CN118862889A - A bidding data filling management method, device, equipment and medium - Google Patents

Abstract

The embodiments of the present specification disclose a bidding data filling and management method, device, equipment and medium, including: filling the bidding source data according to preset fields to obtain the bidding filling data to be evaluated; if the filling rate of the preset fields in the bidding filling data to be evaluated is less than a first threshold, determining the unfilled preset fields of the bidding filling data to be evaluated; inputting the unfilled preset fields into a pre-trained information extraction model to obtain first management data; if the filling rate of the preset fields in the first management data is not less than the first threshold, judging whether the accuracy of the preset fields in the first management data is less than a second threshold; if the accuracy of the preset fields in the first management data is less than the second threshold, performing regular management on the filling data in the first management data to obtain second management data; if the accuracy of the preset fields in the second management data is not less than the second threshold, obtaining qualified bidding data.

Description

A bidding data filling management method, device, equipment and medium

Technical Field

The present invention relates to the field of computer technology, and in particular to a bidding data filling management method, device, equipment and medium.

Background Art

With the rapid development of information technology, the amount and format of tender notice information have become increasingly complex. This information usually contains a large amount of data, such as project name, tenderer, bid deadline, required documents and qualifications, etc. The format of this information may vary from tenderer to tenderer, making it difficult to process uniformly.

In order to better promote the flow and efficient use of market information, high-quality data governance is required for these tender notice information. However, traditional manual recognition and processing methods cannot meet the requirements of high efficiency and high quality. Manual processing is slow, prone to errors, and difficult to process large amounts of data.

Therefore, a new approach is needed to process these tender announcement information to improve the efficiency and quality of data governance.

Summary of the invention

One or more embodiments of this specification provide a bidding data filling management method, device, equipment and medium for solving the technical problems raised by the background technology.

One or more embodiments of this specification adopt the following technical solutions:

One or more embodiments of this specification provide a bidding data filling management method, the method comprising:

Fill the tender source data according to the preset fields to obtain the tender filling data to be evaluated;

If the filling rate of the preset field in the tender filling data to be evaluated is less than a first threshold, determining the unfilled preset field of the tender filling data to be evaluated;

Inputting the unfilled preset fields into a pre-trained information extraction model to obtain first governance data;

If the filling rate of the preset field in the first governance data is not less than the first threshold, determining whether the accuracy rate of the preset field in the first governance data is less than the second threshold;

If the accuracy of the preset field in the first managed data is less than the second threshold, regularization management is performed on the padding data in the first managed data to obtain second managed data;

If the accuracy of the preset field in the second governance data is not less than the second threshold, qualified bidding data is obtained.

It should be noted that the embodiments of this specification have the following beneficial effects through the above contents:

Improve the efficiency and quality of data filling: Through information extraction models and regular governance, large amounts of data can be processed automatically, avoiding errors and inefficiencies that may occur during manual processing.

Ensure the accuracy and completeness of data: By judging the accuracy of filled data and regularizing it, the accuracy and completeness of data can be improved to ensure that the quality of data meets the requirements.

Promote the flow and efficient use of market information: Through high-quality data governance of tender announcement information, it is possible to better promote the flow and efficient use of market information and improve market transaction efficiency.

Reduce labor costs and workload: Avoid traditional manual identification processing methods, reduce labor costs and workload, and improve work efficiency.

Furthermore, the step of inputting the unfilled preset fields into a pre-trained information extraction model to obtain the first governance data includes:

Extracting data from the unfilled preset fields through a pre-trained information extraction model to obtain data to be filled in the unfilled preset fields;

The unfilled preset fields are filled with data according to the data to be filled to obtain first governance data.

It should be noted that the embodiments of this specification have the following beneficial effects through the above contents:

Improve the accuracy of data filling: Through the pre-trained information extraction model, the data to be filled in the preset fields that are not filled can be extracted more accurately, thereby improving the accuracy of data filling.

Reduce manual intervention: Automatic data extraction and filling reduces the need for manual intervention and improves work efficiency.

Quickly process large amounts of data: A large number of unpopulated preset fields can be quickly processed to improve the efficiency of data processing.

Furthermore, the tender filling data to be evaluated is an HTML format document, and the unfilled preset fields are extracted by a pre-trained information extraction model to obtain the data to be filled in the unfilled preset fields, including:

The tender filling data to be evaluated is input into the information extraction model, and the information corresponding to the unfilled preset fields is extracted through a pointer network to achieve named entity recognition, relationship extraction, event extraction, and attribute sentiment extraction.

It should be noted that the embodiments of this specification have the following beneficial effects through the above contents:

Improve the accuracy of data extraction: Using pre-trained information extraction models and pointer networks, you can more accurately extract information corresponding to unfilled preset fields. These models are trained to learn the semantic and structural features of HTML formatted documents to better identify and extract the required data.

Implement multi-task data extraction: Through multiple tasks such as named entity recognition, relationship extraction, event extraction, and attribute sentiment extraction, richer and more comprehensive information can be obtained. This helps to gain a deeper understanding of all aspects of the tender filling data to be evaluated, and provides stronger support for subsequent analysis and decision-making.

Adapt to HTML format: Since the data filled in the tender to be evaluated is in HTML format, this method can directly process this common web page format. No additional format conversion or preprocessing is required, which reduces the complexity and workload of data processing.

Improve data management efficiency: The automatic data extraction process can significantly improve the efficiency of data management. Compared with manual data extraction, the model can quickly process a large number of HTML documents, saving time and labor costs.

Improve data quality: Accurately extracting the data to be filled helps improve data quality. High-quality data is essential for accurate analysis and decision-making, and can reduce errors and uncertainties.

Providing a comprehensive information perspective: The extraction results of multiple tasks can provide a more comprehensive information perspective. For example, named entity recognition can identify important entities, relationship extraction can reveal the relationship between them, event extraction can understand related events, and attribute sentiment extraction can obtain the evaluation and emotional tendency of things.

Furthermore, before extracting data from the unfilled preset fields by using a pre-trained information extraction model to obtain the data to be filled in the unfilled preset fields, the method further includes:

The BeautifulSoup class in the bs4 module of Python is used to remove the script tag content and the style tag content in the tender filling data to be evaluated, and remove the class attributes and id attributes of other tags.

It should be noted that the embodiments of this specification have the following beneficial effects through the above contents:

Improve data cleanliness: Removing the content of the script and style tags can reduce the interference of irrelevant information and improve the cleanliness and readability of the data. These tags usually contain information related to page layout and style, which is not critical for data extraction and analysis.

Reduce data redundancy: Removing the class and id attributes of other tags can reduce data redundancy. These attributes are usually used for page style design and interactive control, and have little impact on the meaning of the data itself.

Improve model efficiency: Cleaning and simplifying HTML data can improve the efficiency and accuracy of information extraction models. The model does not need to process unnecessary tag attributes, so it can focus more on extracting meaningful content.

Enhanced data consistency: By uniformly removing specific tag content and attributes, it helps ensure data consistency and standardization. This is very important for subsequent data analysis and processing, and can reduce differences caused by different tag structures.

Easier data parsing: Simplified HTML data is easier to parse and process. Removing irrelevant tags and attributes can make the data easier to understand and operate by other tools and algorithms.

Improve the accuracy of information extraction: Reducing interference factors can improve the information extraction model's ability to extract key information, thereby improving the accuracy of the extraction results.

Reduce the difficulty of model learning: Cleaning data can reduce the difficulty of learning information extraction models. Reducing unnecessary noise and interference can make it easier for the model to learn meaningful features and patterns.

Furthermore, performing regularization on the padding data in the first governance data to obtain second governance data includes:

The fill data in the first governance data is subjected to address splitting, amount extraction, date standardization, mobile phone number extraction, and preset symbol filtering by regular expressions to obtain second governance data.

It should be noted that the embodiments of this specification have the following beneficial effects through the above contents:

Improve data quality: Through regular governance, noise, errors and inconsistencies in data can be removed to improve data accuracy and reliability.

Extract key information: Regular expressions can be used for address splitting, amount extraction, date standardization, mobile phone number extraction, etc., to help quickly and accurately extract key information from the data.

Data Standardization: Date standardization and filtering with preset symbols help convert data into a uniform format, making it easier to analyze and compare.

Enhanced data consistency: Regular governance can ensure that the populated data follows certain rules and patterns, thereby enhancing the consistency and repeatability of the data.

Simplified data processing: The processed second governance data is more standardized and concise, which is convenient for subsequent data analysis, storage and transmission.

Improve data availability: By improving data quality and consistency, formal governance makes data more valuable and better supports decision making and business processes.

Furthermore, if the accuracy of the preset field in the second governance data is less than the second threshold, the method further includes:

Extracting information from the filling data in the second governance data by using a preset large model to obtain third governance data;

If the accuracy of the preset field in the third governance data is not less than the second threshold, qualified bidding data is obtained.

It should be noted that the embodiments of this specification have the following beneficial effects through the above contents:

Improve the accuracy of data governance: By presetting large models for information extraction, the accuracy of data governance can be further improved. Large models usually have more powerful language understanding and information extraction capabilities, and can extract the required information from the populated data more comprehensively and accurately.

Reduce misjudgments and omissions: When the accuracy of the preset field in the second governance data is less than the threshold, using a large model for further processing can reduce the occurrence of misjudgments and omissions. The large model can use its learned patterns and knowledge to better handle complex and ambiguous information, thereby improving the quality of data governance.

Providing bidding data that is more in line with the conditions: By extracting information from the second governance data, the accuracy of the preset fields in the third governance data obtained is not less than the second threshold, which means that the obtained data is more in line with the bidding conditions and requirements. This helps to improve the quality and reliability of bidding data and provide stronger support for bidding decisions.

Enhance the availability and value of data: Accurate and compliant bidding data has higher availability and value. This data can be used for further analysis, comparison and decision-making, helping the tenderer to better understand the situation of potential suppliers and make wise choices.

Improve the efficiency and effectiveness of the bidding process: Accurate bidding data can reduce delays and problems caused by wrong or inaccurate data during the bidding process. At the same time, qualified data can better meet bidding needs and improve the effectiveness and efficiency of the bidding process.

Reduce risks and costs: By ensuring the accuracy and compliance of bidding data, the risks and costs caused by incorrect data can be reduced. For example, avoiding cooperation with unqualified suppliers can reduce potential legal risks and economic losses.

Improve data governance processes and methods: Identifying and improving deficiencies in the data governance process will help to continuously optimize data governance processes and methods. By introducing advanced technologies such as big models, the ability and level of data governance can be improved to adapt to changing business needs and data challenges.

Furthermore, extracting information from the fill data in the second governance data by using a preset large model to obtain third governance data includes:

Define the role of the large model and provide a task description for information extraction from the second governance data;

According to the task description of extracting information from the second governance data based on the large model, information is extracted from the filling data in the second governance data to obtain third governance data.

It should be noted that the embodiments of this specification have the following beneficial effects through the above contents:

Improve the accuracy of information extraction: By clarifying the role of the big model and providing a detailed task description, the big model can better understand the type and context of information to be extracted. This helps improve the accuracy of the extraction results and reduce misjudgments and omissions.

Reduced manual intervention: Defining the role of a large model can make the information extraction process more automated, reducing the need for manual intervention. This can improve work efficiency and save time and human resources.

Improve data quality: Accurate information extraction helps improve the quality of third-party governance data. This can make subsequent data analysis, decision-making, and business processes more reliable and effective.

Support better decision making: By providing more accurate and detailed information, third-party governance data can provide stronger support for decision making, helping decision makers make more informed choices and reduce the risk of making decisions based on inaccurate or incomplete information.

Enhance model interpretability: Clarifying the role and task description of large models helps enhance model interpretability. This allows people to better understand how the model makes decisions and extracts information, increasing trust and understanding of the model results.

Adapt to different data requirements: By flexibly defining the roles and task descriptions of the large model, it can adapt to different information extraction requirements and data characteristics. This makes the method more general and scalable.

One or more embodiments of this specification provide a bidding data filling management device, including:

A data filling unit fills the tender source data according to preset fields to obtain tender filling data to be evaluated;

an unfilled field determining unit, which determines unfilled preset fields of the tender filling data to be evaluated if the filling rate of the preset fields in the tender filling data to be evaluated is less than a first threshold;

A first management unit inputs the unfilled preset field into a pre-trained information extraction model to obtain first management data;

an accuracy judgment unit, which judges whether the accuracy of the preset field in the first governance data is less than a second threshold value if the filling rate of the preset field in the first governance data is not less than a first threshold value;

A second governance unit, if the accuracy of the preset field in the first governance data is less than the second threshold, performs regularization governance on the padding data in the first governance data to obtain second governance data;

The bidding data determining unit obtains bidding data that meets the conditions if the accuracy of the preset field in the second governance data is not less than the second threshold.

One or more embodiments of this specification provide a bidding data filling management device, including:

at least one processor; and,

a memory communicatively connected to the at least one processor; wherein,

The memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to:

Fill the tender source data according to the preset fields to obtain the tender filling data to be evaluated;

If the filling rate of the preset field in the tender filling data to be evaluated is less than a first threshold, determining the unfilled preset field of the tender filling data to be evaluated;

Inputting the unfilled preset fields into a pre-trained information extraction model to obtain first governance data;

If the filling rate of the preset field in the first governance data is not less than the first threshold, determining whether the accuracy rate of the preset field in the first governance data is less than the second threshold;

If the accuracy of the preset field in the first managed data is less than the second threshold, regularization management is performed on the padding data in the first managed data to obtain second managed data;

If the accuracy of the preset field in the second governance data is not less than the second threshold, qualified bidding data is obtained.

One or more embodiments of this specification provide a non-volatile computer storage medium storing computer executable instructions, which can achieve the following when executed by a computer:

Fill the tender source data according to the preset fields to obtain the tender filling data to be evaluated;

If the filling rate of the preset field in the tender filling data to be evaluated is less than a first threshold, determining the unfilled preset field of the tender filling data to be evaluated;

Inputting the unfilled preset fields into a pre-trained information extraction model to obtain first governance data;

If the filling rate of the preset field in the first governance data is not less than the first threshold, determining whether the accuracy rate of the preset field in the first governance data is less than the second threshold;

If the accuracy of the preset field in the first managed data is less than the second threshold, regularization management is performed on the padding data in the first managed data to obtain second managed data;

If the accuracy of the preset field in the second governance data is not less than the second threshold, qualified bidding data is obtained.

At least one of the above technical solutions adopted in the embodiments of this specification can achieve the following beneficial effects:

Improve the efficiency and quality of data filling: Through information extraction models and regular governance, large amounts of data can be processed automatically, avoiding errors and inefficiencies that may occur during manual processing.

Ensure the accuracy and completeness of data: By judging the accuracy of filled data and regularizing it, the accuracy and completeness of data can be improved to ensure that the quality of data meets the requirements.

Promote the flow and efficient use of market information: Through high-quality data governance of tender announcement information, it is possible to better promote the flow and efficient use of market information and improve market transaction efficiency.

Reduce labor costs and workload: Avoid traditional manual identification processing methods, reduce labor costs and workload, and improve work efficiency.

BRIEF DESCRIPTION OF THE DRAWINGS

In order to more clearly illustrate the technical solutions in the embodiments of this specification or the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art description. Obviously, the drawings described below are only some embodiments recorded in this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. In the drawings:

FIG1 is a flow chart of a bidding data filling management method provided by one or more embodiments of this specification;

FIG2 is a schematic diagram of a structured governance process provided by one or more embodiments of this specification;

FIG3 is a schematic diagram of the structure of a bidding data filling and management device provided by one or more embodiments of this specification;

FIG4 is a schematic diagram of the structure of a bidding data filling management device provided by one or more embodiments of this specification.

DETAILED DESCRIPTION

The embodiments of this specification provide a bidding data filling management method, device, equipment and medium.

In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments of this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this specification.

Figure 1 is a flow chart of a bidding data filling management method provided by one or more embodiments of this specification, which can be executed by a bidding data filling management system. Certain input parameters or intermediate results in the process allow manual intervention and adjustment to help improve accuracy.

The method steps of the embodiment of this specification are as follows:

S102, filling the bidding source data according to the preset fields to obtain the bidding filling data to be evaluated.

In the examples of this specification, the above contents can be implemented by the following specific implementation schemes:

Determine the preset fields: Specify the preset fields to be filled in. These fields may be specific information in the bidding project, such as project name, bidding number, budget amount, bidding deadline, etc. The specific preset fields can be set according to actual needs.

Collect tender source data: Obtain tender source data containing relevant information. This can be data extracted from tender documents, spreadsheets, databases or other data sources.

Extract field values from source data: Extract the values of corresponding fields from the bidding source data according to the preset fields. This step can be completed using data extraction tools, programming languages, or database query statements.

Data cleaning and preprocessing: Clean and preprocess the extracted values, such as handling null values, invalid values, etc. Ensure that the extracted values are accurate and usable.

Fill preset fields: Fill the preprocessed values into the reserved preset fields accordingly. You can use data insert statements, data update operations, or other appropriate methods to fill the value domain into the specified field.

S104: If the filling rate of the preset fields in the tender filling data to be evaluated is less than a first threshold, determine unfilled preset fields of the tender filling data to be evaluated.

In the examples of this specification, the above contents can be implemented by the following specific implementation schemes:

Determine the first threshold: Determine a reasonable first threshold based on actual conditions. This threshold can be set based on business needs, data quality requirements or other relevant factors. For example, the first threshold can be set to 70%.

Field fill rate calculation: For each preset field, calculate its fill rate in the tender fill data to be evaluated. The fill rate calculation can be completed by counting the ratio of the number of records with actual values in the field to the total number of records.

Comparing the filling rate with the first threshold: comparing the filling rate of each preset field with the first threshold. If the filling rate is less than the first threshold, determining that the preset field is an unfilled preset field.

Unfilled preset fields determination: Unfilled preset fields refer to those preset fields that are not filled with actual data during the data filling process. When filling data, if some preset fields do not receive corresponding data values, unfilled preset fields will be formed. This may be due to the lack of relevant information in the data source, or errors or omissions in the filling process. Subsequent processing is mainly for the situation of "errors or omissions in the filling process".

S106, inputting the unfilled preset fields into a pre-trained information extraction model to obtain first governance data.

In an embodiment of the present specification, data can be first extracted from the unfilled preset fields through a pre-trained information extraction model to obtain the data to be filled in the unfilled preset fields; then, data is filled in the unfilled preset fields according to the data to be filled to obtain the first governance data.

In the examples of this specification, the above contents can be implemented by the following specific implementation schemes:

Model training:

-Choose an information extraction model suitable for the data extraction task, such as a neural network-based model.

- Use existing labeled data to train the model to improve the accuracy and generalization ability of the model.

- Adjust model hyperparameters such as learning rate, number of layers, number of nodes, etc. to optimize model performance.

Data extraction:

- Input the unpopulated preset fields into the pre-trained information extraction model.

-The model analyzes and extracts the data to be filled in the tender to be evaluated, and obtains the data to be filled in the preset fields that have not been filled.

Data filling:

-Match and associate the data to be filled with the unfilled preset fields.

-Use the appropriate filling method to fill the data to be filled into the unfilled preset fields.

-It can be filled using rules, machine learning algorithms, etc.

It should be noted that the embodiments of this specification have the following beneficial effects through the above contents:

Improve the accuracy of data filling: Through the pre-trained information extraction model, the data to be filled in the preset fields that are not filled can be extracted more accurately, thereby improving the accuracy of data filling.

Reduce manual intervention: Automatic data extraction and filling reduces the need for manual intervention and improves work efficiency.

Quickly process large amounts of data: A large number of unpopulated preset fields can be quickly processed to improve the efficiency of data processing.

Furthermore, the tender filling data to be evaluated can be a document in HTML format. Through a pre-trained information extraction model, data is extracted from the unfilled preset fields. When the data to be filled in the unfilled preset fields is obtained, the tender filling data to be evaluated can be input into the information extraction model, and the information corresponding to the unfilled preset fields can be fragmented through a pointer network to realize named entity recognition, relationship extraction, event extraction, and attribute sentiment extraction.

It should be noted that the above content can be implemented through the following specific implementation plans:

1. Objectives

Through the pre-trained information extraction model, data extraction is performed on the bidding data to be evaluated that has not filled in the preset fields, realizing named entity recognition, relationship extraction, event extraction and attribute sentiment extraction.

2. Data Preparation

1. Documents required for tender evaluation, in HTML format.

2. Extract text data containing preset fields to be filled.

3. Model Training

1. Use a large amount of HTML formatted text data to train the information extraction model.

2. The training method can be supervised learning, unsupervised learning or reinforcement learning.

4. Data Extraction

1. Analyze the data filled in the bidding to be evaluated.

2. Input the parsed text into the information extraction model.

3. Extract fragments of information corresponding to unfilled preset fields through the pointer network.

5. Extraction Type

1. Named entity recognition: Identify entities such as names of people and places in the text.

2. Relationship extraction: extract the relationship between different entities in the text.

3. Event extraction: extract event information from text.

4. Attribute sentiment extraction: obtain sentiment information about attributes in the text.

6. Output the data to be filled

The extracted data to be filled will be sorted and output for subsequent other operations.

It should be noted that the embodiments of this specification have the following beneficial effects through the above contents:

Improve the accuracy of data extraction: Using pre-trained information extraction models and pointer networks, you can more accurately extract information corresponding to unfilled preset fields. These models are trained to learn the semantic and structural features of HTML formatted documents to better identify and extract the required data.

Implement multi-task data extraction: Through multiple tasks such as named entity recognition, relationship extraction, event extraction, and attribute sentiment extraction, richer and more comprehensive information can be obtained. This helps to gain a deeper understanding of all aspects of the tender filling data to be evaluated, and provides stronger support for subsequent analysis and decision-making.

Adapt to HTML format: Since the data filled in the tender to be evaluated is in HTML format, this method can directly process this common web page format. No additional format conversion or preprocessing is required, which reduces the complexity and workload of data processing.

Improve data management efficiency: The automatic data extraction process can significantly improve the efficiency of data management. Compared with manual data extraction, the model can quickly process a large number of HTML documents, saving time and labor costs.

Improve data quality: Accurately extracting the data to be filled helps improve data quality. High-quality data is essential for accurate analysis and decision-making, and can reduce errors and uncertainties.

Providing a comprehensive information perspective: The extraction results of multiple tasks can provide a more comprehensive information perspective. For example, named entity recognition can identify important entities, relationship extraction can reveal the relationship between them, event extraction can understand related events, and attribute sentiment extraction can obtain the evaluation and emotional tendency of things.

Furthermore, before extracting data from the unfilled preset fields through the pre-trained information extraction model and obtaining the data to be filled in the unfilled preset fields, the script tag content and the style tag content in the tender filling data to be evaluated can be removed through the BeautifulSoup class in the bs4 module of Python, and the class attributes and id attributes of other tags can be removed.

It should be noted that the above content can be implemented through the following specific implementation plans:

Import the BeautifulSoup class: Use the BeautifulSoup class in Python's bs4 module to parse and process the tender filling data to be evaluated.

Read Data: Reads a file or string containing the data populated by the tender to be evaluated.

Parse the data: Use the constructor of the BeautifulSoup class to parse the data into a BeautifulSoup object.

Remove script tag content: Use the BeautifulSoup object’s methods to find all script tags and remove them using the decompose() method.

Remove style tag content: Use a similar method to find all style tags and remove them.

Remove the class and id attributes of other tags: Traverse all other tags and use the del statement to remove their class and id attributes.

Get the cleaned data: Convert the cleaned BeautifulSoup object back to a string or file for subsequent use.

It should be noted that the embodiments of this specification have the following beneficial effects through the above contents:

Improve data cleanliness: Removing the content of the script and style tags can reduce the interference of irrelevant information and improve the cleanliness and readability of the data. These tags usually contain information related to page layout and style, which is not critical for data extraction and analysis.

Reduce data redundancy: Removing the class and id attributes of other tags can reduce data redundancy. These attributes are usually used for page style design and interactive control, and have little impact on the meaning of the data itself.

Improve model efficiency: Cleaning and simplifying HTML data can improve the efficiency and accuracy of information extraction models. The model does not need to process unnecessary tag attributes, so it can focus more on extracting meaningful content.

Enhanced data consistency: By uniformly removing specific tag content and attributes, it helps ensure data consistency and standardization. This is very important for subsequent data analysis and processing, and can reduce differences caused by different tag structures.

Easier data parsing: Simplified HTML data is easier to parse and process. Removing irrelevant tags and attributes can make the data easier to understand and operate by other tools and algorithms.

Improve the accuracy of information extraction: Reducing interference factors can improve the information extraction model's ability to extract key information, thereby improving the accuracy of the extraction results.

Reduce the difficulty of model learning: Cleaning data can reduce the difficulty of learning information extraction models. Reducing unnecessary noise and interference can make it easier for the model to learn meaningful features and patterns.

S108: If the filling rate of the preset field in the first governance data is not less than the first threshold, determine whether the accuracy of the preset field in the first governance data is less than the second threshold.

In an embodiment of the present specification, if the fill rate of the preset field in the first governance data is not less than the first threshold, it is determined whether the accuracy of the preset field is less than the second threshold. The second threshold can be set according to the actual situation. For example, the second threshold is set to 80%. Regarding the method of evaluating the accuracy of the data in the preset field, it can be done by comparing with a reliable data source or using data verification rules. The calculation method of determining the accuracy rate can be determined by the ratio of the number of correct data to the total number of data.

S110: If the accuracy of the preset field in the first managed data is less than the second threshold, regularization management is performed on the padding data in the first managed data to obtain second managed data.

In the embodiment of the present specification, the fill-in data in the first governance data can be subjected to address splitting, amount extraction, date standardization, mobile phone number extraction, and preset symbol filtering through regular expressions to obtain the second governance data.

It should be noted that the above content can be implemented through the following specific implementation plans:

Use regular expressions to split addresses:

-Define common patterns for addresses, such as patterns that contain information such as province, city, district, street, etc.

- Use regular expressions to match and extract individual parts of an address.

Use regular expressions to extract amounts:

-Define common formats for amounts, such as formats with currency symbols, decimal points, thousands separators, etc.

-Use regular expressions to match and extract the numeric portion of an amount field.

Use regular expressions to normalize dates:

-Define common formats for dates, such as year-month-day, month/day/year, etc.

-Use regular expressions to match and convert dates to a uniform format, such as ISO format.

Use regular expressions to extract mobile phone numbers:

- Define common formats for mobile phone numbers, such as 11 digits.

-Use regular expressions to match and extract the mobile number field.

Use regular expressions to filter preset symbols:

-Define the list of symbols that need to be filtered.

- Use regular expressions to remove these symbols from your data.

Finally, the above processed results can be combined into the second governance data.

It should be noted that the embodiments of this specification have the following beneficial effects through the above contents:

Improve data quality: Through regular governance, noise, errors and inconsistencies in data can be removed to improve data accuracy and reliability.

Extract key information: Regular expressions can be used for address splitting, amount extraction, date standardization, mobile phone number extraction, etc., to help quickly and accurately extract key information from the data.

Data Standardization: Date standardization and filtering with preset symbols help convert data into a uniform format, making it easier to analyze and compare.

Enhanced data consistency: Regular governance can ensure that the populated data follows certain rules and patterns, thereby enhancing the consistency and repeatability of the data.

Simplified data processing: The processed second governance data is more standardized and concise, which is convenient for subsequent data analysis, storage and transmission.

Improve data availability: By improving data quality and consistency, formal governance makes data more valuable and better supports decision making and business processes.

S112: If the accuracy of the preset field in the second governance data is not less than the second threshold, qualified bidding data is obtained.

In the embodiment of the present specification, if the accuracy of the preset field in the second governance data is not less than the second threshold, the data is considered to meet the conditions, that is, qualified bidding data is obtained.

Furthermore, if the accuracy of the preset field in the second governance data is less than the second threshold, the preset large model can be used to extract information from the fill-in data in the second governance data to obtain third governance data; if the accuracy of the preset field in the third governance data is not less than the second threshold, qualified bidding data is obtained.

It should be noted that the above content can be implemented through the following specific implementation plans:

Use the preset big model to extract information: If the accuracy of the preset field in the second governance data is less than the second threshold, the preset big model is enabled. The preset big model uses machine learning or natural language processing technology to extract information from the fill data in the second governance data. The extracted information may include but is not limited to keywords, entities, relationships, etc.

The third governance data is obtained according to the extraction result: the information extracted from the preset large model is integrated into the second governance data to form the third governance data.

Detecting the accuracy of the preset field in the third governance data: re-detecting the accuracy of the preset field in the third governance data, and comparing the detected accuracy with the second threshold.

Determine qualified bidding data: If the accuracy of the preset fields in the third governance data is not less than the second threshold, mark these data as qualified bidding data. If the accuracy of the preset fields in the third governance data is still less than the second threshold, the data governance can be re-performed through the big model until the accuracy of the preset fields in the governance data is not less than the second threshold.

It should be noted that the large model used in the embodiments of this specification needs to ensure that the large model has been fully trained and optimized to improve the accuracy of information extraction.

It should be noted that the embodiments of this specification have the following beneficial effects through the above contents:

Improve the accuracy of data governance: By presetting large models for information extraction, the accuracy of data governance can be further improved. Large models usually have more powerful language understanding and information extraction capabilities, and can extract the required information from the populated data more comprehensively and accurately.

Reduce misjudgments and omissions: When the accuracy of the preset field in the second governance data is less than the threshold, using a large model for further processing can reduce the occurrence of misjudgments and omissions. The large model can use its learned patterns and knowledge to better handle complex and ambiguous information, thereby improving the quality of data governance.

Providing bidding data that is more in line with the conditions: By extracting information from the second governance data, the accuracy of the preset fields in the third governance data obtained is not less than the second threshold, which means that the obtained data is more in line with the bidding conditions and requirements. This helps to improve the quality and reliability of bidding data and provide stronger support for bidding decisions.

Enhance the availability and value of data: Accurate and compliant bidding data has higher availability and value. This data can be used for further analysis, comparison and decision-making, helping the tenderer to better understand the situation of potential suppliers and make wise choices.

Improve the efficiency and effectiveness of the bidding process: Accurate bidding data can reduce delays and problems caused by wrong or inaccurate data during the bidding process. At the same time, qualified data can better meet bidding needs and improve the effectiveness and efficiency of the bidding process.

Reduce risks and costs: By ensuring the accuracy and compliance of bidding data, the risks and costs caused by incorrect data can be reduced. For example, avoiding cooperation with unqualified suppliers can reduce potential legal risks and economic losses.

Improve data governance processes and methods: Identifying and improving deficiencies in the data governance process will help to continuously optimize data governance processes and methods. By introducing advanced technologies such as big models, the ability and level of data governance can be improved to adapt to changing business needs and data challenges.

Furthermore, when extracting information from the fill-in data in the second governance data through the preset big model to obtain the third governance data, the role of the big model can be defined first, and a task description for extracting information from the second governance data can be provided; then, according to the big model and the task description for extracting information from the second governance data, information can be extracted from the fill-in data in the second governance data to obtain the third governance data.

It should be noted that the above content can be implemented through the following specific implementation plans:

Clarify the roles and tasks of the big model:

Role definition: Determine the specific role of the big model in this process based on the characteristics of the governance data and the needs of information extraction. For example, if you need to extract keywords from the text, you can define the big model as a keyword extractor.

Task description: Describe the information extraction task in detail, including the type of information to be extracted (such as dates, names of people, places, etc.), extraction rules or patterns, etc. Make sure the task description is clear so that the big model can understand and execute it.

Use large models for information extraction:

The big model is called and the second governance data is used as input. The big model analyzes and processes the populated data according to the task description, and uses its built-in algorithms and models to extract the required information. The big model generates the extraction results, which are the third governance data.

Verify and evaluate the extraction results:

Verify and evaluate the third-party governance data generated by the large model. You can ensure that the extracted information is accurate by comparing it with reference data or other verification methods. Based on the evaluation results, adjust and optimize the parameters or task description of the large model to improve the extraction effect.

It should be noted that the embodiments of this specification have the following beneficial effects through the above contents:

Improve the accuracy of information extraction: By clarifying the role of the big model and providing a detailed task description, the big model can better understand the type and context of information to be extracted. This helps improve the accuracy of the extraction results and reduce misjudgments and omissions.

Reduced manual intervention: Defining the role of a large model can make the information extraction process more automated, reducing the need for manual intervention. This can improve work efficiency and save time and human resources.

Improve data quality: Accurate information extraction helps improve the quality of third-party governance data. This can make subsequent data analysis, decision-making, and business processes more reliable and effective.

Support better decision making: By providing more accurate and detailed information, third-party governance data can provide stronger support for decision making, helping decision makers make more informed choices and reduce the risk of making decisions based on inaccurate or incomplete information.

Enhance model interpretability: Clarifying the role and task description of large models helps enhance model interpretability. This allows people to better understand how the model makes decisions and extracts information, increasing trust and understanding of the model results.

Adapt to different data requirements: By flexibly defining the roles and task descriptions of the large model, it can adapt to different information extraction requirements and data characteristics. This makes the method more general and scalable.

It should be noted that the tender and procurement announcement information currently available on the market has the problem of excessive amount of information and inconsistent format. Only through high-quality data governance and the extraction of valuable data can we better promote the flow of market information and its efficient use.

Traditional manual recognition and processing simply cannot meet the requirements of high efficiency and high quality. Now with the emergence of big model technology, using artificial intelligence to assist data governance is a very effective and innovative solution.

For the large amount of non-standard bidding and procurement announcement text data collected, the data governance process of the embodiment of this specification is applied to output the required high-quality data set to provide an accurate data foundation for various scenario applications.

Glossary:

(1) Filling rate: The percentage of key fields with values in each type of announcement information to the total number of fields. Only when this ratio is reached can the subsequent governance be normal. Based on a large amount of data statistics, 70% is used as the threshold.

(2) Accuracy: The percentage of accurate key fields in each type of announcement information to the total number of fields. Only when this ratio is reached can it be stored in the inventory as a data resource. Based on a large amount of data statistics, 80% is the minimum threshold.

It should be noted that the key fields of the embodiments of this specification may be the preset fields mentioned above. The key fields are pre-set fields and can be set according to different scenarios.

The accuracy can be calculated by comparing a series of indicators. For example, the following special requirements:

Amount: cannot be a negative value, the winning bid amount must be less than or equal to the budget amount, and a warning review is required if it is less than 1 or greater than 10 million; Company name: cannot contain special characters or special words; Contact person: cannot contain special characters, must be Chinese characters, and a warning review is required if it is greater than 4; Contact information: numbers + special characters; Time and date: standard format.

The overall process is shown in Figure 2, which shows a structured governance process diagram. The overall process is divided into five parts, specifically:

(1) Original information: Receive the original information of the bidding and procurement announcement to be processed.

(2) Task control: Different governance tasks are called according to different information formats and businesses.

(3) Data governance: specific data governance processes and methods.

(4) Quality control: Determine the quality effect after treatment and decide whether to continue to improve treatment based on the quality effect.

(5) Data resources: storage of bidding and procurement data after governance is completed.

Information classification:

Announcement type: It is currently divided into 6 categories according to the bidding and procurement stage, including procurement intention, planned construction, bidding notice, bidding announcement, bidding results, and other announcements. Among them, bidding notices include procurement notices, prequalification announcements, prequalification results, demonstration opinions, and demand announcements; bidding announcements include open bidding announcements, invitation bidding announcements, inquiry announcements, competitive negotiation announcements, competitive consultation announcements, single source announcements, bidding announcements, and correction announcements; bidding results include winning bid announcements, transaction announcements, bid cancellation announcements, bid failure announcements, and result changes; other announcements include contract announcements, acceptance announcements, violation announcements, and others.

Key information: project information, purchasing unit information, supplier information, agency information, and review expert information. Project information includes 20 fields, purchasing unit information includes 9 fields, supplier information includes 8 fields, agency information includes 7 fields, and review expert information includes 1 field.

Overall overview: Data quality statistics are performed on the received source data, and different data governance tasks are initiated according to the different data quality. When the filling rate of key fields of source data is greater than 70%, the regular governance-large model governance-quality control task process is executed. When the filling rate of key fields of source data is less than 70%, the UIE model governance-regular governance-large model governance-quality control task process is adopted to accurately produce high-quality data sets.

Description of technology and process points:

(1) Statistical analysis of the filling rate of key fields of source data: Classify by announcement type, define the announcement type dictionary: type_info = {'Announcement type 1':'value1', 'Announcement type 2':'value2',,,}, determine all fields under the source data type, define the dictionary key_info = {'Key field 1':'value1', 'Key field 2':'value2',,,}, determine the key fields of the source data, nested loops traverse type_info and info, and cumulatively calculate the total data volume of the key fields of each announcement type and the total data volume of all fields to obtain the key field filling rate.

(2) Based on the fill rate of key fields, different preset governance tasks are automatically started through the BlockingScheduler().add_job() method.

Data governance: including regularization governance, information extraction model governance and general large model information extraction governance.

Regular governance:

Overall overview: Use regular expressions to split, intercept, match and other operations on source data. This includes address splitting, amount extraction, date standardization, mobile phone number extraction, and filtering of useless information such as special symbols.

Description of key technologies and process points:

(1) Address splitting: Use the detailed address and the INSTR function to find the location of key fields such as province, city, and district. Then use the SUBSTR function to split the detailed address to obtain independent key fields such as province, city, and district.

(2) Amount extraction: In a text expression that contains amount numbers, use the LEAST and Locate functions to find the position of the first number, use the GREATEST and CHAR_LENGTH functions to find the position of the last number, and finally use the SUBSTR function to separate the pure numbers in the text. Then, based on the amount unit information in the text, the accurate amount information is finally determined.

(3) Mobile phone number filtering: Refer to step (2) to extract pure digital information, and then combine LENGTH(phone_number)=11AND SUBSTRING(phone_number,1,2)IN('13','14','15','17','18') with REGEXP'^(13\d|14[57]|15[^4\D]|17[13678]|18\d)\d{8}$' to get the correct mobile phone number.

(4) Special symbol filtering: Use the replace function to remove special symbols from information items.

About information extraction model governance:

Overall overview: Use the Universal Information Extraction (UIE) model to extract specific information from HTML documents and return the information extraction results in a structured form. In order to improve the accuracy of the information extraction model, an HTML simplification method is designed to simplify the HTML original text, remove most of the HTML syntax structure, and retain the format symbols. While making full use of the format information in the HTML original text, by shortening the length of the HTML document and removing invalid information, the memory and time consumed by the model reasoning process are reduced, and the reasoning accuracy is improved.

Description of key technologies and process points:

(1) HTML simplification: Use the BeautifulSoup class in Python's bs4 module to simplify the HTML format, remove the script, style and other tags in the HTML document, and then remove the class, id and other attributes of other tags. The specific tags and attribute names to be removed can be designed according to actual needs.

(2) UIE information extraction model: The simplified HTML document is used as input and passed into the UIE information extraction model to extract the information contained in the bidding and procurement announcement, such as the name, address, and contact information of the purchaser, the name, address, and contact information of the agency, etc. The UIE information extraction model uses a pointer network to implement span extraction, thereby achieving the extraction of multiple tasks such as named entity recognition (NER), relation extraction (RE), event extraction (EE), and attribute sentiment extraction (ABSA). Through the structured schema director, different information extraction tasks and specific content to be extracted are implemented.

Information Extraction (UIE) model: This is a model used to extract information from text. It can identify and extract key elements and data from HTML documents.

Specific information: This refers to the specific data or elements that the model needs to extract. This specific information may depend on the specific application scenario.

Extraction: The model identifies and extracts specific information by analyzing the structure and content of HTML documents.

Structured form: The extracted information will be organized in a specific structure, such as a dictionary or other data structure containing fields and values.

Return information extraction results: The final model will return the extracted structured information to the caller or other processing program for subsequent use and analysis.

About general large model information extraction management:

Overall overview: For fields that the UIE model does not parse accurately, a large model can be used for information extraction. With the help of the language understanding and code understanding capabilities of the general large model, the task of extracting information from the bidding and procurement announcement is converted into a large model dialogue task through the prompt word engineering, so that the large model returns the information in the HTML document to the user in a structured form. Then, through a small amount of annotated data, LORA fine-tuning is used to further enhance the reasoning accuracy of the large model in the task of extracting information from the bidding and procurement announcement. Similar to 4.2, before using the large model for information extraction, the HTML document is simplified to improve reasoning efficiency and accuracy.

Description of key technologies and process points:

(1) HTML simplification: Same as the HTML simplification mentioned above.

(2) General large model: Use existing open source general large models as the basic model, such as qwen, chatglm, etc., and select parameter scales such as 1.5b, 7b, and 14b according to resource allocation. Use the language understanding and code understanding capabilities of the open source general large model to extract bidding and procurement announcement information.

(3) Prompt word engineering: Construct standardized prompt words, define the role you want the big model to play and the scope of its capabilities, provide background information and context to guide the big model, clearly point out the tasks you want the big model to perform, and give relevant question and answer examples.

(4) LoRa fine-tuning: Construct a small amount of samples with labeled information (100-1000 items), perform LoRa fine-tuning on the general large model, and merge the generated LoRa fine-tuning into the original large model. Use the existing labeled question and answer samples to further improve the accuracy and format standardization of the large model annotation.

Quality Control:

Overall overview: When a batch of source data goes through a complete data governance task process, data resources to be confirmed are generated, and quality statistical analysis is performed on the data resources to be confirmed. When the key field filling rate is greater than 70% and the key field accuracy is greater than 80%, the data resources are output. When the key field filling rate is lower than 70%, the UIE model is labeled with a data set, supervised and fine-tuned to improve the information extraction effect of the UIE model, and then the data governance task is restarted. When the key field accuracy is less than 80%, the regular governance process is corrected and maintained to improve the accuracy of regular parsing, and the prompt words of the large model are corrected to improve the reasoning effect of the large model on specific problems, and then the data governance task is restarted.

Description of key technologies and process points:

(1) Supervised fine-tuning: Use the labeled data set to fine-tune the general large model. Use the LoRa fine-tuning method to connect a network in parallel on the basis of the general large model to improve the model's reasoning ability while keeping the original model unchanged. Different LoRa fine-tuning methods can be connected in parallel for different labeling tasks to achieve lightweight fine-tuning conversion of the model.

(2) Prompt word correction: The prompt words are optimized again. The instructions are conveyed to the big model more clearly through the segmented description of the instruction task and the use of special symbols such as emphasis marks. The zero shot and few shot methods are used to write task examples into the prompt words to improve the reasoning effect of the big model on specific problems.

This process can greatly improve the efficiency of managing large amounts of data. After using this process, only 4 technicians are needed to manage tens of millions of data. Through model fine-tuning and algorithm application, real-time management of 2 announcements/s can be achieved on a single-card computing machine, with a management quality of more than 84%.

Figure 3 is a structural diagram of a bidding data filling and management device provided by one or more embodiments of this specification, including: a data filling unit 302, an unfilled field determination unit 304, a first management unit 306, an accuracy judgment unit 308, a second management unit 310 and a bidding data determination unit 312.

The data filling unit 302 fills the tender source data with data according to the preset fields to obtain tender filling data to be evaluated;

An unfilled field determining unit 304 is configured to determine unfilled preset fields of the tender filling data to be evaluated if the filling rate of the preset fields in the tender filling data to be evaluated is less than a first threshold;

The first management unit 306 inputs the unfilled preset field into a pre-trained information extraction model to obtain first management data;

The accuracy judgment unit 308 judges whether the accuracy of the preset field in the first governance data is less than a second threshold if the filling rate of the preset field in the first governance data is not less than a first threshold;

A second governance unit 310, if the accuracy of the preset field in the first governance data is less than the second threshold, performs regularization governance on the padding data in the first governance data to obtain second governance data;

The bidding data determining unit 312 obtains bidding data that meets the conditions if the accuracy of the preset field in the second governance data is not less than the second threshold.

FIG4 is a schematic diagram of the structure of a bidding data filling management device provided by one or more embodiments of this specification, including:

at least one processor; and,

a memory communicatively connected to the at least one processor; wherein,

The memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to:

Fill the tender source data according to the preset fields to obtain the tender filling data to be evaluated;

If the filling rate of the preset field in the tender filling data to be evaluated is less than a first threshold, determining the unfilled preset field of the tender filling data to be evaluated;

Inputting the unfilled preset fields into a pre-trained information extraction model to obtain first governance data;

If the filling rate of the preset field in the first governance data is not less than the first threshold, determining whether the accuracy rate of the preset field in the first governance data is less than the second threshold;

If the accuracy of the preset field in the first managed data is less than the second threshold, regularization management is performed on the padding data in the first managed data to obtain second managed data;

If the accuracy of the preset field in the second governance data is not less than the second threshold, qualified bidding data is obtained.

One or more embodiments of this specification provide a non-volatile computer storage medium storing computer executable instructions, which can achieve the following when executed by a computer:

Fill the tender source data according to the preset fields to obtain the tender filling data to be evaluated;

If the filling rate of the preset field in the tender filling data to be evaluated is less than a first threshold, determining the unfilled preset field of the tender filling data to be evaluated;

Inputting the unfilled preset fields into a pre-trained information extraction model to obtain first governance data;

If the filling rate of the preset field in the first governance data is not less than the first threshold, determining whether the accuracy rate of the preset field in the first governance data is less than the second threshold;

If the accuracy of the preset field in the first managed data is less than the second threshold, regularization management is performed on the padding data in the first managed data to obtain second managed data;

If the accuracy of the preset field in the second governance data is not less than the second threshold, qualified bidding data is obtained.

Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device, equipment, and non-volatile computer storage medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

The above description is only one or more embodiments of this specification and is not intended to limit this specification. For those skilled in the art, one or more embodiments of this specification may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of one or more embodiments of this specification shall be included in the scope of the claims of this specification.

Claims (10)

1. A bidding data filling management method, characterized by comprising: Fill the tender source data according to the preset fields to obtain the tender filling data to be evaluated; If the filling rate of the preset field in the tender filling data to be evaluated is less than a first threshold, determining the unfilled preset field of the tender filling data to be evaluated; Inputting the unfilled preset fields into a pre-trained information extraction model to obtain first governance data; If the filling rate of the preset field in the first governance data is not less than the first threshold, determining whether the accuracy rate of the preset field in the first governance data is less than the second threshold; If the accuracy of the preset field in the first managed data is less than the second threshold, regularization management is performed on the padding data in the first managed data to obtain second managed data; If the accuracy of the preset field in the second governance data is not less than the second threshold, qualified bidding data is obtained. 2. The method according to claim 1, characterized in that the step of inputting the unfilled preset fields into a pre-trained information extraction model to obtain the first governance data comprises: Extracting data from the unfilled preset fields through a pre-trained information extraction model to obtain data to be filled in the unfilled preset fields; The unfilled preset fields are filled with data according to the data to be filled to obtain first governance data. 3. The method according to claim 2 is characterized in that the tender filling data to be evaluated is an HTML format document, and the unfilled preset fields are extracted by a pre-trained information extraction model to obtain the data to be filled in the unfilled preset fields, including: The tender filling data to be evaluated is input into the information extraction model, and the information corresponding to the unfilled preset fields is extracted through a pointer network to achieve named entity recognition, relationship extraction, event extraction, and attribute sentiment extraction. 4. The method according to claim 3, characterized in that before extracting data from the unfilled preset fields through the pre-trained information extraction model to obtain the data to be filled in the unfilled preset fields, the method further comprises: The BeautifulSoup class in the bs4 module of Python is used to remove the script tag content and the style tag content in the tender filling data to be evaluated, and remove the class attributes and id attributes of other tags. 5. The method according to claim 1, wherein the performing regularization on the padding data in the first governance data to obtain the second governance data comprises: The fill data in the first governance data is subjected to address splitting, amount extraction, date standardization, mobile phone number extraction, and preset symbol filtering by regular expressions to obtain second governance data. 6. The method according to claim 1, characterized in that if the accuracy of the preset field in the second governance data is less than the second threshold, the method further comprises: Extracting information from the filling data in the second governance data by using a preset large model to obtain third governance data; If the accuracy of the preset field in the third governance data is not less than the second threshold, qualified bidding data is obtained. 7. The method according to claim 6, characterized in that the step of extracting information from the fill data in the second governance data by using a preset large model to obtain the third governance data comprises: Define the role of the large model and provide a task description for information extraction from the second governance data; According to the task description of extracting information from the second governance data based on the large model, information is extracted from the filling data in the second governance data to obtain third governance data. 8. A bidding data filling management device, characterized by comprising: A data filling unit fills the tender source data according to preset fields to obtain tender filling data to be evaluated; an unfilled field determining unit, which determines unfilled preset fields of the tender filling data to be evaluated if the filling rate of the preset fields in the tender filling data to be evaluated is less than a first threshold; A first management unit inputs the unfilled preset field into a pre-trained information extraction model to obtain first management data; an accuracy judgment unit, which judges whether the accuracy of the preset field in the first governance data is less than a second threshold value if the filling rate of the preset field in the first governance data is not less than a first threshold value; A second governance unit, if the accuracy of the preset field in the first governance data is less than the second threshold, performs regularization governance on the padding data in the first governance data to obtain second governance data; The bidding data determining unit obtains bidding data that meets the conditions if the accuracy of the preset field in the second governance data is not less than the second threshold. 9. A bidding data filling management device, characterized by comprising: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to: Fill the tender source data according to the preset fields to obtain the tender filling data to be evaluated; If the filling rate of the preset field in the tender filling data to be evaluated is less than a first threshold, determining the unfilled preset field of the tender filling data to be evaluated; Inputting the unfilled preset fields into a pre-trained information extraction model to obtain first governance data; If the filling rate of the preset field in the first governance data is not less than the first threshold, determining whether the accuracy rate of the preset field in the first governance data is less than the second threshold; If the accuracy of the preset field in the first managed data is less than the second threshold, regularization management is performed on the padding data in the first managed data to obtain second managed data; If the accuracy of the preset field in the second governance data is not less than the second threshold, qualified bidding data is obtained. 10. A non-volatile computer storage medium, characterized in that it stores computer executable instructions, which when executed by a computer can achieve: Fill the tender source data according to the preset fields to obtain the tender filling data to be evaluated; If the filling rate of the preset field in the tender filling data to be evaluated is less than a first threshold, determining the unfilled preset field of the tender filling data to be evaluated; Inputting the unfilled preset fields into a pre-trained information extraction model to obtain first governance data; If the filling rate of the preset field in the first governance data is not less than the first threshold, determining whether the accuracy rate of the preset field in the first governance data is less than the second threshold; If the accuracy of the preset field in the first managed data is less than the second threshold, regularization management is performed on the padding data in the first managed data to obtain second managed data; If the accuracy of the preset field in the second governance data is not less than the second threshold, qualified bidding data is obtained.