Intelligent bid evaluation method and system based on knowledge enhancement large language model

Abstract

The invention belongs to the technical field of digital bid evaluation, and provides an intelligent bid evaluation method and system based on a knowledge enhancement large language model.The method comprises the steps that firstly, a to-be-evaluated file library of project bid inviting and tendering files is established; then, files in the to-be-reviewed file library are analyzed, file slices are generated, and a reviewed file knowledge base is formed; then, text slices in the file knowledge base are retrieved and reviewed through key fields, and input cue words of the knowledge enhancement large language model are generated in combination with the key fields and the file slices; then, interacting the input cue word with a pre-trained knowledge enhancement large language model to obtain all key field information; then, scoring the bidding file according to a preset bid evaluation algorithm for all the obtained key field information; and finally, sorting the scores of all the bidding files in a descending order to generate an intelligent bid evaluation report. According to the method, the application of the knowledge-enhanced large language model in the aspects of bidding and tendering key information extraction and intelligent bid evaluation is realized, the flexibility and adaptability are relatively high, the influence of human factors on bid evaluation work is reduced, and the fairness and justice of the bid evaluation process and result are actively promoted.

Description

Technical Field

[0001] The present invention relates to the technical field of digital bid evaluation, and in particular to an intelligent bid evaluation method and system based on a knowledge-enhanced large language model. Background Art

[0002] Currently, the review of bidding documents mainly relies on manual review, which is not only time-consuming and labor-intensive, but also easily affected by the subjective judgment of the reviewer, which may lead to inconsistency and deviation in the review results. In addition, traditional text recognition and analysis tools often have low accuracy when processing specific bidding documents, and it is difficult to effectively identify and extract key information in the documents, such as contract terms, technical specifications, and bidding requirements.

[0003] With the increase in bidding activities, the main challenge facing the modern bidding process is how to improve the accuracy of document processing and the efficiency of decision-making. Although some automated tools, such as optical character recognition (OCR) technology, are currently used, these tools often have problems with insufficient accuracy and insufficient information extraction depth when processing highly complex bidding documents.

[0004] Therefore, the present invention aims to establish a system that can deeply analyze the information of bidding documents to improve the accuracy of recognition and the efficiency of information extraction, and promote the realization of intelligent bid evaluation.

[0005] Existing document parsing technologies mainly rely on OCR technology and basic keyword search to process bidding documents. Most of these technologies remain at the level of simple text conversion and lack a deep understanding of document structure and semantic content, resulting in unsatisfactory results when processing bidding documents with complex formats and professional terms, and cannot effectively perform logical parsing of contract terms, compliance checks, and intelligent bid evaluation. Summary of the invention

[0006] The purpose of the present invention is to propose an intelligent bid evaluation method and system based on a knowledge-enhanced large language model, which can efficiently extract multi-format target fields of bidding documents, greatly improve the efficiency of extracting key bid evaluation fields, and effectively increase the bid evaluation speed.

[0007] The technical solution adopted by the present invention to achieve the above-mentioned purpose is: An intelligent bid evaluation method based on a knowledge-enhanced large language model comprises the following steps: S1. Establish a library of project bidding documents to be reviewed; S2. Analyze the files in the to-be-reviewed file library, generate file slices, and form a review file knowledge base; S21. Perform text recognition on the documents stored in the review document library. Use document parsing technology to automatically scan the documents in the review document library and convert them into editable text format; S22. Based on the text slicing model, perform text slicing processing on the editable files in the review file library to obtain structured text data, store it in the file vector library, and form a review file knowledge base; Here, a text slicing model is used to analyze and structure the editable text generated by S21, dividing the editable text content into multiple text slices, and converting them into structured text data vectors and storing them in the knowledge base of the document to be reviewed.

[0008] S3: Generate input prompt words for knowledge-enhanced large language models; S31. Search the knowledge base of the documents to be reviewed through preset key fields to obtain text slices with a similarity higher than a specified threshold; the preset key fields include: bid price, bid project, project number, bidder, legal person of the bidding company, name of the legal person's ID card, validity period of the legal person's ID card, name of the safety license unit, validity period of the safety production license, name of the business license unit, legal representative of the business license, validity period of the business license, project duration (delivery period), name of the tenderer, validity period of the bid, amount of deposit, name of the basic deposit account opening license account, name of the legal representative of the basic deposit account opening license, project manager qualification certificate, type of letter of guarantee, validity period of the bid deposit; S32. Combining the preset key fields, the text slices obtained in S31, and the preset prompt words to generate input prompt words for the knowledge-enhanced large language model.

[0009] S4: Interact with the pre-trained knowledge-enhanced large language model based on the input prompt words generated in S3 to obtain all key field information; The pre-trained knowledge-enhanced large language model is an open source large model that has been trained on the Internet.

[0010] S5: Based on all the key field information obtained in S4, the bid document is scored according to the preset bid evaluation algorithm; The preset bid evaluation algorithm is to score the field according to whether the field has passed the test, where a passing test item is scored 10 points, and a failing test is scored 0 points, and the reference evaluation score of the bid document is accumulated; The inspection items include: whether the bid price exceeds the maximum price limit in the bidding documents, verification of the legal representative's ID card, verification of the safety license, verification of the business license, whether the project number is consistent with the bidding documents, whether the project name is consistent with the bidding documents, whether the project duration is greater than the requirements in the bidding documents, whether the bid validity period meets the requirements of the bidding documents, whether the project manager has the qualifications specified in the bidding documents, verification of the basic deposit account opening license, and verification of the bid electronic letter of guarantee.

[0011] S6: Based on the evaluation scores of S5, all bidding documents are sorted in descending order and an intelligent bidding evaluation report is generated.

[0012] The present invention also provides an intelligent bid evaluation system based on a knowledge-enhanced large language model, comprising a text slicing module, a knowledge-enhanced large language model module and a bid evaluation module; The text slicing module is used to parse the files in the library of files to be reviewed, generate file slices, and form a review file knowledge base.

[0013] Knowledge-enhanced large language model module, used to obtain all key field information; The bid evaluation module scores the bid documents based on key field information and the preset bid evaluation algorithm, sorts the scores of all bid documents in descending order, and generates an intelligent bid evaluation report.

[0014] The beneficial effects of the present invention are: The present invention has significant advantages in the processing and evaluation of bidding documents by integrating text slicing, knowledge base retrieval and large language models.

[0015] The present invention combines document parsing technology with a text slicing model, which not only improves the accuracy of text recognition, but also effectively processes and recognizes complex formats and professional terms in bidding documents. This combination ensures that structured key information can be extracted efficiently and accurately from documents.

[0016] By searching the knowledge base, the preset key fields are compared with the text slices to obtain the similarity between the key fields and the text slices. The system can then select the text slices with higher similarity. The knowledge base search ensures the relevance of the slice content with the key fields and can identify and extract key text slice information. In addition, by using the deep learning capabilities of the large language model, the system can further understand and process this information and provide more accurate and detailed data analysis.

[0017] This system can automatically evaluate and score bidding documents intelligently. This process is not only fast and reduces the need for manual operation, but also increases the transparency and fairness of the evaluation process. The automated scoring mechanism eliminates human bias and ensures the consistency of the evaluation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0019] The present invention will be further described below in conjunction with the accompanying drawings: An intelligent bid evaluation method based on a knowledge-enhanced large language model comprises the following steps: S1. Establish a library of project bidding documents to be reviewed.

[0020] S2. Analyze the files in the to-be-reviewed file library, generate file slices, and form a review file knowledge base; S21. Perform text recognition on the documents stored in the review document library. Use document parsing technology to automatically scan the documents in the review document library and convert them into editable text format; S22. Based on the text slicing model, perform text slicing processing on the editable files in the review file library to obtain structured text data, store it in the file vector library, and form a review file knowledge base; Here, a text slicing model is used to analyze and structure the editable text generated by S21, dividing the editable text content into multiple text slices, and converting them into structured text data vectors and storing them in the knowledge base of the document to be reviewed.

[0021] The text slicing model structured document data by identifying document text information through OCR, segmenting sentences and words according to Chinese and English punctuation marks, specifying the maximum number of sentences contained in each segment, splitting the text document into slices according to the number of sentences and storing them in a text slice list, and converting the text slice list into a vector list.

[0022] S3: Generate input prompt words for knowledge-enhanced large language models; S31. Search the knowledge base of the documents to be reviewed through preset key fields to obtain text slices with a similarity higher than a specified threshold; the preset key fields include: bid price, bid project, project number, bidder, legal person of the bidding company, name of the legal person's ID card, validity period of the legal person's ID card, name of the safety license unit, validity period of the safety production license, name of the business license unit, legal representative of the business license, validity period of the business license, project duration (delivery period), name of the tenderer, validity period of the bid, amount of deposit, name of the basic deposit account opening license account, name of the legal representative of the basic deposit account opening license, project manager qualification certificate, type of letter of guarantee, validity period of the bid deposit; S32. Combining the preset key fields, the text slices obtained in S31, and the preset prompt words to generate input prompt words for the knowledge-enhanced large language model.

[0023] S4: Interact with the pre-trained knowledge-enhanced large language model based on the input prompt words generated in S3 to obtain all key field information; Among them, enter the prompt word to obtain the fields in the bidding document, and assemble each field and the obtained value into a key-value pair to be returned as a json dictionary. If the accurate key field information is not obtained, do not return the approximate value, and return the corresponding field with an empty string. If the field is a validity period, it is best to obtain the start and end dates, or directly return the deadline.

[0024] The pre-trained knowledge-enhanced large language model is an open source large model that has been trained on the Internet.

[0025] S5: Based on all the key field information obtained in S4, the bid document is scored according to the preset bid evaluation algorithm; the preset bid evaluation algorithm is to score the field according to whether the field passes the test, with 10 points for passing a test item and 0 points for failing the test, and the reference evaluation score of the bid document is obtained by accumulating the score; the test items include: whether the bid price exceeds the maximum price limit of the bidding document, the legal representative's ID card verification (whether the legal representative's name matches the legal representative's ID card photo, whether the legal representative's ID card is within the validity period), safety license verification (whether the name of the safety production license unit is consistent with the bidder's name, whether the safety production license is within the validity period), business license verification (whether the business license unit name is consistent with the bidder's name, whether ... Whether the name of the legal representative of the business license is consistent with that of the legal representative of the bidder, whether the business license is within the validity period), whether the project number is consistent with that in the bidding documents, whether the project name is consistent with that in the bidding documents, whether the project duration is greater than the requirements in the bidding documents, whether the bidding validity period meets the requirements in the bidding documents, whether the project manager has the qualifications specified in the bidding documents, basic deposit account opening license verification (whether the basic deposit account name is consistent with that of the bidder, whether the legal representative is consistent with that of the bidder), electronic bidding guarantee verification (whether the amount of the guarantee is consistent in uppercase and lowercase, whether the amount of the guarantee is consistent with the requirements in the bidding documents, whether the payment time is within the validity period, and whether the bidder, project name, and guarantee type in the guarantee are consistent with the requirements in the bidding documents).

[0026] S6: Based on the evaluation scores of S5, all bid documents are sorted in descending order to generate an intelligent bid evaluation report. The report not only includes the score of each bid project, but also lists the basis for the score and the analysis results of key fields in detail, providing comprehensive and objective decision support for the bid evaluation committee. Example

[0027] After the designated bidding document is uploaded to the review document library, the key fields of the bidding document are obtained according to the key fields to be extracted: "Bid Price": "￥31720044.12", "Bidding Project": "Smart Water Comprehensive Management Platform Project in a Certain City", "Project Number": "ZHGHZB—2021—0813", "Bidder": "Beijing Technology Co., Ltd." "Legal person of the bidding company": "Wu Mouxiao", "Legal person ID card name": "Wu Mouxiao", "Validity period of legal person ID card": "Long term", "Security License Unit Name": "Beijing Technology Co., Ltd." "Validity period of safety production license": "May 22, 2022", "Business License Unit Name": "Beijing Technology Co., Ltd." "Legal representative of business license": "Wu Mouxiao", "Business license validity period": "Long-term", "Project duration (delivery time)": "6 months", "Name of the Tenderer": "Water Conservancy Capital Construction Project Department of a Certain City", "Bidding validity period": "90 days", "Deposit amount": "150,000 yuan", "Basic deposit account opening license account name": "Beijing Technology Co., Ltd.", "Name of the legal representative of the basic deposit account opening license": "Wu Mouxiao", "Project Manager Qualification Certificate": "Level 3 general contractor for water conservancy and hydropower engineering construction; Level 1 professional contractor for electronic and intelligent engineering", "Guarantee Type": "", "Bid bond validity period": "" } According to the requirements of the bidding documents regarding key fields, determine whether the content of the key fields obtained is qualified.

[0028] For example, if the requirement for "bid price" in the bidding document is "bid price ≤ 32 million yuan", and the content of the target field is "bid price 40 million yuan", then the content of the target field "bid price" is judged to be unqualified and the score is 0; For example, if the requirement for "project duration" in the bidding document is "project duration ≤ 6 months", and the content of the target field is "project duration ≤ 12 months", the content of the target field "project duration" is judged to be unqualified and the score is 0; For another example, if the requirement regarding "validity period of business license" in the bidding documents is "validity period of business license ≥ 12 months", and the content of the target field is "validity period of business license = long-term", then the content of the target field "validity period of business license" is judged to be qualified and the score is 10 points.

[0029] The reference evaluation score of the bidding document is obtained by accumulating the comprehensive scores of each item.

[0030] The above-mentioned embodiment utilizes the combination of document parsing and text slicing model to improve the accuracy of text recognition, effectively process and recognize the complex formats and professional terms in the bidding documents, and ensure efficient and accurate extraction of structured key information from the documents.

[0031] Then, through knowledge base retrieval, the preset key fields are compared with the text slices to obtain the similarity between the key fields and the text slices, and the text slices with higher similarity are screened out to identify and extract key text slice information. Finally, the deep learning capabilities of the large language model are used to further understand and process this information to obtain more accurate and detailed data analysis results.

[0032] In addition, the present invention also provides an intelligent bid evaluation system based on a knowledge-enhanced large language model based on the intelligent bid evaluation method: An intelligent bid evaluation system based on a knowledge-enhanced large language model, including a text slicing module, a knowledge-enhanced large language model module, and a bid evaluation module; The text slicing module is used to parse the files in the to-be-reviewed file library, generate file slices, and form a review file knowledge base; Knowledge-enhanced large language model module, used to obtain all key field information; The bid evaluation module scores the bid documents based on key field information and the preset bid evaluation algorithm, sorts the scores of all bid documents in descending order, and generates an intelligent bid evaluation report.

Claims

1. An intelligent bid evaluation method based on knowledge-enhanced large language model, characterized by: The following steps are involved: S1. Establish a library of project bidding documents to be reviewed; S2. Analyze the files in the to-be-reviewed file library, generate file slices, and form a review file knowledge base; S3: Generate input prompt words for knowledge-enhanced large language models; S4: Interact with the pre-trained knowledge-enhanced large language model based on the input prompt words generated in S3 to obtain all key field information; S5: Based on all the key field information obtained in S4, the bid document is scored according to the preset bid evaluation algorithm; S6: Based on the evaluation scores of S5, all bidding documents are sorted in descending order and an intelligent bidding evaluation report is generated.

2. According to claim 1, the intelligent bid evaluation method based on knowledge-enhanced large language model is characterized by: The specific steps of S2 forming the review document knowledge base include: S21. Perform text recognition on the documents stored in the review document library. Use document parsing technology to automatically scan the documents in the review document library and convert them into editable text format; S22. Based on the text slicing model, text slicing is performed on the editable files in the review document library to obtain structured text data, which is stored in the file vector library to form a review document knowledge base; the text slicing model analyzes and structures the editable text generated by S21, divides the editable text content into multiple text slices, and converts them into structured text data vectors and stores them in the knowledge base of files to be reviewed.

3. According to claim 1, the intelligent bid evaluation method based on knowledge-enhanced large language model is characterized by: The specific steps of generating the input prompt word include: S31. Search the knowledge base of the documents to be reviewed through preset key fields to obtain text slices with a similarity higher than a specified threshold; the preset key fields include: bid price, bid project, project number, bidder, legal person of the bidding company, name of the legal person's ID card, validity period of the legal person's ID card, name of the safety license unit, validity period of the safety production license, name of the business license unit, legal representative of the business license, validity period of the business license, project duration (delivery period), name of the tenderer, validity period of the bid, amount of deposit, name of the basic deposit account opening license account, name of the legal representative of the basic deposit account opening license, project manager qualification certificate, type of letter of guarantee, validity period of the bid deposit; S32. Combining the preset key fields, the text slices obtained in S31, and the preset prompt words to generate input prompt words for the knowledge-enhanced large language model.

4. According to claim 1, the intelligent bid evaluation method based on knowledge-enhanced large language model is characterized by: The pre-trained knowledge-enhanced large language model is an open source large model that has been trained on the Internet.

5. According to claim 1, the intelligent bid evaluation method based on knowledge-enhanced large language model is characterized by: The preset bid evaluation algorithm is to score the field according to whether the field passes the test, with 10 points for passing a test item and 0 points for failing the test, and the reference evaluation score of the bid document is accumulated.

6. According to claim 5, the intelligent bid evaluation method based on knowledge-enhanced large language model is characterized by: The inspection items include: whether the bid price exceeds the maximum price limit in the bidding documents, verification of the legal representative's ID card, verification of the safety license, verification of the business license, whether the project number is consistent with the bidding documents, whether the project name is consistent with the bidding documents, whether the project duration is greater than the requirements in the bidding documents, whether the bid validity period meets the requirements of the bidding documents, whether the project manager has the qualifications specified in the bidding documents, verification of the basic deposit account opening license, and verification of the bid electronic letter of guarantee.

7. An intelligent bid evaluation system based on knowledge-enhanced large language model, characterized by: It includes text slicing module, knowledge-enhanced large language model module and bid evaluation module; The text slicing module is used to parse the files in the to-be-reviewed file library, generate file slices, and form a review file knowledge base; Knowledge-enhanced large language model module, used to obtain all key field information; The bid evaluation module scores the bid documents based on key field information and the preset bid evaluation algorithm, sorts the scores of all bid documents in descending order, and generates an intelligent bid evaluation report.

Images