CN114997708B

CN114997708B - Method and device for channel allocation according to user problems

Info

Publication number: CN114997708B
Application number: CN202210737327.1A
Authority: CN
Inventors: 龙翀; 于浩淼; 王颖
Original assignee: Alipay Hangzhou Information Technology Co Ltd
Current assignee: Alipay Hangzhou Information Technology Co Ltd
Priority date: 2022-06-27
Filing date: 2022-06-27
Publication date: 2025-04-01
Anticipated expiration: 2042-06-27
Also published as: CN114997708A

Abstract

The embodiment of the specification provides a channel distribution method and device for user questions, wherein the method comprises the steps of obtaining user questions of a user question, predicting the manual satisfaction degree of a user for manual service after the user enters a first manual channel according to the user questions by means of a pre-trained manual satisfaction degree prediction model, predicting the self-help solution rate of the user questions after the user enters the self-help channel according to the user questions by means of a pre-trained self-help solution rate prediction model, determining target channels in multiple channels according to the manual satisfaction degree and the self-help solution rate, and distributing the user questions to the target channels, wherein the multiple channels at least comprise the first manual channel and the self-help channel.

Description

Channel allocation method and device for user problem

Technical Field

The embodiment of the specification relates to the technical field of computers, in particular to a method and a device for channel allocation aiming at user problems.

Background

With the increasing business of companies, how to provide customers with a good customer service experience is a concern for most companies. However, on different days (e.g., weekdays, weekends, holidays, etc.), or different time periods of the same day (e.g., daytime, evening), the frequency of customer dialing or using the line is different, and peak periods necessarily cause tremendous stress to customer service personnel. If the scheduling is bad, the waiting time of the user can be greatly prolonged, and even the appeal (also called user problem) of the user can not be solved in time, thereby greatly influencing the user experience.

Therefore, a reasonable and reliable scheme is urgently needed, and the user problems can be distributed to proper channels, so that the customer service dispatching efficiency can be improved, and the user problems can be timely solved.

Disclosure of Invention

The embodiment of the specification provides a channel allocation method and device for user problems, which can allocate the user problems to proper channels, so that customer service scheduling efficiency can be improved, and the user problems can be timely solved.

According to the first aspect, the embodiment of the specification provides a channel distribution method for user questions, which comprises the steps of obtaining user questions of a user question, predicting manual satisfaction of a user for manual service after the user enters a first manual channel according to the user questions by means of a pre-trained manual satisfaction prediction model, predicting self-help solution rate of the user questions after the user enters a self-help channel according to the user questions by means of a pre-trained self-help solution rate prediction model, determining target channels in various channels according to the manual satisfaction and the self-help solution rate, and distributing the user questions to the target channels, wherein the various channels at least comprise the first manual channel and the self-help channel.

In some embodiments, the determining a target channel from a plurality of channels according to the manual satisfaction and the self-help resolution includes determining the first manual channel as the target channel if the manual satisfaction reaches a satisfaction threshold and the self-help resolution does not reach a resolution threshold, determining the self-help channel as the target channel if the manual satisfaction does not reach a satisfaction threshold and the self-help resolution reaches a resolution threshold, and selecting one of the first manual channel and the self-help channel as the target channel if the manual satisfaction reaches a satisfaction threshold and the self-help resolution reaches a resolution threshold.

In some embodiments, the selecting a channel from the first artificial channel and the self-service channel as the target channel includes obtaining idle degree information of the first artificial channel, and selecting a channel from the first artificial channel and the self-service channel as the target channel according to the idle degree information.

In some embodiments, the selecting a channel from the first artificial channel and the self-service channel as the target channel includes determining whether the current time is in a set busy time period, and if so, selecting the self-service channel as the target channel.

In some embodiments, the plurality of channels further includes a second artificial channel, and the determining a target channel from among the plurality of channels according to the artificial satisfaction and the self-help resolution includes determining the second artificial channel as the target channel if the artificial satisfaction does not reach a satisfaction threshold and the self-help resolution does not reach a resolution threshold.

In some embodiments, the method further comprises obtaining behavior information and/or factor information related to the user problem, wherein the behavior information comprises first behavior data generated by the user in a target application to which the user problem belongs and/or second behavior data generated by the target application for the user, the factor information comprises user factors of the user, service factors related to services to which the user problem belongs and/or experience factors of the user under the user problem, and the predicting the manual satisfaction degree of manual service after the user enters a first manual channel according to the user problem by using a pre-trained manual satisfaction degree prediction model comprises predicting the manual satisfaction degree by using the manual satisfaction degree prediction model according to the user problem and the behavior information and/or factor information, and the predicting the self-help resolution of the user after the user enters a self-help channel according to the user problem by using a pre-trained solution rate prediction model comprises predicting the self-help resolution of the user problem after the user enters the self-help channel according to the self-help resolution rate prediction model and the self-help resolution rate by using the user rate prediction model and the self-help resolution rate.

In some embodiments, the artificial satisfaction prediction model comprises a plurality of satisfaction prediction models, and the predicting the artificial satisfaction according to the user problem and the behavior information and/or factor information by using the artificial satisfaction prediction model comprises predicting a plurality of satisfaction sub-results according to the user problem and the behavior information and/or factor information by using the plurality of satisfaction prediction models, and determining the artificial satisfaction according to the plurality of satisfaction sub-results.

In some embodiments, the plurality of satisfaction prediction models includes a first hybrid network and at least one of a first Bert model, a first XGBoost model.

In some embodiments, the predicting a plurality of satisfaction sub-results based on the user problem and the behavior information and/or factor information using the plurality of satisfaction prediction models includes predicting a first satisfaction sub-result based on the user problem and the behavior information and/or factor information using the first hybrid network.

In some embodiments, the plurality of satisfaction prediction models includes the first Bert model, and the predicting, using the plurality of satisfaction prediction models, a plurality of satisfaction sub-results based on the user question, and the behavior information and/or factor information, further includes predicting, using the first Bert model, a second satisfaction sub-result based on the user question.

In some embodiments, the plurality of satisfaction prediction models includes the first XGBoost model, and the predicting, using the plurality of satisfaction prediction models, a plurality of satisfaction sub-results based on the user question, and the behavior information and/or factor information, further includes predicting, using the first XGBoost model, a third satisfaction sub-result based on the factor information.

In some embodiments, the self-help resolution prediction model comprises a plurality of resolution prediction models, and the predicting the self-help resolution according to the user problem and the behavior information and/or factor information by using the self-help resolution prediction model comprises predicting a plurality of resolution sub-results according to the user problem and the behavior information and/or factor information by using the plurality of resolution prediction models, and determining the self-help resolution according to the plurality of resolution sub-results.

In some embodiments, the plurality of resolution prediction models includes a second hybrid network and at least one of a second Bert model, a second XGBoost model.

In some embodiments, the predicting a plurality of resolution sub-results from the user problem and the behavior information and/or factor information using the plurality of resolution prediction models includes predicting a first resolution sub-result from the user problem and the behavior information and/or factor information using the second hybrid network.

In some embodiments, the plurality of resolution prediction models includes the second Bert model, and the predicting, with the plurality of resolution prediction models, a plurality of resolution sub-results based on the user problem, and the behavior information and/or factor information, further includes predicting, with the second Bert model, a second resolution sub-result based on the user problem.

In some embodiments, the plurality of resolution prediction models includes the second XGBoost model, and the predicting, using the plurality of resolution prediction models, a plurality of resolution sub-results based on the user issue, and the behavior information and/or factor information, further includes predicting, using the second XGBoost model, a third resolution sub-result based on the factor information.

In some embodiments, the first hybrid network is obtained by training an initial hybrid network by adopting a first training step of acquiring a plurality of first training samples, wherein any first training sample comprises a first historical user problem, and at least one of a manual satisfaction label and factor information associated with the first historical user problem, wherein for the first training samples in the plurality of first training samples, the first historical user problem in the first training sample and the behavior information and/or factor information associated with the first historical user problem are used as input, the manual satisfaction label associated with the first historical user problem is used as a training label, and training the initial hybrid network to obtain the first hybrid network.

In some embodiments, the first Bert model is obtained by training an initial Bert model using a second training step of training the initial Bert model using, for a first training sample of the plurality of first training samples, a first historical user question in the first training sample as input, a manual satisfaction label associated with the first historical user question as a training label, and obtaining a first Bert model.

In some embodiments, the arbitrary first training samples include factor information, the first XGBoost model is obtained by training the initial XGBoost model using a third training step of training the initial XGBoost model with respect to a first training sample of the plurality of first training samples, taking factor information associated with a first historical user problem in the first training sample as input, and taking a manual satisfaction label associated with the first historical user problem as a training label, to obtain a first XGBoost model.

In some embodiments, the second hybrid network is obtained by training an initial hybrid network by adopting a fourth training step of acquiring a plurality of second training samples, wherein any second training sample comprises a second historical user problem, self-help resolution labels associated with the second historical user problem and at least one of behavior information and factor information, taking the second historical user problem in the second training sample and the behavior information and/or factor information associated with the second historical user problem as input for the second training sample in the plurality of second training samples, taking the self-help resolution labels associated with the second historical user problem as training labels, and training the initial hybrid network to obtain the second hybrid network.

In some embodiments, the second Bert model is obtained by training an initial Bert model using a fifth training step of training the initial Bert model to obtain a second Bert model by taking a second historical user problem in a second training sample of the plurality of second training samples as input, and a self-help resolution label associated with the second historical user problem as a training label.

In some embodiments, the arbitrary second training samples include factor information, the second XGBoost model is obtained by training the initial XGBoost model using a sixth training step of training the initial XGBoost model to obtain a second XGBoost model for a second training sample of the plurality of second training samples, taking factor information associated with a second historical user problem in the second training sample as input, and taking a self-help resolution label associated with the second historical user problem as a training label.

In a second aspect, the embodiment of the specification provides a channel allocation device for user questions, which comprises an acquisition unit, a manual satisfaction prediction unit, a self-help solution rate prediction unit and a channel allocation unit, wherein the acquisition unit is configured to acquire user questions of user questions, the manual satisfaction prediction unit is configured to predict manual satisfaction of manual service after the user enters a first manual channel according to the user questions by using a pre-trained manual satisfaction prediction model, the self-help solution rate prediction unit is configured to predict self-help solution rate of the user questions after the user enters the self-help channel according to the user questions by using a pre-trained self-help solution rate prediction model, and the channel allocation unit is configured to determine a target channel in multiple channels according to the manual satisfaction and the self-help solution rate and allocate the user questions to the target channel, wherein the multiple channels at least comprise the first manual channel and the self-help channel.

In a third aspect, embodiments of the present specification provide a computer-readable storage medium having stored thereon a computer program, wherein the computer program, when executed in a computer, causes the computer to perform a method as described in any of the implementations of the first aspect.

In a fourth aspect, embodiments of the present specification provide a computing device comprising a memory and a processor, wherein the memory has executable code stored therein, and wherein the processor, when executing the executable code, implements a method as described in any of the implementations of the first aspect.

In a fifth aspect, embodiments of the present specification provide a computer program, wherein the computer program, when executed in a computer, causes the computer to perform a method as described in any of the implementations of the first aspect.

According to the channel distribution method and device for the user questions, after the user questions asked by the user are obtained, the manual satisfaction degree of the user on manual service after entering the first manual channel can be predicted according to the user questions, and the self-help solution rate of the user questions after entering the self-help channel can be predicted. Then, a more appropriate target channel can be determined among a plurality of channels, including at least a first artificial channel and a self-help channel, with reference to the artificial satisfaction and the self-help resolution, and the user problem is distributed to the target channel. Therefore, the user problems can be distributed to the proper channels, so that the customer service dispatching efficiency can be improved, and the user problems can be timely solved.

Drawings

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

FIG. 1 is an exemplary system architecture diagram in which some embodiments of the present description may be applied;

FIG. 2 is a schematic diagram of a channel selection method;

FIG. 3 is a flow chart of one embodiment of a method of channel allocation for user questions;

FIG. 4 is a flow chart of another embodiment of a method of channel allocation for user questions;

FIG. 5 is a schematic illustration of a targeting channel determination process;

Fig. 6 is a schematic structural view of an apparatus for channel distribution for user problems.

Detailed Description

The present specification is further described in detail below with reference to the drawings and examples. It is to be understood that the specific embodiments described herein are merely illustrative of the application and are not limiting of the application. The described embodiments are only some of the embodiments of the present description and not all of the embodiments. All other embodiments, which can be made by one of ordinary skill in the art without undue burden from the present disclosure, are intended to be within the scope of the present application based on the embodiments herein.

For convenience of description, only a portion related to the present invention is shown in the drawings. Embodiments and features of embodiments in this specification may be combined with each other without conflict.

As described above, if customer service scheduling is not good, waiting time of a user is greatly prolonged, and even a user's appeal cannot be timely solved, so that user experience is greatly affected.

Based on the above, some embodiments of the present disclosure provide a method for channel allocation for user problems, by which the user problems can be allocated to appropriate channels, so that customer service scheduling efficiency can be improved, and the user problems can be solved in time. In particular, FIG. 1 illustrates an exemplary system architecture diagram suitable for use with these embodiments.

As shown in fig. 1, the system architecture may include a server, which may be, for example, assigned to a customer service dispatch system, etc., and is not specifically limited herein.

In practice, the server may obtain a user question of a user question, such as user question a shown in fig. 1. The user question a may be a question asked by the user through an online consultation manner, or may be a question asked by a hot line call manner, which is not limited herein. When the user question a is a question asked by making a hot line call, the user question a may be obtained by text-converting voice information of the user.

Then, the server may predict, according to the user problem a, the degree of manual satisfaction with the manual service after the user enters the first manual channel (such as the common manual channel shown in fig. 1), and predict the self-help resolution of the user problem a after the user enters the self-help channel.

Individual personnel services under common personnel channels can typically address most of the user's problems. Artificial satisfaction may be specifically understood as the probability of whether a user is satisfied with an artificial service after entering a common artificial channel. Under the self-service channel, the user can solve the problem in various self-service modes, for example, the problem can be solved by using an intelligent robot to answer with a person, a short message link is sent, and/or a question and answer page of an APP (Application program) is sent, etc. The self-help resolution can be specifically understood as the probability of whether the user problem can be solved after the user enters the self-help channel.

Then, the server can determine the target channel from a plurality of channels according to the predicted manual satisfaction and self-help resolution. Among them, the target channel is a channel suitable for the user problem a. As an example, as shown in fig. 2, when the manual satisfaction is high and the self-help resolution is low, a general manual channel may be selected as the target channel. When the manual satisfaction is low and the self-help resolution is high, the self-help channel can be selected as a target channel. When the manual satisfaction and the self-help resolution are high, one channel can be selected from common manual channels and self-help channels to serve as a target channel. In practice, thresholds, such as satisfaction threshold and resolution threshold, can be used as the basis for measuring the degree of manual satisfaction and self-help resolution. It should be noted that the threshold may be defined manually or may be obtained by system statistics, which is not specifically limited herein.

In some embodiments, user problem A may be of great concern when both manual satisfaction and self-help resolution are low. For example, the above-described various channels may further include a second artificial channel, such as the special artificial channel shown in fig. 2, which may be selected as the target channel when both the artificial satisfaction and the self-help resolution are low.

The service capacity of the artificial customer service under the special artificial channel is generally higher than that under the common artificial channel. For the user problem which cannot be solved through the common artificial channel and the self-service channel, the user problem can be processed by the artificial customer service with stronger service capability by distributing the user problem to the special artificial channel, so that the resolution of the user problem and the satisfaction of the user can be improved.

After determining the target channel, the server may assign user question a to the target channel. Therefore, the user problems can be distributed to the proper channels, so that the customer service dispatching efficiency can be improved, and the user problems can be timely solved.

In the following, specific implementation steps of the above method are described in connection with specific embodiments.

Referring to FIG. 3, a flow 300 of one embodiment of a method of channel allocation for user questions is shown. The execution subject of the method may be the server shown in fig. 1, and the method includes the following steps:

step 302, obtaining user questions asked by a user;

step 304, predicting the manual satisfaction degree of the user on the manual service after entering the first manual channel according to the user problem by utilizing a pre-trained manual satisfaction degree prediction model;

Step 306, predicting the self-help solution rate of the user problem after the user enters the self-help channel according to the user problem by using a pre-trained self-help solution rate prediction model;

And 308, determining a target channel in a plurality of channels according to the manual satisfaction degree and the self-service resolution ratio, and distributing the user problems to the target channel, wherein the plurality of channels at least comprise a first manual channel and a self-service channel.

The above steps are further described below.

In step 302, user questions asked by the user may be obtained in real time. For example, user questions asked by the user via online consultation may be obtained. Or the voice information of the user can be obtained and subjected to text conversion so as to obtain the user problem. Wherein the voice message may be generated by making a hot line call.

Next, in step 304, a pre-trained artificial satisfaction prediction model may be utilized to predict, based on the user's questions, the user's artificial satisfaction with the artificial service after entering the first artificial channel. The first artificial channel may include, for example, a common artificial channel as previously described. The artificial satisfaction may be a number within [0,1 ].

As an example, the artificial satisfaction prediction model may be obtained by training an initial first machine learning model. The first machine learning model may be an untrained or untrained model. Further, the first machine learning model may include, for example, a Bert model, a hybrid network (MixNet), a convolutional neural network (Convolutional Neural Networks, CNN), or the like. Based on this, in step 304, the user questions may be input into the manual satisfaction prediction model such that the manual satisfaction prediction model predicts and outputs the manual satisfaction.

When the first machine learning model is trained, a plurality of first historical user questions collected in advance can be respectively used as input, and manual satisfaction labels respectively associated with the plurality of first historical user questions are used as training labels, so that the first machine learning model is trained, and a manual satisfaction prediction model is obtained. Specifically, in the training process, the prediction loss may be determined based on the artificial satisfaction label as the training label and the prediction result of the first machine learning model, and the network parameters in the first machine learning model may be adjusted with the goal of reducing the prediction loss.

The first historical user question may be a user question addressed through a first artificial channel. In practice, after the user ends the session with the artificial customer service in the first artificial channel, the user may evaluate the satisfaction level of the artificial customer service. Wherein the evaluated satisfaction level may be included in a plurality of satisfaction levels, which may be, in order from high to low, very satisfactory, general, unsatisfactory, very unsatisfactory. Based on this, the manual satisfaction label associated with the first historical user issue may be determined based on the satisfaction level associated with the first historical user issue.

As an example, when the satisfaction level of the first historical user problem is very satisfactory or satisfied, for example, a value of 1 may be set as the manual satisfaction tag of the first historical user problem, and when the satisfaction level of the first historical user problem is general, unsatisfactory or very unsatisfactory, for example, a value of 0 may be set as the manual satisfaction tag of the first historical user problem.

As another example, when the satisfaction level of the first historical user problem is very satisfactory, for example, a value of 1 may be set as the manual satisfaction tag of the first historical user problem, when the satisfaction level of the first historical user problem is very unsatisfactory, for example, a value of 0 may be set as the manual satisfaction tag of the first historical user problem, and when the satisfaction level of the first historical user problem is satisfactory, general, or unsatisfactory, for example, a value of 0.5 may be set as the manual satisfaction tag of the first historical user problem.

It should be appreciated that the manual satisfaction label may be set according to the actual situation, and is not specifically limited herein.

Next, in step 306, a self-help resolution of the user's problem after the user enters the self-help channel may be predicted from the user's problem using a pre-trained self-help resolution prediction model. Wherein, the self-help resolution may be a value within [0,1 ].

As an example, the self-help resolution prediction model may be obtained by training an initial second machine learning model. The second machine learning model may be an untrained or untrained model. Further, the second machine learning model may include, for example, a Bert model, a hybrid network, a convolutional neural network, or the like. Based on this, in step 306, the user issue may be input into the self-help resolution prediction model such that the self-help resolution prediction model predicts and outputs a self-help resolution.

When the second machine learning model is trained, a plurality of pre-collected second historical user questions can be respectively used as input, and self-help resolution labels respectively associated with the plurality of second historical user questions are used as training labels, so that the second machine learning model is trained, and a self-help resolution prediction model is obtained. Specifically, in the training process, the prediction loss may be determined based on the self-help resolution label as the training label and the prediction result of the second machine learning model, and the network parameters in the second machine learning model may be adjusted with the goal of reducing the prediction loss.

The second historical user issue may be a user issue addressed through a self-service channel. In practice, the user problem addressed through the self-service channel is usually addressed or not addressed. Based on this, the self-help resolution label associated with the second historical user issue may be determined based on the resolution status of the second historical user issue. For example, when the solution status of the second historical user problem is solved, for example, a value of 1 may be set as a self-help resolution label of the second historical user problem, and when the solution status of the second historical user problem is unresolved, for example, a value of 0 may be set as a self-help resolution label of the second historical user problem.

Next, in step 308, a target channel may be determined from a plurality of channels including at least a first artificial channel and a self-help channel based on the artificial satisfaction and the self-help resolution, and the user question may be assigned to the target channel.

Specifically, if the manual satisfaction reaches the satisfaction threshold and the self-service resolution does not reach the resolution threshold, the first artificial channel may be determined to be the target channel. If the manual satisfaction does not reach the satisfaction threshold and the self-help resolution reaches the resolution threshold, the self-help channel can be determined to be the target channel. If the manual satisfaction reaches the satisfaction threshold and the self-help resolution reaches the resolution threshold, a channel can be selected from the first manual channel and the self-help channel to serve as a target channel.

Further, when one of the first artificial channel and the self-service channel is selected as the target channel, various selection methods may be adopted.

In one example, one channel may be randomly selected as the target channel among the first artificial channel and the self-service channel.

In another example, the idle degree information of the first artificial channel may be obtained, and one channel may be selected from the first artificial channel and the self-service channel as the target channel according to the idle degree information. For example, whether the first artificial channel has the artificial customer service in the idle state or not may be determined according to the idle degree information, and if the determination result is yes, the first artificial channel is determined as the target channel. And if the determination result is negative, determining the self-service channel as the target channel. Therefore, when the first artificial channel has the artificial customer service in the idle state, the user problem is preferentially distributed to the first artificial channel, and the user problem is ensured to be solved in time.

In yet another example, it may be determined whether the current time is in a set busy period, and if the determination is yes, a self-service channel may be selected as the target channel. In practice, if a user requests help from a person in a busy period, queuing is generally required, and the user's appeal may not be resolved in time. By adopting the implementation mode, the user problems can be distributed to the self-service channels preferentially in a busy time period, and the user problems are ensured to be solved in time.

In some embodiments, the plurality of channels may further comprise a second artificial channel, which may comprise, for example, a special artificial channel as previously described. In step 308, if the manual satisfaction does not reach the satisfaction threshold and the self-help resolution does not reach the resolution threshold, then the second manual channel may be determined to be the target channel.

In practice, the business capability of the artificial customer service under the second artificial channel is generally higher than the business capability of the artificial customer service under the first artificial channel. For the user problem which can not be solved by the first artificial channel and the self-service channel, the user problem can be processed by the artificial customer service with stronger service capability by distributing the user problem to the second artificial channel, so that the resolution of the user problem and the satisfaction of the user can be improved.

In some embodiments, if the manual satisfaction does not reach the satisfaction threshold and the self-help resolution does not reach the resolution threshold, then the user questions may be assigned to the first manual channel and the self-help channel in step 308. Therefore, the user problem can be processed in a mode of combining the first artificial channel and the self-service channel, and the resolution ratio of the user problem and the satisfaction degree of the user can be improved.

In the embodiment corresponding to fig. 3, after obtaining the user question asked by the user, the manual satisfaction degree of the user for manual service after entering the first manual channel can be predicted according to the user question, and the self-help solution rate of the user question after entering the self-help channel can be predicted. Then, a more appropriate target channel can be determined among a plurality of channels, including at least a first artificial channel and a self-help channel, with reference to the artificial satisfaction and the self-help resolution, and the user problem is distributed to the target channel. Therefore, the user problems can be distributed to the proper channels, so that the customer service dispatching efficiency can be improved, and the user problems can be timely solved.

Referring further to FIG. 4, a flow 400 of another embodiment of a method for channel allocation for user problems is shown, the method may be performed by a server as shown in FIG. 1, the method comprising the steps of:

Step 402, obtaining user questions asked by the user, and behavior information and/or factor information related to the user questions;

Step 404, predicting the manual satisfaction of the user with respect to the manual service after entering the first manual channel according to the user problem, and the behavior information and/or factor information related to the user problem by using a pre-trained manual satisfaction prediction model;

Step 406, predicting the self-help solution rate of the user problem after the user enters the self-help channel according to the user problem, the behavior information and/or factor information related to the user problem by using a pre-trained self-help solution rate prediction model;

and step 408, determining a target channel in a plurality of channels according to the manual satisfaction degree and the self-service resolution ratio, and distributing the user problems to the target channel, wherein the plurality of channels at least comprise a first manual channel and a self-service channel.

The above steps are further described below.

In step 402, user questions asked by the user, as well as behavioral information and/or factor information related to the user questions, may be obtained in real-time. Here, for explanation of the user problem, reference may be made to the previous related description, and the description is not repeated here.

The behavior information related to the user question may include, for example, first behavior data generated by the user in a target application to which the user question belongs, and/or second behavior data generated by the target application for the user. The target application may be, for example, an application where an online consultation portal or a hotline phone portal used when the user asks a user question, or an application associated with a user identifier provided when the user asks a user question, etc., which is not limited herein. In addition, the target application may be any class of application, such as a payment class application, a social class application, a game class application, or an educational class application, among others.

The first behavior data may specifically include behavior data generated in the target application by a user recently, e.g., for a few days (e.g., about 2,3, 4, or 5 days, etc.), or a preset number of times (e.g., about 30, 40, or 50, etc.), which may include click behavior data, and/or browse behavior data, for example. The second behavior data may specifically include behavior data generated by the target application recently for the user, and the behavior data may include, for example, information actively pushed to the user by the target application, such as a bill, a reminder, and/or an advertisement actively pushed to the user. It should be noted that the first behavior data and the second behavior data may be respectively embodied as a behavior sequence, and each piece of data in the behavior sequence may be ordered according to a time sequence.

Factor information related to a user problem may include, for example, a user factor of the user, a business factor related to a business to which the user problem belongs, and/or an experience factor of the user under the user problem.

The user factors may include, for example, the age, gender, and/or familiarity with the target application of the user, etc. The familiarity with the target application may be represented by, for example, the total number of times the user logs in to the target application, the number of times the user recently logs in to the target application, the service life of the user to the target application, or the time the user last logs in to the target application.

The business factors may include, for example, a business identifier of a business to which the user problem belongs, a scene identifier of a business scene to which the business belongs, a target query corresponding to historical satisfaction and/or historical resolution. Wherein the target question is a question corresponding to the user question, which may be included in a question library. It should be noted that the question may be simply called a standard question, and a plurality of standard questions may be stored in the question bank.

In practice, the historical satisfaction degree corresponding to the target mark can be determined according to the manual satisfaction degree of a plurality of first historical user questions corresponding to the target mark under the first manual channel. The historical resolution corresponding to the target questions may be determined based on the self-help resolution of the plurality of second historical user questions corresponding to the target questions under the self-help channel. It should be appreciated that the historical satisfaction may be used as a reference factor in predicting manual satisfaction and the historical resolution may be used as a reference factor in predicting self-help resolution.

Experience factors may include, for example, the number of interaction rounds, intent recognition results, number of repeated help, and/or number of repeated accesses, among others.

The number of interaction rounds may be understood as the number of rounds that a user interacts with the first intelligent robot for a user problem before channel distribution is performed on the user problem. It is noted that the first intelligent robot may be a robot for preliminary handling of user problems, which is typically different from intelligent robots under self-service channels. In practice, it is possible for the user to interact with the first intelligent robot for several rounds before channel distribution is performed for the user's problem. For example, the hotline phone service and/or the online consultation service may have their own first intelligent robots, which may for example query the user for information when he dials a hotline phone or makes an online consultation, so that he responds. The information that the first intelligent robot inquires about the user may include, for example, but not limited to, a service type to which the user problem that the user needs to ask belongs, and/or content of the user problem, etc., which is not specifically limited herein.

The intention recognition result may be understood as a result of the first intelligent robot performing intention recognition on the user according to the user problem. The number of repeated help calls is understood to be the number of times the user repeatedly dials a hotline phone for the user's question. The number of repeated accesses may be understood as the number of times the user repeatedly accesses the target application for the user problem, in other words, the number of repeated accesses may be the number of times the user repeatedly consults online for the user problem.

It should be noted that the behavior information and the factor information described above may be used as auxiliary data for predicting the manual satisfaction and the self-help resolution, where the auxiliary data helps to improve the accuracy of the manual satisfaction and the self-help resolution that are predicted finally.

Next, in step 404, a pre-trained artificial satisfaction prediction model may be utilized to predict artificial satisfaction with the artificial service after the user enters the first artificial channel based on the user problem, and behavior information and/or factor information related to the user problem. Wherein, the artificial satisfaction can be a numerical value within [0,1 ].

As an example, the artificial satisfaction prediction model may be obtained by training an initial hybrid network. The initial hybrid network may be an untrained or untrained model. Based on this, the user problem, and behavior information and/or factor information related to the user problem may be input into the manual satisfaction prediction model, so that the manual satisfaction prediction model predicts and outputs the manual satisfaction according to the input information. When factor information is input into the artificial satisfaction prediction model, normalization processing can be performed on the factor information, and then the factor information after normalization processing can be input into the artificial satisfaction prediction model.

It should be noted that, when training the initial hybrid network, a plurality of first training samples may be obtained, where any first training sample may include a first historical user problem, and an artificial satisfaction label associated with the first historical user problem and at least one of behavior information and factor information. For a first training sample in the plurality of first training samples, a first historical user problem in the first training sample and behavior information and/or factor information associated with the first historical user problem can be used as input, a manual satisfaction label associated with the first historical user problem is used as a training label, and the initial hybrid network is trained to obtain a manual satisfaction prediction model. Specifically, during the training process, a prediction loss may be determined based on the artificial satisfaction label as a training label and a prediction result of the initial hybrid network, and network parameters in the initial hybrid network may be adjusted with the goal of reducing the prediction loss.

Next, in step 406, a self-help resolution of the user's problem after the user enters the self-help channel may be predicted from the user's problem, and from behavioral information and/or factor information related to the user's problem, using a pre-trained self-help resolution prediction model. Wherein, the self-help resolution may be a value within [0,1 ].

As an example, the self-help resolution prediction model may be obtained by training an initial hybrid network. The initial hybrid network may be an untrained or untrained model. Based on this, the self-help resolution prediction model may be input with the user problem, and behavior information and/or factor information related to the user problem, such that the self-help resolution prediction model predicts and outputs a self-help resolution.

It should be noted that, when training the initial hybrid network, a plurality of second training samples may be obtained, where any second training sample may include a second historical user problem, and a self-help resolution label associated with the second historical user problem and at least one of behavior information and factor information. For a second training sample in the plurality of second training samples, a second historical user problem in the second training sample and behavior information and/or factor information associated with the second historical user problem can be used as input, a self-help solution rate label associated with the second historical user problem is used as a training label, and the initial hybrid network is trained to obtain a self-help solution rate prediction model. Specifically, during the training process, the prediction loss may be determined based on the self-help resolution label as the training label and the prediction result of the initial hybrid network, and the network parameters in the initial hybrid network may be adjusted with the goal of reducing the prediction loss.

It should be noted that in predicting the manual satisfaction, the business factor used may include historical satisfaction corresponding to the target query. In predicting self-help resolution, the business factor used may include a historical resolution corresponding to the target bid. For example, if factor information is used as input and the factor information includes a business factor, the business factor may include, for example, historical satisfaction corresponding to the target query in step 404. In step 406, if the factor information is input and the factor information includes a business factor, the business factor may include, for example, a historical resolution corresponding to the target query.

Next, in step 408, a target channel may be determined from a plurality of channels including at least a first artificial channel and a self-help channel based on the artificial satisfaction and the self-help resolution, and the user question may be assigned to the target channel. The specific processing and technical effects of step 408 are similar to those of step 308 in the corresponding embodiment of fig. 3, and reference may be made to the related description of step 308, which is not repeated herein.

In the embodiment corresponding to fig. 4, the accuracy of the final prediction result can be further improved by using the user questions asked by the user and the behavior information and/or factor information related to the user questions as the reference data for the manual satisfaction and the self-help solution prediction. Therefore, the user problems can be distributed to more proper channels, so that the customer service dispatching efficiency can be further improved, and the user problems can be timely solved.

In the corresponding embodiment of fig. 4, as an implementation manner, the artificial satisfaction prediction model may include a plurality of satisfaction prediction models, and step 404 may further include predicting a plurality of satisfaction sub-results according to the user problem and behavior information and/or factor information related to the user problem by using the plurality of satisfaction prediction models, and determining the artificial satisfaction of the user with respect to the artificial service after entering the first artificial channel according to the plurality of satisfaction sub-results.

In practice, the plurality of satisfaction prediction models may include, for example, a first hybrid network, and at least one of a first Bert model, a first XGBoost model, and the like. As shown in fig. 5, which illustrates that the plurality of satisfaction prediction models described above include a first hybrid network, a first Bert model, and a first XGBoost model. Wherein fig. 5 is a schematic diagram of a targeting channel determination process.

As shown in fig. 5, a first satisfaction sub-result may be predicted from the user problem, and the behavior information and/or factor information related to the user problem, using the first hybrid network. For example, the user questions, as well as behavioral information and/or factor information related to the user questions, may be input into the first hybrid network such that the first hybrid network predicts and outputs a first satisfaction sub-result.

In addition, when the plurality of satisfaction prediction models includes the first Bert model, the first Bert model may be used to predict the second satisfaction sub-result according to the user problem. For example, the user questions may be input into the first Bert model such that the first Bert model predicts and outputs the second satisfaction sub-results.

In addition, in the case of acquiring factor information related to a user problem, when the above-described plurality of satisfaction prediction models includes the first XGBoost model, the third satisfaction sub-result may be predicted from the factor information using the first XGBoost model. For example, the factor information may be input into the first XGBoost model such that the first XGBoost model predicts and outputs a third satisfaction sub-result.

It should be appreciated that the predictions of the first hybrid network, the first Bert model, and the first XGBoost model, respectively, may all be referred to as artificial satisfaction. In order to distinguish from the final predicted artificial satisfaction, in the present embodiment, the predicted result is referred to as a satisfaction sub-result.

In practice, the first hybrid network, the first Bert model, and the first XGBoost model may be separately obtained by training.

For example, the first hybrid network may be obtained by training an initial hybrid network using a first training step of obtaining a plurality of first training samples, wherein any first training sample includes a first historical user problem, and at least one of a manual satisfaction label associated with the first historical user problem and behavior information and factor information, using the first historical user problem in the first training sample and the behavior information and/or factor information associated with the first historical user problem as input for a first training sample in the plurality of first training samples, using the manual satisfaction label associated with the first historical user problem as a training label, and training the initial hybrid network to obtain the first hybrid network. Wherein the initial hybrid network may be an untrained or untrained model. Specifically, during the training process, a prediction loss may be determined based on the artificial satisfaction label as a training label and a prediction result of the initial hybrid network, and network parameters in the initial hybrid network may be adjusted with the goal of reducing the prediction loss.

The first Bert model may be obtained by training an initial Bert model using a second training step of training the initial Bert model using a first historical user problem in a first training sample of the plurality of first training samples as input, using a manual satisfaction label associated with the first historical user problem as a training label, and obtaining the first Bert model. Wherein the initial Bert model may be an untrained or untrained completed model. Specifically, in the training process, the prediction loss may be determined based on the artificial satisfaction label as the training label and the prediction result of the initial Bert model, and the network parameters in the initial Bert model may be adjusted with the goal of reducing the prediction loss.

When any first training sample includes factor information, the first XGBoost model may be obtained by training the initial XGBoost model using a third training step of training the initial XGBoost model to obtain a first XGBoost model by taking factor information associated with a first historical user problem in the first training sample as input and taking a manual satisfaction label associated with the first historical user problem as a training label for a first training sample of the plurality of first training samples. Wherein the initial XGBoost model may be an untrained or untrained completed model. Specifically, during the training process, a prediction loss may be determined based on the artificial satisfaction label as a training label and the prediction result of the initial XGBoost model, and the network parameters in the initial XGBoost model may be adjusted with the goal of reducing the prediction loss.

After predicting the plurality of satisfaction sub-results, a determination of the manual satisfaction with the manual service after the user enters the first manual channel may be made based on the plurality of satisfaction sub-results, as shown in fig. 5. For example, an average of the plurality of satisfaction sub-results may be determined as the manual satisfaction. For another example, when the plurality of satisfaction prediction models are respectively provided with a weight value, the plurality of satisfaction sub-results may be weighted and summed according to the weight value, and the resulting sum value may be determined as the manual satisfaction.

In the embodiment corresponding to fig. 4, as an implementation manner, the self-help resolution prediction model may include a plurality of resolution prediction models, and step 406 may further include predicting a plurality of resolution sub-results according to the user problem and behavior information and/or factor information related to the user problem by using the plurality of resolution prediction models, and determining a self-help resolution of the user problem after the user enters the self-help channel according to the plurality of resolution sub-results.

In practice, the plurality of solution prediction models may include, for example, a second hybrid network, and at least one of a second Bert model, a second XGBoost model, and the like. As shown in fig. 5, which illustrates the above-described plurality of solution prediction models including a second hybrid network, a second Bert model, and a second XGBoost model.

As shown in fig. 5, the first resolution sub-result may be predicted from the user problem, and the behavior information and/or factor information related to the user problem, using the second hybrid network. For example, the user problem, as well as behavioral information and/or factor information related to the user problem, may be input into the second hybrid network such that the second hybrid network predicts and outputs the first resolution sub-result.

In addition, when the plurality of resolution prediction models includes the second Bert model, the first resolution may be predicted according to the user problem using the second Bert model. For example, the user problem may be input into a second Bert model such that the second Bert model predicts and outputs a second resolution sub-result.

In addition, in the case of acquiring factor information related to a user problem, when the above-described plurality of resolution prediction models includes the second XGBoost model, the first resolution may be predicted from the factor information using the second XGBoost model. For example, the factor information may be input into the second XGBoost model, such that the second XGBoost model predicts and outputs a third resolution sub-result.

It should be appreciated that the predicted results of each of the second hybrid network, the second Bert model, and the second XGBoost model may all be referred to as self-help resolution. In order to distinguish from the final predicted self-help resolution, in the present embodiment, this predicted result is referred to as a resolution sub-result.

In practice, the second hybrid network, the second Bert model, and the second XGBoost model may be separately obtained by training.

For example, the second hybrid network may be obtained by training the initial hybrid network using a fourth training step of obtaining a plurality of second training samples, wherein any second training sample includes a second historical user problem, and a self-help resolution label associated with the second historical user problem and at least one of behavior information and factor information, training the initial hybrid network using, as input, the second historical user problem in the second training sample and the behavior information and/or factor information associated with the second historical user problem for a second training sample in the plurality of second training samples, and using, as training label, the self-help resolution label associated with the second historical user problem, to obtain the second hybrid network. Wherein the initial hybrid network may be an untrained or untrained model. Specifically, during the training process, the prediction loss may be determined based on the self-help resolution label as the training label and the prediction result of the initial hybrid network, and the network parameters in the initial hybrid network may be adjusted with the goal of reducing the prediction loss.

The second Bert model may be obtained by training the initial Bert model using a fifth training step of training the initial Bert model to obtain a second Bert model using a second historical user problem in the second training sample as input, a self-help resolution label associated with the second historical user problem as a training label, and a second training step of training the initial Bert model using the second historical user problem as input. Wherein the initial Bert model may be an untrained or untrained completed model. Specifically, in the training process, the prediction loss can be determined based on the self-help resolution label as the training label and the prediction result of the initial hybrid network, and the network parameters in the initial Bert model can be adjusted with the goal of reducing the prediction loss.

When any second training sample includes factor information, a second XGBoost model may be obtained by training the initial XGBoost model using a sixth training step of training the initial XGBoost model to obtain a second XGBoost model by taking factor information associated with a second historical user problem in a second training sample of the plurality of second training samples as input and a self-help resolution label associated with the second historical user problem as a training label. Wherein the initial XGBoost model may be an untrained or untrained completed model. Specifically, during the training process, the prediction loss may be determined based on the self-help resolution label as the training label and the prediction result of the initial XGBoost model, and the network parameters in the initial XGBoost model may be adjusted with the goal of reducing the prediction loss.

After predicting the plurality of resolution sub-results, a self-help resolution of the user's problem after the user enters the self-help channel may be determined from the plurality of resolution sub-results as shown in fig. 5. For example, an average of the plurality of resolution sub-results may be determined as the self-resolution. For another example, when the plurality of solution rate prediction models are respectively provided with a weight value, the plurality of solution rate sub-results may be weighted and summed according to the weight value, and the resulting sum value may be determined as the self-service solution rate.

With continued reference to FIG. 5, after obtaining the final manual satisfaction using the plurality of satisfaction prediction models described above and obtaining the final self-help resolution using the plurality of resolution prediction models described above, a target channel may be determined from the plurality of channels described above based on the manual satisfaction and the self-help resolution. Here, for a specific determination method of the target channel, reference may be made to the description of step 308 in the corresponding embodiment of fig. 3, which is not repeated herein.

It should be noted that, for the three models of the Bert model, the hybrid network and the XGBoost model, since the Bert model has a better prediction effect on text, the XGBoost model has a better prediction effect on factors, and the hybrid network can predict according to various information, the satisfaction degree prediction models and the resolution ratio prediction models are obtained by training the initial three models, and the prediction results of the three models can be integrated to obtain a more accurate final prediction result.

With further reference to fig. 6, the present description provides one embodiment of an apparatus for channel allocation for user problems, which may be applied to a server as shown in fig. 1.

As shown in fig. 6, the apparatus 600 for channel allocation for user problems of the present embodiment includes an acquisition unit 601, a manual satisfaction prediction unit 602, a self-help resolution prediction unit 603, and a channel allocation unit 604. The system comprises an acquisition unit 601, a manual satisfaction prediction unit 602, a self-help resolution prediction unit 603 and a channel distribution unit 604, wherein the acquisition unit 601 is configured to acquire user questions of user questions, the manual satisfaction prediction unit 602 is configured to predict manual satisfaction of manual service after a user enters a first manual channel according to the user questions by using a pre-trained manual satisfaction prediction model, the self-help resolution prediction unit 603 is configured to predict self-help resolution of the user questions after the user enters the self-help channel according to the user questions by using a pre-trained self-help resolution prediction model, and the channel distribution unit 604 is configured to determine target channels in multiple channels according to the manual satisfaction and the self-help resolution and distribute the user questions to the target channels, wherein the multiple channels at least comprise the first manual channel and the self-help channels.

In some embodiments, the channel allocation unit 604 may be further configured to determine the first artificial channel as the target channel if the manual satisfaction reaches the satisfaction threshold and the self-help resolution does not reach the resolution threshold, determine the self-help channel as the target channel if the manual satisfaction does not reach the satisfaction threshold and the self-help resolution reaches the resolution threshold, and select one of the first artificial channel and the self-help channel as the target channel if the manual satisfaction reaches the satisfaction threshold and the self-help resolution reaches the resolution threshold.

In some embodiments, the channel allocation unit 604 may be further configured to obtain the free-degree information of the first artificial channel, and select one of the first artificial channel and the self-service channel as the target channel according to the free-degree information.

In some embodiments, the channel allocation unit 604 may be further configured to determine whether the current time is in a set busy period, and if so, select the self-service channel as the target channel.

In some embodiments, the plurality of channels may further include a second artificial channel, and the channel allocation unit 604 may be further configured to determine the second artificial channel as the target channel if the artificial satisfaction does not reach the satisfaction threshold and the self-help resolution does not reach the resolution threshold.

In some embodiments, the obtaining unit 601 may be further configured to obtain behavior information and/or factor information related to the user problem, the behavior information may include first behavior data generated by the user in a target application to which the user problem belongs and/or second behavior data generated by the target application for the user, the factor information may include a user factor of the user, a business factor related to a business to which the user problem belongs and/or an experience factor of the user under the user problem, and the manual satisfaction prediction unit 602 may be further configured to predict the manual satisfaction using a manual satisfaction prediction model based on the user problem and the behavior information and/or the factor information, and the self-help resolution prediction unit 603 may be further configured to predict the self-help resolution using a self-help resolution prediction model based on the user problem and the behavior information and/or the factor information.

In some embodiments, the manual satisfaction prediction model may comprise a plurality of satisfaction prediction models, and the manual satisfaction prediction unit 602 may be further configured to predict a plurality of satisfaction sub-results from the user problem, and the behavior information and/or factor information, using the plurality of satisfaction prediction models, and determine the manual satisfaction from the plurality of satisfaction sub-results.

In some embodiments, the plurality of satisfaction prediction models may include a first hybrid network and at least one of a first Bert model, a first XGBoost model.

In some embodiments, the manual satisfaction prediction unit 602 may be further configured to predict the first satisfaction sub-result from the user question, and the behavior information and/or factor information, using the first hybrid network.

In some embodiments, the plurality of satisfaction prediction models may include a first Bert model, and the manual satisfaction prediction unit 602 may be further configured to predict the second satisfaction sub-result based on the user problem using the first Bert model.

In some embodiments, the plurality of satisfaction prediction models may include a first XGBoost model, and the manual satisfaction prediction unit 602 may be further configured to predict the third satisfaction sub-result from the factor information using the first XGBoost model.

In some embodiments, the self-resolution prediction model may include a plurality of resolution prediction models, and the self-resolution prediction unit 603 may be further configured to predict a plurality of resolution sub-results from the user problem and the behavior information and/or factor information using the plurality of resolution prediction models, and determine a self-resolution from the plurality of resolution sub-results.

In some embodiments, the plurality of solution prediction models may include a second hybrid network and at least one of a second Bert model, a second XGBoost model.

In some embodiments, the self-help resolution prediction unit 603 may be further configured to predict the first resolution sub-result from the user questions, and the behavior information and/or factor information, using the second hybrid network.

In some embodiments, the plurality of resolution prediction models may include a second Bert model, and the self-help resolution prediction unit 603 may be further configured to predict a second resolution sub-result from the user problem using the second Bert model.

In some embodiments, the plurality of resolution prediction models may include a second XGBoost model, and the self-help resolution prediction unit 603 may be further configured to predict a third resolution sub-result from the factor information using the second XGBoost model.

In some embodiments, the first hybrid network may be obtained by training an initial hybrid network using a first training step of obtaining a plurality of first training samples, wherein any first training sample includes a first historical user problem, and a manual satisfaction label associated with the first historical user problem, and at least one of behavior information, factor information, training the initial hybrid network using the first historical user problem in the first training sample, and the behavior information and/or factor information associated with the first historical user problem as input, using the manual satisfaction label associated with the first historical user problem as a training label, and obtaining the first hybrid network.

In some embodiments, the first Bert model is obtained by training an initial Bert model using a second training step of training the initial Bert model using, for a first training sample of the plurality of first training samples, a first historical user question in the first training sample as input, a manual satisfaction label associated with the first historical user question as a training label, to obtain the first Bert model.

In some embodiments, any first training sample includes factor information, the first XGBoost model may be obtained by training the initial XGBoost model using a third training step of training the initial XGBoost model to obtain a first XGBoost model for a first training sample of the plurality of first training samples, taking factor information associated with a first historical user problem in the first training sample as input, and taking a manual satisfaction label associated with the first historical user problem as training label.

In some embodiments, the second hybrid network may be obtained by training the initial hybrid network using a fourth training step of obtaining a plurality of second training samples, wherein any second training sample includes a second historical user problem, and a self-help resolution label associated with the second historical user problem and at least one of behavior information and factor information, for a second training sample in the plurality of second training samples, taking the second historical user problem in the second training sample, and behavior information and/or factor information associated with the second historical user problem as input, taking the self-help resolution label associated with the second historical user problem as a training label, and training the initial hybrid network to obtain the second hybrid network.

In some embodiments, the second Bert model may be obtained by training the initial Bert model using a fifth training step of training the initial Bert model using, for a second training sample of the plurality of second training samples, a second historical user problem in the second training sample as input, a self-help resolution label associated with the second historical user problem as a training label, to obtain a second Bert model.

In some embodiments, the arbitrary second training samples include factor information, the second XGBoost model may be obtained by training the initial XGBoost model using a sixth training step of training the initial XGBoost model to obtain a second XGBoost model for a second training sample of the plurality of second training samples, taking factor information associated with a second historical user problem in the second training sample as input, and a self-help resolution label associated with the second historical user problem as training label.

In the embodiment of the apparatus corresponding to fig. 6, the specific processing of each unit and the technical effects brought by the processing may refer to the related description of the method embodiment in the foregoing, and will not be repeated herein.

The present specification also provides a computer-readable storage medium having stored thereon a computer program, wherein the computer program, when executed in a computer, causes the computer to perform the method for channel allocation for user problems described in the above respective method embodiments.

The embodiments of the present disclosure also provide a computing device, including a memory and a processor, where the memory stores executable code, and the processor executes the executable code to implement the channel allocation method for user problem described in the foregoing method embodiments respectively.

The embodiments of the present specification also provide a computer program, wherein the computer program, when executed in a computer, causes the computer to execute the method for channel allocation for user problems described in the above method embodiments, respectively.

Those of skill in the art will appreciate that in one or more of the above examples, the functions described in the various embodiments disclosed herein may be implemented in hardware, software, firmware, or any combination thereof. When implemented in software, these functions may be stored on or transmitted over as one or more instructions or code on a computer-readable medium.

In some cases, the actions or steps recited in the claims can be performed in a different order than in the embodiments and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

While the foregoing detailed description has described the objects, aspects and advantages of the embodiments disclosed herein in further detail, it should be understood that the foregoing detailed description is merely illustrative of the embodiments disclosed herein and is not intended to limit the scope of the embodiments disclosed herein, but rather any modifications, equivalents, improvements or the like that may be made to the embodiments disclosed herein are intended to be included within the scope of the embodiments disclosed herein.

Claims

1. A method of channel allocation for user questions, comprising:

Acquiring user questions asked by a user;

Predicting the manual satisfaction degree of the user for manual service after entering a first manual channel according to the user problem by utilizing a pre-trained manual satisfaction degree prediction model;

Predicting the self-help solution rate of the user problem after the user enters a self-help channel according to the user problem by using a pre-trained self-help solution rate prediction model;

If the manual satisfaction degree reaches a satisfaction degree threshold value and the self-service resolution ratio does not reach a resolution ratio threshold value, determining the first manual channel as a target channel;

If the manual satisfaction does not reach the satisfaction threshold and the self-help resolution reaches the resolution threshold, determining the self-help channel as the target channel;

if the manual satisfaction reaches a satisfaction threshold and the self-help resolution reaches a resolution threshold, selecting a channel from the first manual channel and the self-help channel as the target channel;

the user questions are assigned to the target channels.

2. The method of claim 1, wherein the selecting one of the first artificial channel and the self-service channel as the target channel comprises:

acquiring the idle degree information of the first artificial channel;

And selecting a channel from the first artificial channel and the self-service channel as the target channel according to the idle degree information.

3. The method of claim 1, wherein the selecting one of the first artificial channel and the self-service channel as the target channel comprises:

determining whether the current moment is in a set busy time period;

And if the self-service channel is the target channel, selecting the self-service channel as the target channel.

4. The method of claim 1, further comprising:

And if the manual satisfaction degree does not reach the satisfaction degree threshold value and the self-service resolution ratio does not reach the resolution ratio threshold value, determining a second manual channel as the target channel.

5. The method of one of claims 1-4, further comprising:

Obtaining behavior information and/or factor information related to the user problem, wherein the behavior information comprises first behavior data generated by the user in a target application to which the user problem belongs and/or second behavior data generated by the target application for the user, the factor information comprises user factors of the user, service factors related to services to which the user problem belongs and/or experience factors of the user under the user problem, and

The method for predicting the manual satisfaction degree of the user for the manual service after entering the first manual channel according to the user problem by utilizing a pre-trained manual satisfaction degree prediction model comprises the following steps:

Predicting the manual satisfaction according to the user problem, the behavior information and/or the factor information by using the manual satisfaction prediction model;

The self-help solution rate prediction model for predicting the self-help solution rate of the user problem after the user enters the self-help channel according to the user problem comprises the following steps:

And predicting the self-help resolution according to the user problem, the behavior information and/or the factor information by using the self-help resolution prediction model.

6. The method of claim 5, wherein the artificial satisfaction prediction model comprises a plurality of satisfaction prediction models, and

The predicting the manual satisfaction according to the user problem, the behavior information and/or the factor information by using the manual satisfaction prediction model comprises the following steps:

predicting a plurality of satisfaction sub-results according to the user problem, the behavior information and/or the factor information by using the plurality of satisfaction prediction models;

And determining the manual satisfaction according to the plurality of satisfaction sub-results.

7. The method of claim 6, wherein the plurality of satisfaction prediction models comprises a first hybrid network and at least one of a first Bert model, a first XGBoost model.

8. The method of claim 7, wherein predicting a plurality of satisfaction sub-results from the user question, and the behavioral information and/or factor information using the plurality of satisfaction prediction models comprises:

And predicting a first satisfaction degree sub-result according to the user problem, the behavior information and/or the factor information by using the first hybrid network.

9. The method of claim 8, wherein the plurality of satisfaction prediction models includes the first Bert model, and

The predicting, by using the multiple satisfaction prediction models, multiple satisfaction sub-results according to the user problem, the behavior information and/or the factor information, further includes:

and predicting a second satisfaction degree sub-result according to the user problem by using the first Bert model.

10. The method of claim 8 or 9, wherein the plurality of satisfaction prediction models includes the first XGBoost model, and

And predicting a third satisfaction degree sub-result according to the factor information by using the first XGBoost model.

11. The method of claim 5, wherein the self-help resolution prediction model comprises a plurality of resolution prediction models, and

The predicting the self-help resolution by using the self-help resolution prediction model according to the user problem, the behavior information and/or the factor information comprises the following steps:

Predicting a plurality of resolution sub-results according to the user problems, the behavior information and/or the factor information by using the plurality of resolution prediction models;

and determining the self-help resolution according to the plurality of resolution sub-results.

12. The method of claim 11, wherein the plurality of solution prediction models comprises a second hybrid network and at least one of a second Bert model, a second XGBoost model.

13. The method of claim 12, wherein predicting a plurality of resolution sub-results from the user problem, and the behavior information and/or factor information using the plurality of resolution prediction models, comprises:

And predicting a first resolution sub-result according to the user problem, the behavior information and/or the factor information by using the second hybrid network.

14. The method of claim 13, wherein the plurality of resolution prediction models includes the second Bert model, and

Predicting a plurality of resolution sub-results according to the user problem, the behavior information and/or the factor information by using the plurality of resolution prediction models, and further comprising:

And predicting a second resolution sub-result according to the user problem by using the second Bert model.

15. The method of claim 13 or 14, wherein the plurality of solution prediction models includes the second XGBoost model, and

and predicting a third resolution sub-result according to the factor information by using the second XGBoost model.

16. The method of claim 7, wherein the first hybrid network is obtained by training an initial hybrid network using a first training step of:

Acquiring a plurality of first training samples, wherein any first training sample comprises a first historical user problem, and a manual satisfaction label and at least one of behavior information and factor information which are related to the first historical user problem;

And for a first training sample in the plurality of first training samples, taking a first historical user problem in the first training sample and behavior information and/or factor information related to the first historical user problem as input, taking a manual satisfaction label related to the first historical user problem as a training label, and training the initial hybrid network to obtain a first hybrid network.

17. The method of claim 16, wherein the first Bert model is obtained by training an initial Bert model using the following second training step:

And for a first training sample in the plurality of first training samples, taking a first historical user problem in the first training sample as input, taking a manual satisfaction label associated with the first historical user problem as a training label, and training the initial Bert model to obtain a first Bert model.

18. The method of claim 16 or 17, wherein the arbitrary first training samples comprise factor information, the first XGBoost model being obtained by training an initial XGBoost model using the following third training step:

And for a first training sample in the plurality of first training samples, taking factor information related to a first historical user problem in the first training sample as input, taking a manual satisfaction label related to the first historical user problem as a training label, and training the initial XGBoost model to obtain a first XGBoost model.

19. The method of claim 12, wherein the second hybrid network is obtained by training an initial hybrid network using a fourth training step of:

acquiring a plurality of second training samples, wherein any second training sample comprises a second historical user problem, a self-help resolution label associated with the second historical user problem and at least one of behavior information and factor information;

And for a second training sample in the plurality of second training samples, taking a second historical user problem in the second training sample and behavior information and/or factor information related to the second historical user problem as input, taking a self-help resolution label related to the second historical user problem as a training label, and training the initial hybrid network to obtain a second hybrid network.

20. The method of claim 19, wherein the second Bert model is obtained by training an initial Bert model using the following fifth training step:

and for a second training sample in the plurality of second training samples, taking a second historical user problem in the second training sample as input, taking a self-help resolution label associated with the second historical user problem as a training label, and training the initial Bert model to obtain a second Bert model.

21. The method of claim 19 or 20, wherein the arbitrary second training samples comprise factor information, the second XGBoost model being obtained by training the initial XGBoost model using the following sixth training step:

And for a second training sample in the plurality of second training samples, taking factor information related to a second historical user problem in the second training sample as input, taking a self-help resolution label related to the second historical user problem as a training label, and training the initial XGBoost model to obtain a second XGBoost model.

22. An apparatus for channel allocation for user questions, comprising:

an acquisition unit configured to acquire a user question asked by a user;

The manual satisfaction prediction unit is configured to predict the manual satisfaction of the user on the manual service after entering the first manual channel according to the user problem by using a pre-trained manual satisfaction prediction model;

the self-help resolution prediction unit is configured to predict the self-help resolution of the user problem after the user enters a self-help channel according to the user problem by using a pre-trained self-help resolution prediction model;

The channel allocation unit is configured to determine the first artificial channel as a target channel if the manual satisfaction degree reaches a satisfaction degree threshold and the self-help resolution rate does not reach a resolution rate threshold, determine the self-help channel as the target channel if the manual satisfaction degree does not reach the satisfaction degree threshold and the self-help resolution rate reaches the resolution rate threshold, select one channel from the first artificial channel and the self-help channel as the target channel if the manual satisfaction degree reaches the satisfaction degree threshold and the self-help resolution rate reaches the resolution rate threshold, and allocate the user problem to the target channel.

23. A computer readable storage medium having stored thereon a computer program, wherein the computer program, when executed in a computer, causes the computer to perform the method of any of claims 1-21.

24. A computing device comprising a memory and a processor, wherein the memory has executable code stored therein, which when executed by the processor, implements the method of any of claims 1-21.