searching of mobile phone numbers is not needed, screening types are selected one by one, complaint processing time is greatly shortened, processing efficiency is improved, in addition, the key information is utilized to classify the complaint instant messaging messages, relevant images of the historical complaint of the users are combined to conduct identification, the properties of the complaint instant messaging messages can be judged more accurately, accordingly, accurate screening strategies are implemented, and misjudgment and omission possibility are reduced.

Fig. 2 is a flow chart of a method for dynamically shielding instant messaging messages according to an embodiment of the present invention. As shown in fig. 2, the method includes the following steps S110 to S160.

S110, acquiring complaint instant communication message contents submitted by a user;

Specifically, the instant communication message includes a short message or a multimedia message, etc. In an instant messaging service platform or related business system, a special complaint instant messaging receiving interface is developed. The interface can be integrated with various channels such as an instant messaging gateway, an APP built-in complaint function module, a webpage form and the like, and can receive complaint instant messaging messages submitted by users in real time. For example, through an API interface with an instant messaging gateway, when a user sends a complaint instant messaging message to a specified number, the instant messaging gateway pushes instant messaging message content to the interface. The received complaint instant message content is stored in a database for subsequent processing and analysis. A relational database (e.g., mySQL) or a non-relational database (e.g., mongo db) may be employed, with the selection being made based on traffic requirements and data size.

That is, through multiple channel integration, the complaint instant messaging content submitted by the user can be ensured to be timely and comprehensively obtained, and omission of complaint information caused by channel limitation is avoided. In addition, complaint instant communication messages are stored in a database, so that centralized management and maintenance of data are facilitated, and a data base is provided for subsequent extraction, classification and other operations.

In one embodiment, the obtaining the content of the complaint instant messaging message submitted by the user includes:

and the complaint information submitted by the user is uniformly received by the channel complaint entrance of the official APP, the instant messaging information platform and the customer service hotline so as to obtain the contents of the complaint instant messaging information.

Specifically, the method is used for interfacing with the complaint entrance of the official APP channel, and an interface special for receiving complaint information is developed in the back-end service of the official APP. The interface needs to define a clear data format, for example, it is specified that the complaint information includes fields such as user ID, mobile phone number (optionally, if APP is bound to mobile phone number), complaint content text, complaint time, etc. A simple and clear complaint page is designed at the front end of the APP, and a user can fill in the complaint content in the page and submit the complaint content. The front end sends complaint information filled in by the user to the back end server in a specified format through calling a complaint interface developed by the back end. And docking the front end of the APP with the rear end interface, and performing comprehensive test. The test content comprises complaint information submission under normal conditions and processing abnormal conditions (such as network interruption, data format errors and the like), so that the stability and the reliability of the interface are ensured. And after receiving complaint information sent by the APP, the back-end server stores the complaint information into a database. A relational database (e.g., mySQL) may be selected to store complaint information, and a unique identifier may be assigned to each complaint record, facilitating subsequent queries and management.

And (3) docking the channel complaint entrance of the instant messaging platform, namely setting a special complaint receiving number in the instant messaging platform, wherein the number is used for receiving the complaint instant messaging message sent by the user. And interfacing the instant messaging platform with an instant messaging processing system within the enterprise. Real-time receiving and forwarding of instant messaging messages are achieved through an API or a protocol provided by the instant messaging gateway. For example, when a user sends a complaint instant message to a complaint receiving number, the instant message gateway forwards the instant message content to an instant message processing system within the enterprise. After receiving the instant communication message, the instant communication message processing system analyzes the content of the instant communication message. Since the instant messaging content is usually plain text, it is necessary to identify key information therein, such as contact information (if not explicitly provided) that the user may have implicit in the instant messaging, complaint core content, and so on. The analysis can be performed by technical means such as regular expression, character string matching and the like. And storing the analyzed complaint information into a database and correlating the complaint information with user information (if any). If the instant messaging message does not explicitly contain the user identity information, the user information can be perfected through subsequent further communication or data comparison with other systems of the enterprise.

And (3) abutting against a customer service hotline channel complaint entrance, namely recording complaint calls of the user in a customer service hotline system. The recorded speech is converted into text content using speech recognition techniques (e.g., APIs provided by the science fiction, hundred degree speech recognition, etc.). And extracting key information from the converted text content, and identifying key information such as complaint objects, complaint reasons and the like similarly to the processing of the instant messaging message content. This may be achieved by keyword extraction, named entity recognition, etc. methods in Natural Language Processing (NLP) technology. The extracted key information is integrated with basic information (such as caller number, user ID, etc. which can be obtained by customer service system) of the user to form complete complaint information. And storing the integrated complaint information into a database, and uniformly managing the complaint information with the complaint information of other channels, so that the subsequent inquiry and analysis are convenient.

And (3) unified receiving and integrating, namely uniformly receiving and distributing complaint information from different channels such as APP, an instant messaging platform, a customer service hotline and the like by using a message queue (e.g. RabbitMQ, kafka) or a middleware technology. Complaint information of different channels is firstly sent to a message queue, and the message queue forwards the message to a subsequent processing module according to a certain rule. In a subsequent processing module, complaint information from different channels is integrated. And associating complaint information of the same user according to the user identification (such as a mobile phone number, a user ID and the like) to form a complete user complaint history record.

That is, by interfacing with complaint entrances of a plurality of channels such as an official APP, an instant messaging platform, a customer service hotline, etc., a complaint mode that a user may use can be covered on the whole, ensuring that any piece of complaint information is not missed. The complaint can be conveniently submitted by young users who like to use the APP or old users who are used to complaint through instant messaging or telephone, and the acquisition rate of the complaint information is improved. In addition, a plurality of convenient complaint ways are provided for the user, and the user can select the most suitable mode to complain according to own preference and actual conditions. For example, if the user finds out the problem in the conversation, the complaint can be carried out through a customer service hotline without switching other equipment or applications, so that the complaint experience of the user is improved. In addition, after the complaint information submitted by the user is uniformly received, all the complaint information is stored in the same database, so that centralized management of data is realized, and thus, the complaint information can be conveniently inquired, counted and analyzed, such as counting the complaint number, complaint type distribution and the like of different channels, and data support is provided for decision making of enterprises.

S120, extracting complaint instant messaging content to obtain key information;

Specifically, the NLP technology is utilized to analyze and process the complaint instant messaging content. Firstly, word segmentation is carried out on an instant messaging message text, sentences are divided into single words, then part-of-speech tagging is carried out, grammar types of each word are determined, and then key entities involved in the instant messaging message, such as complaint objects (company names, product names and the like), complaint reasons (quality problems, service attitudes and the like), complaint time and the like, are extracted through a named entity recognition technology. The extracted key information is further processed and converted into feature vectors which can be used for classification. For example, the complaint reasons can be encoded, different complaint reasons are mapped into different values, and the complaint time is standardized so as to meet the requirement of model input.

That is, through NLP technology, key information can be accurately extracted from complaint instant messaging information, and powerful support is provided for subsequent classification and complaint property identification. In addition, the extracted key information is converted into the feature vector, so that the standardization processing of the data is realized, and the subsequent machine learning or deep learning model processing is facilitated.

In an embodiment, the extracting complaint instant message content to obtain key information includes:

preprocessing complaint instant messaging message content, including removing irrelevant characters, advertisement links and unified text formats, so as to obtain preprocessed content;

Specifically, extraneous characters are removed by using regular expressions to identify and remove extraneous characters in instant messaging message content. For example, special symbols (e.g., #, $,% etc.), extra spaces, tabs, line breaks, etc. And writing a regular expression mode, such as r '[ ] w\s ]' (matching non-alphanumeric and blank characters), and replacing the matched characters with blank character strings by traversing the instant messaging message text, thereby achieving the purpose of removing irrelevant characters. According to the service requirement, defining an allowed character set, and only retaining the characters belonging to the character set in the instant messaging message content. For example, if only Chinese, english, numerals and common punctuation marks need to be reserved, a set containing the characters can be constructed, then the instant messaging text is traversed, and characters not in the set are filtered.

Ad links are removed-ad links typically have a specific format, such as beginning with http://, https:// or contain common domain name suffixes (e.g.,. Com,. Cn, etc.). These link patterns are matched by regular expressions, and the matched links are deleted from the instant messaging message content. A blacklist of advertisement links is maintained, including known advertisement website links. When preprocessing instant communication message content, checking whether the instant communication message contains a link in a blacklist, and if so, removing the link.

Unified text format-all letters in instant messaging message content are uniformly converted into upper-case or lower-case form so as to prevent subsequent processing problem caused by inconsistent upper-case or lower-case. For example, the character string is processed using lower () or upper () methods of Python. The uniform coding format of the instant communication message content is ensured, and the instant communication message content is converted into UTF-8 coding. If the instant messaging message content has different coding formats, a coding conversion library (such as chardet library in Python for detecting codes, and then converting by using an encode () method and a decode () method) can be used to convert the instant messaging message content into a unified coding format.

And extracting keywords, semantic analysis and user emotion from the preprocessed content, and integrating to form key information.

Specifically, extracting keywords, namely performing word segmentation on the preprocessed instant messaging message text, and then counting the occurrence frequency of each word. Words with high occurrence frequency and certain representativeness are selected as keywords. The word may be segmented using existing segmentation tools (e.g., jieba parts of speech library of Python) and then the word frequency may be counted using a dictionary or counter.

Semantic analysis, namely converting each Word in the preprocessed instant messaging message text into a Word vector, wherein Word2Vec, gloVe and the like are commonly used Word vector models. The entire instant messaging text is represented as a text vector for semantic analysis by a weighted average of word vectors or other aggregation. A pre-trained word vector model or a self-trained word vector model may be used. And carrying out similarity calculation (such as cosine similarity) on the text vector of the instant messaging message text and a predefined semantic category vector, and judging the semantic category to which the instant messaging message text belongs according to the similarity. For example, semantic categories such as "product quality problem" and "service attitude problem" may be defined, and the main semantics of the instant messaging message may be determined by calculating the similarity between the text of the instant messaging message and these category vectors.

And (3) carrying out emotion analysis on the user, namely constructing an emotion dictionary which comprises positive emotion words, negative emotion words and neutral emotion words, and assigning corresponding emotion scores to each word. Matching the preprocessed instant messaging message text with an emotion dictionary, counting the number and sum of emotion scores of positive emotion words and negative emotion words, and judging emotion tendencies (such as positive, negative or neutral) of the user according to a certain rule. Using the labeled instant messaging data set with emotion labels, a machine learning model (e.g., naive bayes, support vector machine, deep learning model, etc.) is trained to classify the emotion. And inputting the preprocessed instant messaging message text into a trained model, and outputting the emotion type of the user by the model.

And integrating the extracted keywords, the semantic analysis result and the emotion analysis result of the user to form key information. For example, keywords may be presented in a list, semantic analysis results are represented as specific semantic categories, user emotion analysis results are represented as positive, negative, or neutral, and then these information are combined into a structured data structure (e.g., dictionary, JSON object, etc.) to form key information.

That is, noise information such as irrelevant characters, advertisement links and the like is removed, and the text format is unified, so that the data quality of the complaint instant messaging content can be improved, and the interference and errors in the subsequent processing process are reduced. For example, removing the advertising links may avoid mistaking the advertising content as part of the complaint content, thereby improving the accuracy of the key information extraction. The preprocessed text has uniform format, is clean and tidy, and is more beneficial to the subsequent operations of keyword extraction, semantic analysis, user emotion analysis and the like. For example, unifying the case and coding format may avoid algorithm processing anomalies due to format inconsistencies. In addition, the extracted keywords can directly reflect the core content in the complaint instant messaging message, and help enterprises to quickly locate the problem points of user complaints. The meaning of the complaint instant communication message can be deeply understood by semantic analysis, and the complaint instant communication message is classified into specific semantic categories, so that enterprises can grasp the type distribution of the complaint of users macroscopically, and a basis is provided for formulating a targeted solution. The user emotion analysis can enable enterprises to know the emotion states of the users during complaints, whether anger, dissatisfaction or disappointment and the like. According to the emotion of the user, enterprises can adopt different coping strategies, and user satisfaction is improved. By integrating the key words, semantic analysis and user emotion to form key information, a comprehensive and accurate user complaint portrait is provided for enterprises. Enterprises can formulate more scientific and reasonable complaint treatment strategies and product improvement schemes according to the key information, and the service quality and the competitiveness of the enterprises are improved.

S130, classifying the complaint instant messaging content according to the key information to obtain a classification result;

Specifically, according to the service requirements and the data characteristics, a proper classification model is selected, such as a decision tree, a Support Vector Machine (SVM), a random forest, a Convolutional Neural Network (CNN) and the like. Training the model by using the marked historical complaint instant messaging data, and adjusting parameters of the model to accurately identify the complaint instant messaging messages of different categories. The extracted key information feature vectors are input into a trained classification model, the model classifies complaint instant messaging messages according to the learned features and classification rules, and a classification result is output. The classification results may include high frequency complaints, malicious complaints, general complaints, business consultations, and the like.

That is, through machine learning or deep learning model, automatic classification of complaint instant messaging messages is realized, classification efficiency and accuracy are greatly improved, and workload and subjectivity of manual classification are reduced. Through a model trained by a large amount of historical data, complaint instant messaging messages of different categories can be accurately identified, and a reliable basis is provided for the follow-up shielding strategy implementation.

In an embodiment, the classifying the complaint instant messaging content according to the key information to obtain the classification result includes:

Matching keywords in the key information with a predefined classified keyword library to obtain a keyword matching result;

Specifically, the business scope, the common complaint types and the past complaint data of the enterprise are deeply analyzed, and keywords corresponding to different complaint types are determined. For example, if the instant messaging message contains keywords such as "arrears", "repayment", "overdue", the instant messaging message is primarily classified as "furcation type" complaints, and if the instant messaging message contains keywords such as "marketing", "preferential", "activity", the instant messaging message is classified as "marketing type" complaints. For certain specific scenarios or business requirements, more accurate keyword matching may be required. For example, for instant messaging messages involving lawsuits, it may be desirable to match more specific keywords of "litigation," "law," "court," etc. Keywords are collected from multiple channels, such as enterprise internal documents, industry reports, complaint analysis by competitors, etc. And (5) sorting and de-duplication the collected keywords, and ensuring the accuracy and the integrity of a keyword library. Meanwhile, a corresponding complaint category label is allocated to each keyword.

Simple string matching algorithms, such as exact or fuzzy matching, are used. The fuzzy matching allows certain character differences, such as using an edit distance algorithm to measure the similarity of two character strings, and the matching is considered successful when the similarity exceeds a certain threshold. Traversing all keywords in the key information and matching the keywords with the classified keyword library. Recording the complaint category matched with each keyword, and counting the number of the keywords matched with each complaint category.

Carrying out overall analysis on the semantics in the key information to obtain a semantic analysis result;

Specifically, a suitable pre-trained semantic understanding model is selected, such as BERT, GPT, etc. The models are pre-trained on large-scale text data, and have strong semantic understanding capability. And fine tuning the pre-training model according to complaint data and business requirements of the enterprise. Training the model by using the annotated complaint data set, and adjusting parameters of the model to enable the model to better understand the semantics of the complaint instant messaging message and classify the complaint instant messaging message into the correct complaint category.

The text content in the key information is input into a fine-tuned semantic understanding model, and the model converts the text into a vector representation, wherein the vector contains semantic information of the text. And carrying out similarity calculation (such as cosine similarity) on the text vector and a predefined complaint category vector, and predicting the complaint category to which the text belongs according to the similarity. For example, the similarity between the text vector and the category vector such as the commodity quality problem, the logistics distribution problem and the like is calculated, and the category with the highest similarity is selected as the semantic analysis result.

Carrying out emotion analysis on the emotion of the user in the key information to obtain an emotion analysis result;

Specifically, a large amount of complaint instant messaging information data with emotion labels is collected, and the emotion tendency of each instant messaging information is marked manually, such as positive, negative or neutral. These annotation data will be used to train the emotion analysis model. Emotion analysis models are constructed using machine learning algorithms (e.g., naive bayes, support vector machines) or deep learning algorithms (e.g., recurrent neural network RNNs and variants LSTM, GRU thereof). Dividing the marked data set into a training set and a testing set, training the model by using the training set, and evaluating the performance of the model by using the testing set.

And extracting features of texts in the key information, such as lexical features (such as emotion dictionary matching), syntactic features (such as sentence structure analysis) and the like. And inputting the extracted features into a trained emotion analysis model, and outputting emotion tendency types of the text, namely emotion analysis results, by the model.

And combining the keyword matching result, the semantic analysis result and the emotion analysis result to obtain a classification result.

Specifically, different weights are allocated to the keyword matching result, the semantic analysis result and the emotion analysis result according to business experience and data analysis results of enterprises. For example, if keyword matching is more accurate in the past classification, higher weight may be given, while emotion analysis results may not distinguish some complaint categories more heavily, giving lower weight. And carrying out weighted summation on the keyword matching result, the semantic analysis result and the emotion analysis result according to the respective weights by adopting a weighted summation method, and determining the final complaint category according to the summation result. Voting mechanisms may also be employed, for example, if two or more of the keyword matching, semantic analysis, and emotion analysis results point to the same complaint category, then that category is taken as the final classification result.

That is, the contents of complaint instant messaging messages and the intention of users can be more comprehensively understood by combining three-dimensional information of keyword matching, semantic analysis and emotion analysis. The information of each dimension provides different visual angles, and supplements each other, so that the classification accuracy is improved. For example, keyword matching may be capable of locating some obvious complaints quickly, semantic analysis may be capable of understanding the overall meaning of the text deeply, emotion analysis may reflect the emotional state of the user, and comprehensive information may be used to determine the complaint category more accurately. A single classification approach may have limitations that can easily lead to erroneous decisions. The combination of the multidimensional information can effectively reduce the misjudgment. For example, if the matching of keywords is only relied on, some instant messaging messages containing similar keywords but different actual semantics can be misclassified, and after semantic analysis is combined, the true meaning of the instant messaging messages can be more accurately judged, so that misjudgment is avoided. In addition, the complaint instant messaging information of the user has various expression modes, and the conditions of incomplete keywords, fuzzy semantics, complex emotion expression and the like can exist. By combining multiple analysis methods, these different expression modes can be better adapted. For example, even if keyword matching is not ideal, semantic analysis and emotion analysis can still provide valuable information, help determine complaint categories, and enhance the robustness of the classification system to different complaint instant messaging messages. With the development of business and the variation of customer complaint habits, the content and form of complaint instant messaging messages also vary. The classification method combining the multidimensional information can better adapt to the changes, and the higher classification performance is kept by adjusting the weight and optimizing the model.

S140, judging whether the current user has a history complaint related image or not based on the classification result, if the current user does not have the history complaint related image, constructing a user image, and performing conventional shielding processing on the content of the instant communication message of the current complaint;

Specifically, a unique user identifier, such as a mobile phone number, a user ID, etc., is extracted from the complaint instant messaging message. And according to the unique user identification, inquiring in the user portrait database. The user portrait database stores information such as historical complaint records, complaint frequency, complaint type preference and the like of the user. If the historical complaint related record of the user exists in the query result, the current user is judged to have the historical complaint related portrait.

That is, by inquiring the user identification and the portrait database, whether the current user has a history complaint related portrait or not can be rapidly judged, and a foundation is provided for the follow-up complaint property identification and shielding policy implementation. In addition, the establishment of the user portrait database is beneficial to analyzing the historical complaint behaviors of the user, knowing the complaint habits and characteristics of the user and providing support for personalized treatment.

In one embodiment, the determining whether the current user has a history complaint related representation based on the classification result includes:

Extracting user unique identification information from complaint instant messaging information content, and inquiring a historical complaint record of a current user in a database according to the user unique identification information to obtain an inquiry result;

In particular, the content structure and common format of complaint instant communication messages are being studied in depth, determining locations that may contain user unique identification information. For example, the beginning, end, or particular passage of an instant messaging message may contain information such as the user's cell phone number, order number, member account number, etc., which may typically be the unique identification of the user. Aiming at the unique identification information of different types of users, corresponding regular expressions are compiled for matching extraction, the regular expressions r '1[3-9] \d {9}' can be used for matching mobile phone numbers, order numbers can be composed of numbers and letters, specific length and format rules exist, and regular expressions are compiled for extraction according to the rules. For some unique identification information that is not directly presented in a distinct format, such as a user nickname, etc., extraction may be performed in conjunction with natural language processing techniques. For example, named Entity Recognition (NER) algorithms are used to identify words in instant messaging messages that may represent nicknames of users.

The setting database is provided with a table which is specially used for storing user complaint records, and the table comprises relevant fields such as unique identification information fields (such as mobile phone numbers, order numbers and the like) of users, complaint time, complaint content, processing results and the like. And constructing a corresponding database query statement according to the extracted unique identification information of the user. For example, in SQL, the history complaint record of the user may be queried using the select_ FROM complaint _ records WHERE phone _number= 'extracted cell phone number'. After the query statement is executed, a query result is obtained. The query result may be a data set containing a plurality of records, each record representing a historical complaint of the user. The query results are formatted, for example, into a list, dictionary, or other data structure that is easy to handle later.

And screening the historical complaint records similar to the current complaint type from the query result according to the classification result, so as to distinguish whether the current user has the historical complaint related portrait.

Specifically, according to business characteristics of enterprises and complaint processing experiences, similarity rules among different complaint types are defined. For example, for an e-commerce business, a commercial quality problem complaint and an after-market service problem complaint may have some similarity in some cases because the after-market service problem may originate from the commercial quality problem. Similarity rules are quantified, for example, by assigning similarity scores to different types of complaints. If the two complaint types have higher relevance in business logic, a higher similarity score may be given, whereas a lower score is given.

Traversing the queried historical complaint records of the user, and comparing the similarity between the complaint type of each record and the classification result of the current complaint. And calculating the similarity score of the current complaint type and each historical complaint type according to a predefined similarity rule. And setting a similarity threshold, and when the similarity score of the history complaint record exceeds the threshold, considering that the record is similar to the current complaint type, and screening the record.

And counting the number of the screened similar historical complaint records. If the number is greater than zero, the current user is indicated to have a history complaint record similar to the current complaint type, namely, a history complaint related image exists, and if the number is zero, the current user is indicated to have no history complaint record similar to the current complaint type, and no history complaint related image exists. If the history complaint related portrait exists, features of the portrait, such as average processing time of similar history complaint records, satisfaction degree score of users and the like, can be further extracted, so that richer information is provided for subsequent complaint processing.

That is, by extracting the unique user identification information and querying the historical complaint records, it is possible to quickly locate whether the user has similar complaint problems. If the related historical complaint records exist, complaint treatment personnel can directly refer to the past treatment experience and result, so that repeated problem analysis and investigation are avoided, and the complaint treatment efficiency is greatly improved. In addition, businesses may find some potential problems and trends through analysis and screening of historical complaint records. For example, if a user is found to complain about the quality of the same type of commodity many times, it may indicate that the commodity has defects in design or production, and the enterprise may take measures to improve in time, so as to avoid the reappearance of similar problems.

In one embodiment, if the current user does not have a history complaint related portrait, constructing a user portrait, and performing conventional shielding processing on the content of the current complaint instant messaging message;

Specifically, data related to user complaints is collected from a plurality of channels, including, but not limited to, current complaint instant messaging content, basic information of the user (e.g., age, gender, region, etc., if available), and interaction records of the user with the enterprise (e.g., customer service consultation records, feedback records, etc.). The data may be acquired in real-time or periodically by interfacing with various business systems of the enterprise. And cleaning and preprocessing the collected data to remove noise data, repeated data and invalid data. For example, complaint instant messaging message content is processed for word segmentation, stop word removal, etc., for subsequent feature extraction and analysis.

And carrying out natural language processing on the content of the current complaint instant messaging message, and extracting the characteristics such as keywords, topics, emotion tendencies and the like. May be implemented using bag of words model, TF-IDF algorithm, emotion analysis algorithm, etc. For example, whether a complaint instant message is positive, negative or neutral, and the extent of the negative emotion is determined by an emotion analysis algorithm. Users with similar features are grouped using a clustering algorithm (e.g., K-Means clustering). Through cluster analysis, complaint behavior patterns and characteristics of different user groups can be found, and references are provided for constructing user portraits. The constructed user representation is stored in a database for subsequent querying and use.

Conventional masking rules are formulated for all new users (users who do not have a history complaint image). For example, screening complaint instant communication messages that contain sensitive words (e.g., abuse, threats, spurious information, etc.), or screening complaint instant communication messages that are sent during a particular time period (e.g., early morning). And the shielding rules are stored in a rule management system, so that operations such as adding, modifying and deleting are convenient to perform. The rule management system can provide a graphical interface to allow an administrator to intuitively manage the mask rules. When the system receives the current complaint instant communication message, the shielding rule is used for detecting and matching the instant communication message content in real time. And if the content of the complaint instant communication message meets the condition of the shielding rule, shielding the instant communication message. The method of shielding processing can be to mark the instant communication message as a shielding state without subsequent processing and forwarding, or to discard the instant communication message directly without recording in the system.

That is, for users who do not have history complaint related portraits, constructing user images can fill in the data blank, so that enterprises can more comprehensively understand the characteristics and behaviors of users, and the enterprises are helped to establish a more perfect user management system, so that basic data support is provided for subsequent user service, marketing and complaint processing. In addition, the conventional shielding treatment is carried out on the content of the current complaint instant communication message, so that the interference of the ineffective complaint instant communication message can be reduced, the complaint treatment personnel can concentrate more on treating valuable complaints, the efficiency and the quality of the complaint treatment are improved, and the operation cost of enterprises is reduced.

S150, if the current user has a history complaint related portrait, inquiring the history complaint related portrait, and identifying complaint properties of the current complaint instant messaging message content;

Specifically, the historical complaint related portrait information of the user is extracted from a user portrait database, including the type of historical complaint, the frequency of complaint, the resolution of complaint and the like. And combining the classification result of the current complaint instant messaging message and the historical complaint portrait information to identify the complaint property. For example, if the current complaint belongs to a high-frequency complaint category and the history complaint frequency of the user is high, the complaint may be identified as a malicious repeated complaint, and if the current complaint is similar to the history complaint type and the processing result of the history complaint is not satisfied, the complaint upgrade with dissatisfied processing result may be identified.

That is, by combining the historical complaint image and the current complaint classification result, the complaint properties can be more comprehensively and accurately identified, and erroneous judgment caused by single information can be avoided. In addition, according to different complaint properties, a personalized processing mode can be adopted, and user satisfaction and complaint processing efficiency are improved.

In one embodiment, if the current user has a history complaint related representation, querying the history complaint related representation and identifying complaint properties of the current complaint instant message content, including:

positioning a database for storing historical complaint images of the user according to the unique identification information of the user;

Specifically, the database architecture is planned according to the business scale and the data size of the enterprise. For small enterprises or cases with smaller data volume, a centralized database, such as MySQL, can be used to store all the historical complaint portrait data of users on one server, so that the management and maintenance are convenient. For large enterprises or scenes with huge data volume, a distributed database (such as CASSANDRA, MONGODB distributed clusters) is more suitable, and can store data on a plurality of nodes in a scattered manner, so that the performance and expandability of data storage and query are improved. A clear and easily identifiable name, such as user_ complaint _profile_db, is set for the database storing the user's historical complaint images. At the same time, a unique identifier is assigned to the database in the database management system so that the system can locate it quickly and accurately.

An index is established in the database for the unique user identification information. The indexing can greatly increase query speed because the database engine can quickly locate records containing specific user identification information through the indexing. For example, in MySQL, an INDEX may be created for a user identification information field using the CREATE INDEX statement. If an enterprise has multiple databases or tables of data, and the user identification information is scattered across different places, an association map may be created. The table records the correspondence between the unique user identification information and the database or data table storing the historical complaint images thereof. When the historical complaint image of the user needs to be queried, the system firstly queries the associated mapping table, determines a target database or a data table, and then performs subsequent operation.

Performing a query operation, and retrieving historical complaint portraits of the current user from a database, wherein the historical complaint portraits comprise portrayal dimension information of complaint frequency, complaint type preference and sensitivity score;

specifically, if a relational database (e.g., mySQL, oracle) is used, the corresponding SQL query statement is written. For non-relational databases (e.g., mongo db), its specific query syntax is used. After the query result is obtained, the returned data is analyzed. According to the data format (such as JSON, XML, etc.) returned by the database, the data is converted into a data structure which is easy to process inside the program, such as a dictionary or an object in Python. And verifying the analyzed data to ensure the integrity and accuracy of the data. Checking whether each dimension information is empty or out of a reasonable range, if abnormal data is found, corresponding processing can be performed, such as logging, prompting errors or adopting default values.

Matching the content of the current complaint instant messaging message with each dimension information in the historical complaint image, and identifying the image characteristics related to the content of the current complaint instant messaging message to obtain the complaint property.

Specifically, in order to evaluate the relationship between the current complaint and the frequency of the historical complaint more accurately, a time window is set. For example, consider a history of complaints over the past year or half. Counting the complaint times of the user in the time window, and carrying out association analysis with the current complaint. If the current complaint occurs during a period of time when the frequency of user complaints is high, it may mean that the user is more sensitive or dissatisfied with the current problem. And calculating the change trend of the current complaint frequency and the historical complaint frequency. If the complaint frequency of the user suddenly increases, it may indicate that a new problem occurs or the user's satisfaction with the service is greatly reduced, and if the complaint frequency is stable or reduced, it may indicate that the user's response to the current problem is more conventional.

And establishing a mapping relation between the complaint keywords and the complaint types. And performing word segmentation and keyword extraction on the content of the current complaint instant messaging message, and then determining the type possibly belonging to the current complaint according to the mapping relation. And calculating the similarity between the current complaint type and the preference of the historical complaint type. The complaint type may be expressed as vectors using an algorithm such as cosine similarity, and then similarity between the vectors is calculated. If the similarity is higher, the current complaint is more consistent with the historical complaint type preference of the user.

And carrying out emotion analysis on the content of the current complaint instant messaging message, and judging the emotion tendency (such as positive, negative or neutral) of the user. The emotion analysis results are correlated with sensitivity scores in the historical complaint representation. For example, if the emotion analysis results show that the user's emotion is very negative and the sensitivity score of the user in the historical complaint representation is high, then the nature of the current complaint may be considered more severe. The threshold of sensitivity score is dynamically adjusted based on business needs and user feedback of the enterprise. For example, during special periods (e.g., during promotional campaigns), the user's demand for services may be higher, at which time the sensitivity threshold may be appropriately lowered in order to more timely discover and handle problems that may cause user dissatisfaction.

Different weights are assigned to dimensions such as complaint frequency, complaint type preference, and sensitivity score. The weights may be determined based on business experience, data analysis, or expert evaluation. For example, if the business considers that the sensitivity score has a greater impact on the nature of the complaint, a higher weight may be given. And carrying out weighted summation on the matching results of the dimensions according to weights by adopting a weighted summation method, and determining the property of the current complaint according to the summation results. A decision tree algorithm can also be used to gradually judge the complaint properties according to the matching conditions of each dimension. For example, it is first determined whether the complaint frequency exceeds a threshold, and if so, the complaint type preference and sensitivity score are further determined, ultimately resulting in the complaint nature.

That is, by combining information of multiple dimensions of complaint frequency, complaint type preference, sensitivity score, and the like, the nature of the current complaint can be more comprehensively and accurately identified. The one-sided performance possibly brought by single-dimensional analysis is avoided, and the accuracy of complaint property judgment is improved. The historical complaint image of each user is unique, and personalized judgment can be realized by identifying the complaint property based on the image. For example, for users with high complaint frequency and high sensitivity scores, even if the problem of the current complaint appears not large, it may be identified as important complaints, so that it is handled more timely.

S160, implementing a corresponding shielding strategy according to the complaint property.

Specifically, a shielding policy library is established, which contains shielding policies for different complaint properties. For example, for repeated complaints, a strategy of temporarily shielding the complaint authority of the user for a period of time can be adopted, and for the updating of complaints with unsatisfactory processing results, special persons can be arranged to follow-up, and shielding conditions can be properly relaxed. In addition, a corresponding masking policy is selected and executed from a policy repository based on the identified complaint nature. The shielding operation can be realized by modifying the system configuration, calling the shielding interface and the like, and meanwhile, the related information of the shielding operation, such as shielding time, shielding reasons and the like, is recorded.

That is, implementing the corresponding shielding strategy according to the complaint properties can realize accurate shielding, avoid the false shielding of normal complaint users, and effectively reduce the interference of malicious complaints and repeated complaints. In addition, by establishing a strategy library, the shielding strategy can be flexibly adjusted and optimized conveniently so as to adapt to the changes of different service scenes and requirements.

In one embodiment, the complaint properties include high frequency complaints, malicious complaints, or special types.

Specifically, for high-frequency complaint users, all complaint instant communication messages of the high-frequency complaint users can be shielded for a certain time, for malicious complaint users, complaint instant communication messages of the high-frequency complaint users can be permanently shielded or complaint authorities of the high-frequency complaint users can be limited, and for special types of instant communication messages (such as a prompt receipt instant communication message required to be sent by a central office), compliance limits (such as sending at least one message per month) can be set.

That is, by shielding instant messaging messages of high frequency complaints and malicious complaints users, enterprises can concentrate limited complaint processing resources on more valuable and more reasonable complaints, improving the efficiency and quality of complaint processing. Avoiding wasting manpower, material resources and time due to handling a large number of invalid complaints. For special type instant messaging messages, compliance limit is set to ensure that the special type instant messaging messages are sent and processed according to specified requirements, so that the sending and receiving of the important instant messaging messages are preferentially ensured, and the management level of enterprises on special services is improved. In addition, for high-frequency complaints and malicious complaints, shielding or limiting measures are adopted, so that the interference to normal users can be reduced, and the normal users are prevented from being influenced by malicious complaints. In addition, the complaints are reasonably treated and compliance with compliance limiting rules, so that the honour of the interests of the enterprise to the user and compliance with laws and regulations are shown, the good image of the enterprise is built, and the trust and loyalty of the user to the enterprise are enhanced.

In one embodiment, the implementation of the corresponding shielding strategy according to the complaint properties further comprises monitoring the change of the complaint behaviors of the user and dynamically adjusting the shielding strategy according to the user portrait and the behavior change.

Specifically, time windows of different lengths are set to monitor changes in complaint frequency, such as weeks, months, quarters, etc. For example, the number of complaints of the user during the week is counted once a week and compared to the number of complaints of the previous weeks. An algorithm is written to calculate the rate of change of complaint frequency. For example, the ratio of the number of complaints in the current time window to the number of complaints in the previous time window is calculated, and if the ratio is smaller than a certain set threshold (e.g., 0.8), the frequency of complaints is considered to be reduced. Using natural language processing techniques, emotion analysis models are trained based on a large amount of complaint text data to classify the complaint text emotionally (e.g., positive, negative, neutral). When a new complaint instant communication message arrives, real-time evaluation is carried out on the content of the instant communication message by using a trained emotion analysis model, and the complaint emotion of the user is judged. Comparing the evaluation result with the historic complaint emotion, and if the proportion of negative emotion is reduced, considering that the complaint emotion is improved.

The time interval for updating the user representation periodically, such as once a month, is set. And updating each dimension (such as complaint frequency, complaint type preference, sensitivity score and the like) in the user portrait according to the latest complaint behavior data, consumption data, historical interaction data and the like of the user. When the complaint behavior of the user changes significantly (such as the complaint frequency decreases suddenly and greatly), a mechanism for updating the user portrait in real time is triggered. The latest user behavior information is reflected in the user portrait in time so as to more accurately adjust the shielding strategy. Rules engines (e.g., drools) are used to define the adjustment rules for the masking policy. The rule engine can automatically judge whether the shielding strategy needs to be adjusted and how to adjust according to the data of the user portraits and behavior changes. For example, a rule "when the user complaint frequency decreases by 50% and the complaint sensitivity score decreases by 20%, the masking time is shortened by half". When the policy adjustment rule is formulated, factors of multiple dimensions, such as complaint frequency, complaint emotion, complaint sensitivity, shielding time and the like are comprehensively considered. And (3) giving different weights in different dimensions, carrying out weighted calculation to obtain a comprehensive evaluation result, and determining whether to adjust the shielding strategy according to the evaluation result.

Specific conditions for unmasking are configured in the system, such as reduced user complaint behavior (complaint frequency reduced by a certain proportion and number of complaints below a certain threshold), reduced complaint sensitivity (sensitivity score below a certain value), masking time reaching a certain period (e.g. 2 years), etc. The system monitors behavior data of the user in real time, and when the releasing condition is met, the shielding releasing process is automatically triggered. For example, complaint behavior data of all shielded users is checked every day at regular intervals to determine whether a release condition is satisfied. The shielding state of the user is updated in the database, and the shielding mark is changed into normal so that the user can normally receive the instant communication message. At the same time, the time and reason of unmasking are recorded for subsequent querying and auditing.

That is, the dynamic unmasking function can timely adjust the masking policy according to the actual behavior change of the user, so as to avoid excessive masking of the user. When the complaint behaviors of the user are reduced and the releasing conditions are met, the shielding is timely released, so that the user can normally receive the instant communication message, and inconvenience and dissatisfaction of the user are reduced. In addition, through dynamic adjustment of the shielding strategy, enterprises can concentrate more resources on complaints which really need to be processed, the efficiency and the quality of complaint processing are improved, unnecessary processing on users with improved behaviors is avoided, and labor, material and time costs are saved.

For example:

Customer complaints are a common and well-handled problem in the financial field. Different types of complaints may reflect different needs and risk conditions of the customer. For example, high frequency complaints may suggest that customers are generally not satisfied with financial services or that there are defects in the service flow, malicious complaints may present reputation risks and operational disturbances to financial institutions, and specific types of complaints (e.g., prosecution instant messaging-related complaints involving regulatory requirements of the central office) require strict compliance with compliance regulations. The dynamic shielding method of the instant messaging messages can help financial institutions to process complaints more efficiently, optimize resource allocation and reduce risks.

The content of the complaint instant communication message submitted by a client through an official APP is that' the credit card of your bank is calculated reasonably in a stage interest, I pay more per month, I complain for a plurality of times, and I complain is not solved any more.

The bank receives the complaint instant communication message uniformly through the channel complaint entrance of the official APP. Preprocessing the content of the complaint instant messaging message, removing irrelevant characters (such as the part of punctuation marks, which does not affect core meaning, can simplify processing), unifying text formats (such as converting full-angle characters into half-angle characters, and the like), and obtaining the preprocessing content, wherein' the credit card stage interest calculation of your bank is too unreasonable, i am complaint is detained for a plurality of times every month, and the complaint is not solved.

Keywords such as "credit card staged interest", "irrational", "multi-withhold", "complaint several times over are extracted from the pre-processed content.

The whole semantic meaning shows that the calculation mode of the bank credit card stage interest is seriously discontented by the client, and the client considers that money is deducted and complaints are repeated.

The emotion analysis shows that the emotion of the client is more excited and belongs to negative emotion from the expressions of 'unreasonable', 'no resolution of me and complaint' and the like.

And integrating key information, namely integrating key words, semantic analysis and emotion of a user to form key information, namely, solving the problem that the customer is not satisfied with credit card stage interest calculation, and realizing multiple complaints and emotional agitation.

And matching the keywords in the key information with a predefined classified keyword library. For example, the classified keyword library may have keywords related to credit card business and complaints, such as "credit card", "interest", "complaint", and the like, and these keywords can be matched.

And combining semantic analysis results to determine that the complaint is related to credit card staged service and belongs to the category of dissatisfaction of customers on service rules and cost calculation.

And (5) emotion analysis results, namely analyzing emotion as negative emotion.

And (3) synthesizing the classification result, namely synthesizing the analysis, and classifying the complaint instant messaging message into a credit card stage interest complaint-customer dissatisfaction class.

And extracting user unique identification information, such as a mobile phone number of a client or a client number in a banking system, from the complaint instant messaging message content or a system record for receiving the complaint. And inquiring the historical complaint record of the current user in the database according to the unique identification information of the user. Suppose that the query results show that the customer has complained about the credit card stage interest problem 5 times in the past year. And screening historical complaint records similar to the current complaint type from the query result according to the classification result 'credit card stage interest complaint-customer dissatisfaction', and finding out that a plurality of related records exist, so as to show that the current user has the historical complaint related portrait.

And locating and storing a database of the historical complaint images of the user according to the unique identification information of the user. And executing a query operation, and retrieving the historical complaint portrait of the current user from the database. Assuming that the portrait dimension information includes complaint frequencies of 1-2 times per month (average over the past year), complaint type preferences are mainly focused on credit card stage interest and commission, with a sensitivity score of 8 points (10 points full, higher points indicating more sensitivity). And matching the content of the current complaint instant messaging message with each dimension information in the historical complaint image. The current complaint content is related to the credit card stage interest, the customer emotion is excited, and the high-frequency complaint with the risk of the upgrade complaint can be further judged by identifying that the complaint is high-frequency complaint (because the average complaint is 1-2 times per month in the past year and relatively frequent) and the customer sensitivity is higher.

Because the complaint is high-frequency complaint, the bank can set to shield all complaint instant messaging messages within a certain time (such as 3 months) (but needs to ensure that a special channel or personnel follow-up treatment is performed during the shielding period so as to avoid further upgrading of the customer problem. Meanwhile, special customer service personnel are arranged to actively contact with the customer, so that the problem of the customer service personnel is thoroughly known, the problem of unreasonable credit card stage interest calculation is solved as soon as possible, customer satisfaction is improved, and real complaints of the customer are avoided.

Through the instant messaging message dynamic shielding method, banks can quickly classify and process complaints, and more resources are concentrated on solving the actual problems of customers, instead of being bothered by a large number of repeated complaints. By timely identifying high-frequency complaints and customers at risk of upgrading the complaints and taking corresponding measures, further worsening of customer problems can be avoided, and reputation risks and supervision risks faced by banks are reduced. Although the high-frequency complaint users are shielded, special person follow-up processing is arranged, so that the customers feel the importance of banks on the problems, and the customer satisfaction and loyalty are improved.

As another example:

In the medical health field, patient complaints are an important source of feedback for medical institutions to understand quality of service and improve workflow. However, there are also cases of high-frequency complaints, malicious complaints, etc. of some patients, which may waste resources of medical institutions, interfering with normal working order. Meanwhile, some special types of complaints (such as complaints related to patient privacy protection and emergency medical assistance) require special treatment. The instant messaging message dynamic shielding method can help medical institutions to efficiently process complaints, reasonably allocate resources and improve service quality.

A complaint instant communication message submitted by a patient through a hospital official APP is received by a large hospital, and the contents are that the registration system of your hospital is too junk, the expert number is not hung for a plurality of times, the system is crashed each time, the complaint of me is returned for a plurality of times, and then the user needs to find the media exposure- "

The hospital receives the complaint instant communication message uniformly through the complaint entrance of the official APP. Preprocessing the content of the complaint instant messaging message, removing irrelevant characters (such as garbage, which is emotional but does not affect the expression of core meaning, can keep core semantics, mainly processing irrelevant characters such as punctuation and the like), and unifying text formats (such as converting full-angle characters into half-angle characters) to obtain preprocessing content, wherein the problem that a registering system of your hospital hangs up too much garbage and cannot hang up expert numbers for several times is that the system crashes me complaints for several times each time, and then the me needs to find media exposure.

Keyword extraction, namely extracting keywords such as a hospital, a registration system, a hanging-off expert, a system crash, complaints for a plurality of times and a media finding exposure.

The whole semantic meaning is that patients are not satisfied with a hospital registration system, the registration is failed for a plurality of times, the system is crashed, the complaints are not solved for a plurality of times, and the idea of finding media exposure is provided.

Emotion analysis, namely, the emotion of a patient is excited and belongs to negative emotion from the expressions of 'too much garbage', 'how much me needs to find media exposure', and the like.

And integrating key information, namely integrating key words, semantic analysis and emotion of a user to form key information, namely, discontent of a patient on a hospital registration system, multiple registration failures and system breakdown, multiple complaints not solved and emotion activation.

Keyword matching, namely matching keywords in the key information with a predefined classified keyword library. The classified keyword library may have keywords related to hospital registration and complaints, such as a hospital registration system, an expert number, a complaint, and the like, and the keywords can be matched.

And combining semantic analysis results to determine that the complaint is related to the hospital registration system fault and belongs to the category that the patient is not satisfied with the system service.

And (5) emotion analysis results, namely analyzing emotion as negative emotion.

And (3) synthesizing classification results, namely synthesizing the analysis, and classifying the complaint instant communication message into a hospital registration system fault complaint-patient dissatisfaction type.

And extracting user unique identification information, such as a mobile phone number of a patient or a patient number in a hospital system, from complaint instant messaging message content or a system record for receiving complaints.

And inquiring the historical complaint records, namely inquiring the historical complaint records of the current user in a database according to the unique identification information of the user. Suppose that the query results show that the patient has complained about the registering system fault problem 4 times in the past half year.

Screening similar historical complaint records, namely screening the historical complaint records similar to the current complaint type from the query result according to the classification result of hospital registration system fault complaint-patient dissatisfaction, and finding out that a plurality of related records exist, so as to indicate that the current user has the related portrait of the historical complaint.

And locating and storing a database of the historical complaint images of the user according to the unique identification information of the user. And (3) searching the historical complaint portrait, namely executing a query operation and searching the historical complaint portrait of the current user from the database. Assuming that the portrait dimension information includes complaint frequencies of 1-2 times per month (average over the past half year), complaint type preferences are mainly focused on registering system faults, with a sensitivity score of 7 points (10 points full, higher points indicating more sensitivity). And matching the content of the current complaint instant messaging message with each dimension information in the historical complaint image. The current complaint content is related to the registering system fault, the emotion of the patient is excited, and the high-frequency complaint with the risk of the updated complaint is identified by combining the historical complaint frequency and the sensitivity score, wherein the complaint is high-frequency complaint (average complaint 1-2 times per month in the past half year is relatively frequent), and the sensitivity of the patient is relatively high, so that the high-frequency complaint with the risk of the updated complaint can be further judged.

Because the complaint is a high frequency complaint, hospitals can set to shield all complaint instant messaging messages for a certain time (such as 2 months) (but special channels or personnel follow-up treatment is required to be ensured during shielding, such as customer service personnel actively contact patients to know the problem progress). Meanwhile, the information technology department is arranged to carry out comprehensive inspection and optimization on the registration system, the problem of system faults is solved as soon as possible, the success rate of patient registration is improved, and further upgrading of the patient problem is avoided.

Through the instant messaging message dynamic shielding method, hospitals can quickly classify and process complaints, more efforts are put into solving the actual problems of patients, and the complaints are not bothered by a large number of repeated complaints, so that the overall complaint processing efficiency is improved. By identifying high-frequency complaints and patients at risk of upgrading the complaints in time and taking corresponding measures, further worsening of the patient problem can be avoided, and reputation risks and potential medical disputes risks faced by hospitals are reduced. Although the high-frequency complaint patients are shielded, special people follow-up treatment is arranged, so that the patients feel the importance of the hospital on the problems, the satisfaction and the loyalty of the patients are improved, and the doctor-patient relationship is improved.

The method for dynamically shielding the instant messaging messages realizes the rapid processing of the instant messaging messages by automatically acquiring, extracting, classifying and judging the historical complaint images of the users, does not need to manually search mobile phone numbers and select shielding types one by one, greatly shortens the complaint processing time, improves the processing efficiency, classifies the complaint instant messaging messages by utilizing the key information and identifies the relevant images of the historical complaint of the users, can more accurately judge the properties of the complaint instant messaging messages, thereby implementing a more accurate shielding strategy, reducing the possibility of misjudgment and omission, and can timely shield the high-frequency complaint or malicious complaint instant messaging messages, effectively reduce the harassment and intrusion of the users, improve the use experience of the users, simultaneously shield the instant messaging messages according to the preset frequency rule, and also consider the service compliance, and further reduce the manual intervention flow, thereby improving the processing efficiency of enterprises in the aspect of more operation and further improving the processing efficiency of enterprises.

Fig. 3 is a schematic block diagram of an instant messaging dynamic screening apparatus 300 according to an embodiment of the present invention. As shown in fig. 3, the present invention further provides an instant messaging dynamic screening apparatus 300 corresponding to the above instant messaging dynamic screening method. The instant communication message dynamic screening apparatus 300 includes a unit for performing the above-described instant communication message dynamic screening method, and may be configured in a server.

Specifically, referring to fig. 3, the instant messaging dynamic shielding device 300 includes:

an obtaining unit 301, configured to obtain complaint instant messaging content submitted by a user;

an extracting unit 302, configured to extract complaint instant messaging content to obtain key information;

a classification unit 303, configured to classify the complaint instant messaging content according to the key information, so as to obtain a classification result;

A judging unit 304, configured to judge whether a current user has a history complaint related portrait based on the classification result;

a query recognition unit 305, configured to query the historical complaint related representation if the current user has the historical complaint related representation, and recognize the complaint nature of the current complaint instant message content;

and an implementation unit 306, configured to implement a corresponding shielding policy according to the complaint nature.

In an embodiment, the obtaining unit 301 is configured to interface with a channel complaint portal of an official APP, an instant messaging platform and a customer service hotline, and receive complaint information submitted by a user in a unified manner, so as to obtain complaint instant messaging content.

In an embodiment, the extracting unit 302 includes:

the preprocessing module is used for preprocessing the complaint instant messaging message content, including removing irrelevant characters, advertisement links and unified text formats, so as to obtain preprocessed content;

And the extraction and integration module is used for extracting keywords, semantic analysis and user emotion from the preprocessed content and integrating the keywords, the semantic analysis and the user emotion to form key information.

In an embodiment, the classifying unit 303 includes:

The matching module is used for matching keywords in the key information with a predefined classified keyword library so as to obtain a keyword matching result;

the whole analysis module is used for carrying out whole analysis on the semantics in the key information so as to obtain a semantic analysis result;

The analysis module is used for carrying out emotion analysis on the emotion of the user in the key information so as to obtain emotion analysis results;

The combination module is used for combining the keyword matching result, the semantic analysis result and the emotion analysis result to obtain a classification result.

In an embodiment, the determining unit 304 includes:

the extraction and query module is used for extracting the unique user identification information from the complaint instant messaging message content and querying the historical complaint record of the current user in the database according to the unique user identification information so as to obtain a query result;

And the screening and distinguishing module is used for screening the historical complaint records similar to the current complaint type from the query result according to the classification result so as to distinguish whether the current user has the historical complaint related portrait or not.

In an embodiment, the query recognition unit 305 includes:

The positioning module is used for positioning and storing a database of historical complaint images of the user according to the unique identification information of the user;

The execution search module is used for executing query operation and searching historical complaint portraits of the current user from the database, wherein the historical complaint portraits comprise the complaint frequency, the complaint type preference and the portrait dimension information of the sensitivity score;

The matching identification module is used for matching the content of the current complaint instant messaging message with each dimension information in the historical complaint image and identifying the image characteristics related to the content of the current complaint instant messaging message so as to obtain the complaint property.

In one embodiment, the complaint properties include high frequency complaints, malicious complaints, or special types.

It should be noted that, as will be clearly understood by those skilled in the art, the specific implementation process of the instant messaging message dynamic shielding device 300 and each unit may refer to the corresponding description in the foregoing method embodiment, and for convenience and brevity of description, the description is omitted here.

The instant messaging message dynamic barrier apparatus 300 described above may be implemented in the form of a computer program that is executable on a computer device such as that shown in fig. 4.

Referring to fig. 4, fig. 4 is a schematic block diagram of a computer device according to an embodiment of the present application. The computer device 500 may be a server, where the server may be a stand-alone server or may be a server cluster formed by a plurality of servers.

With reference to FIG. 4, the computer device 500 includes a processor 502, memory, and a network interface 505, connected by a system bus 501, where the memory may include a non-volatile storage medium 503 and an internal memory 504.

The non-volatile storage medium 503 may store an operating system 5031 and a computer program 5032. The computer program 5032 includes program instructions that, when executed, cause the processor 502 to perform an instant messaging message dynamic screening method.

The processor 502 is used to provide computing and control capabilities to support the operation of the overall computer device 500.

The internal memory 504 provides an environment for the execution of a computer program 5032 in the non-volatile storage medium 503, which computer program 5032, when executed by the processor 502, causes the processor 502 to perform an instant messaging dynamic screening method.

The network interface 505 is used for network communication with other devices. It will be appreciated by those skilled in the art that the architecture shown in fig. 4 is merely a block diagram of some of the architecture relevant to the present inventive arrangements and is not limiting of the computer device 500 to which the present inventive arrangements may be implemented, and that a particular computer device 500 may include more or fewer components than shown, or may combine some of the components, or have a different arrangement of components.

Wherein the processor 502 is configured to execute a computer program 5032 stored in a memory to implement the steps of:

the method comprises the steps of obtaining complaint instant messaging information content submitted by a user, extracting the complaint instant messaging information content to obtain key information, classifying the complaint instant messaging information content according to the key information to obtain a classification result, judging whether a current user has a historical complaint related image or not based on the classification result, inquiring the historical complaint related image and identifying the complaint property of the current complaint instant messaging information content if the current user has the historical complaint related image, and implementing a corresponding shielding strategy according to the complaint property.

It should be appreciated that in embodiments of the present application, the Processor 502 may be a central processing unit (Central Processing Unit, CPU), the Processor 502 may also be other general purpose processors, digital signal processors (DIGITAL SIGNAL processors, DSPs), application SPECIFIC INTEGRATED Circuits (ASICs), off-the-shelf Programmable gate arrays (Field-Programmable GATEARRAY, FPGA) or other Programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or the like. Wherein the general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.

Those skilled in the art will appreciate that all or part of the flow in a method embodying the above described embodiments may be accomplished by computer programs instructing the relevant hardware. The computer program comprises program instructions, and the computer program can be stored in a storage medium, which is a computer readable storage medium. The program instructions are executed by at least one processor in the computer system to implement the flow steps of the embodiments of the method described above.

Accordingly, the present invention also provides a storage medium. The storage medium may be a computer readable storage medium. The storage medium stores a computer program which, when executed by a processor, causes the processor to perform the steps of:

the method comprises the steps of obtaining complaint instant messaging information content submitted by a user, extracting the complaint instant messaging information content to obtain key information, classifying the complaint instant messaging information content according to the key information to obtain a classification result, judging whether a current user has a historical complaint related image or not based on the classification result, inquiring the historical complaint related image and identifying the complaint property of the current complaint instant messaging information content if the current user has the historical complaint related image, and implementing a corresponding shielding strategy according to the complaint property.

The storage medium may be a U-disk, a removable hard disk, a Read-Only Memory (ROM), a magnetic disk, or an optical disk, or other various computer-readable storage media that can store program codes.

Those of ordinary skill in the art will appreciate that the elements and algorithm steps described in connection with the embodiments disclosed herein may be embodied in electronic hardware, in computer software, or in a combination of the two, and that the elements and steps of the examples have been generally described in terms of function in the foregoing description to clearly illustrate the interchangeability of hardware and software. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.

In the several embodiments provided by the present invention, it should be understood that the disclosed apparatus and method may be implemented in other manners. For example, the device embodiments described above are merely illustrative. For example, the division of each unit is only one logic function division, and there may be another division manner in actual implementation. For example, multiple units or components may be combined or may be integrated into another system, or some features may be omitted, or not performed.

The steps in the method of the embodiment of the invention can be sequentially adjusted, combined and deleted according to actual needs. The units in the device of the embodiment of the invention can be combined, divided and deleted according to actual needs. In addition, each functional unit in the embodiments of the present invention may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit.

The integrated unit may be stored in a storage medium if implemented in the form of a software functional unit and sold or used as a stand-alone product. Based on such understanding, the technical solution of the present invention is essentially or a part contributing to the prior art, or all or part of the technical solution may be embodied in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a terminal, a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention.

While the invention has been described with reference to certain preferred embodiments, it will be understood by those skilled in the art that various changes and substitutions of equivalents may be made and equivalents will be apparent to those skilled in the art without departing from the scope of the invention. Therefore, the protection scope of the invention is subject to the protection scope of the claims.

Claims (10)

1. A method for dynamically shielding instant messaging messages, comprising: Obtain the content of complaint instant messaging messages submitted by users; Extract the content of complaint instant messaging messages to obtain key information; Classify the content of the complaint instant messaging messages according to key information to obtain classification results; Based on the classification results, determine whether the current user has a profile related to historical complaints; If the current user has a history of complaint-related profiles, query the history of complaint-related profiles and identify the nature of the complaint in the current instant messaging message; Implement corresponding blocking strategies based on the nature of the complaint. 2. The method for dynamically shielding instant messaging messages according to claim 1, wherein obtaining the content of the complaint instant messaging message submitted by the user comprises: Connect to the channel complaint portal of the official APP, instant messaging platform and customer service hotline to uniformly receive complaint information submitted by users to obtain the content of the complaint instant messaging message. 3. The method for dynamically shielding instant messaging messages according to claim 1, wherein extracting the content of the complaint instant messaging message to obtain key information comprises: Pre-processing the content of the complaint instant messaging message, including removing irrelevant characters, advertising links, and unifying the text format to obtain pre-processed content; Extract keywords, semantic analysis and user sentiment from pre-processed content and integrate them to form key information. 4. The method for dynamically shielding instant messaging messages according to claim 3, wherein classifying the content of the complaint instant messaging messages based on key information to obtain classification results comprises: Match the keywords in the key information with the predefined classification keyword library to obtain keyword matching results; Conduct a holistic analysis of the semantics in key information to obtain semantic analysis results; Perform sentiment analysis on user emotions in key information to obtain sentiment analysis results; Combine keyword matching results, semantic analysis results, and sentiment analysis results to obtain classification results. 5. The method for dynamic blocking of instant messaging messages according to claim 1, wherein the step of determining whether the current user has a profile related to historical complaints based on the classification result comprises: Extracting user unique identification information from the complaint instant messaging message content, and searching the database for the current user's historical complaint records based on the user unique identification information to obtain a query result; Based on the classification results, historical complaint records similar to the current complaint type are filtered from the query results to distinguish whether the current user has a historical complaint-related profile. 6. The method for dynamically blocking instant messaging messages according to claim 5, wherein if the current user has a historical complaint-related profile, querying the historical complaint-related profile and identifying the nature of the complaint of the current complaint instant messaging message content comprises: Based on the user's unique identification information, locate the database that stores the user's historical complaint profile; Execute a query operation to retrieve the current user's historical complaint profile from the database, including profile dimension information such as complaint frequency, complaint type preference, and sensitivity score; The content of the current complaint instant communication message is matched with the dimensional information in the historical complaint portrait, and the portrait features related to the content of the current complaint instant communication message are identified to obtain the nature of the complaint. 7. The method for dynamically shielding instant messaging messages according to claim 1, wherein the nature of the complaint includes: high-frequency complaints, malicious complaints or special types. 8. An instant messaging message dynamic shielding device, characterized by comprising: An acquisition unit, used to acquire the content of the complaint instant messaging message submitted by the user; An extraction unit, used to extract the content of the complaint instant messaging message to obtain key information; A classification unit, configured to classify the content of the complaint instant messaging message according to key information to obtain a classification result; A judgment unit, used to judge whether the current user has a historical complaint-related profile based on the classification result; A query identification unit is used to query the historical complaint-related portraits if the current user has any, and identify the nature of the complaint in the current complaint instant messaging message; The implementation unit is used to implement corresponding blocking strategies based on the nature of the complaint. 9. A computer device, characterized in that the computer device comprises a memory and a processor, the memory stores a computer program, and the processor implements the method according to any one of claims 1 to 7 when executing the computer program. 10. A storage medium, characterized in that the storage medium stores a computer program, and when the computer program is executed by a processor, it implements the method according to any one of claims 1 to 7.