Bid evaluation early warning method and system based on large language model

Abstract

The invention belongs to the technical field of artificial intelligence, and provides a bid evaluation early warning method and system based on a large language model.The method comprises the steps that firstly, an evaluation file library and a retrieval knowledge base are established, the knowledge base is retrieved through preset key fields, then key field related text slices are obtained, and then the key fields and the related text slices are combined, so that a bid evaluation result is obtained; the method comprises the following steps: generating a big language model interaction prompt word for interacting with a trained general big language model so as to automatically extract key field information, and then carrying out risk detection on each piece of key field information based on the obtained key field information and a preset risk assessment algorithm so as to generate a bid evaluation early warning report. According to the method, efficient bidding file information extraction, bid evaluation risk identification and early warning are realized, the bid evaluation process is optimized, and then the bid evaluation efficiency and the accuracy of the bid evaluation result are remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence technology, and relates to the field of digital bid evaluation technology, and specifically provides a method and system for bid evaluation early warning based on a large language model. Background Art

[0002] With the rapid development of information technology and the advent of the big data era, the number and complexity of bidding documents are increasing. At present, the review process of bidding documents mainly relies on manual review, which not only consumes a lot of time and energy, but is also inevitably affected by the reviewer's personal subjective judgment, resulting in instability and deviation in the review results. The traditional completely manual review method can no longer meet the actual needs of modern bidding. The efficiency, accuracy and fairness of bidding review are crucial to ensuring fair competition in the market and the smooth implementation of projects. Traditional text recognition and analysis tools are often not very accurate when processing specific content in bidding documents, and a large amount of data still needs to be manually checked, which cannot bring significant efficiency improvements to the bid evaluation process.

[0003] At present, traditional text recognition technology mainly relies on optical character recognition (OCR) technology, but this technology cannot directly process and analyze the recognized text. Existing information extraction technologies mainly include two types: rule-based methods and machine learning-based methods. Rule-based methods require manual definition of rule templates, which are suitable for scenarios with clear rules, but are difficult to adapt to complex text structures and changing text content. Machine learning-based methods rely on a large amount of annotated data for model training, but the acquisition cost of annotated data is high, and the generalization ability of the model is limited. The current risk assessment technology mainly relies on expert experience and statistical models. The expert experience method relies on the subjective judgment of the evaluation experts, which has great instability and deviation. The statistical model law requires a large amount of historical data for training, and it is difficult to effectively identify new risk points. Summary of the invention

[0004] The purpose of the present invention is to propose a bid evaluation warning method and system based on a large language model, which aims to efficiently and automatically extract key fields of bidding documents, significantly improve the efficiency of extracting key information in the bid evaluation process, thereby accelerating the bid evaluation speed and improving the accuracy of the bid evaluation work.

[0005] The technical solution adopted by the present invention to achieve the above-mentioned purpose is: A bid evaluation early warning method based on a large language model, comprising the following steps: S1: Establish a review document library; All bid documents to be reviewed are stored in a database to achieve centralized management and rapid retrieval of documents, providing a basis for subsequent automated processing; S2: Use document parsing technology and pre-trained text slicing models to automatically parse documents in the review document library to form a retrieval knowledge base for subsequent retrieval; Through document parsing technology, documents are converted into text format, and the text content is deeply analyzed and structured using the text slicing model; the processed text slices are then converted into vector form and stored in the document vector library to build a retrieval knowledge base; S3: Search the search knowledge base established in S2 through the preset key fields, and screen out text slices with high relevance by comparing the similarity between the key fields and the text slices; S4: Generate prompt words based on the key fields preset in S3, related text slices, and preset prompt words; S5: Based on the prompt words generated by S4 as the input field of the large language model, it interacts with the pre-trained large language model to accurately extract all key field information; S6: Based on all the key field information obtained in S5 and the preset risk assessment algorithm, all the key field information is subjected to risk detection. For the key field information that passes the detection, the key field information is returned normally and a pass detection label is added to the key field information; for the key field information that fails the detection, an early warning is given, and the key field information is returned with a corresponding early warning information label added; S7: Generate an early warning report based on the early warning prompt generated by S6, so that the bid evaluation experts can quickly identify the risk field information and locate and handle the abnormalities during the bid evaluation process.

[0006] Furthermore, the large language model is an open source large model that has been trained on the Internet.

[0007] Furthermore, the risk assessment algorithm includes: 1. Preset risk factors: define risk factors according to the bid evaluation criteria and specific project requirements; 2. Key field matching: Use rule matching or machine learning methods to make judgments based on the extracted key field information; 3. Risk score calculation: Based on the key field matching results, the risk score of each key field is calculated. The higher the score, the greater the risk level. This helps the bid evaluation experts identify high-risk key fields. 4. Risk label generation: Based on the risk score, set the risk threshold and generate warning labels for key fields that exceed the threshold.

[0008] Furthermore, the risk factors are based on an assessment of project goals, budgets, and technical solutions, covering all types of potential issues that may affect project success.

[0009] Furthermore, the rule matching determines whether the key fields meet the requirements through preset conditions; the machine learning method uses historical data to perform risk judgment through training models to improve the accuracy and adaptability of judgments.

[0010] Furthermore, the scoring criteria are set according to the severity and occurrence probability of the risk factors.

[0011] Furthermore, the setting of the risk threshold is adjusted according to the evaluation criteria of different projects to ensure the identification and timely warning of important risk points, so as to alert the evaluation experts and take necessary measures to avoid or reduce risks.

[0012] The present invention also provides a bid evaluation early warning system based on a large language model, comprising a text slicing module, a large language model module and a risk assessment 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 retrieval knowledge base; Large language model module, used to obtain all key field information; The risk assessment module performs risk detection on the review documents based on key field information and according to the preset risk assessment algorithm, adds warning information tags to all key field information, and generates a warning report.

[0013] The beneficial effects of the present invention are: The present invention significantly improves the efficiency, accuracy, and risk detection of bidding and tendering reviews; through automated text parsing and information extraction processes, manual participation is significantly reduced, thereby speeding up the processing of bidding and tendering documents. Compared with the traditional review process that relies entirely on manual labor, the present invention can assist evaluators in completing the review of more documents in a shorter time. By shortening the review cycle, relevant departments and personnel can use the saved time to handle other key tasks, thereby improving overall work efficiency.

[0014] The present invention utilizes a large language model and text slicing technology to achieve a deep understanding of text content and accurately extract key field information from complex bidding documents. This deep understanding exceeds the capabilities of traditional OCR technology and ensures high accuracy of information extraction.

[0015] Through the preset risk assessment method and intelligent analysis of the large language model, the present invention can identify potential problems and risk points, issue early warnings in time, and help reviewers take preventive or countermeasures; the early warning reports and detailed risk assessment results provided can provide powerful decision-making support tools for evaluation experts, helping them to make more reasonable and accurate judgments. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1is a flow chart of the method of the present invention; DETAILED DESCRIPTION

[0017] In order to better illustrate the technical solution of the present invention, the bid evaluation early warning method based on the large language model is further described below in conjunction with the accompanying drawings and specific embodiments: A bid evaluation early warning method based on a large language model, such as Figure 1 As shown, the steps include: S1: Establish a review document library; All bid documents to be reviewed are stored in a database. These documents are usually in PDF format. By building a document library, centralized management and rapid retrieval of documents are achieved, providing a solid foundation for subsequent automated processing.

[0018] S2: Use document parsing technology and pre-trained text slicing models to automatically parse documents in the review document library to form a knowledge base for subsequent retrieval; Through OCR technology, the document is converted into text format, and the text content is deeply analyzed and structured using a text slicing model, i.e., slicing, to obtain structured text slices. The processed text slices are then converted into vector form and stored in a document vector library to build a retrieval knowledge base.

[0019] S3: Search the knowledge base established in S2 through the key fields preset by the system. Filter out text slices with higher relevance by comparing the similarity between the key fields and the text slices; S4: Generate prompt words based on the system preset fields, relevant text slices and preset prompt words in step 3; S5: Based on the prompt words generated in step 4 as the input fields of the large language model, interact with the pre-trained large language model to accurately extract the information of all key fields; S6: Based on the key field information obtained in step 5 and the preset risk assessment algorithm, each field is tested for risk. For fields that pass the test, the system returns the field information normally and adds a pass test label to the field; for fields that fail the test, the system issues an early warning prompt and returns the field information and the corresponding early warning prompt information label.

[0020] The risk assessment algorithm is: 1. Preset risk factors: Based on the bid evaluation criteria and specific project requirements, define a series of risk factors, such as project schedule infeasibility, budget overrun risk, and ambiguity in contract terms. These risk factors are based on the evaluation of project goals, budget, technical solutions, etc., and cover various potential problems that may affect project success; 2. Key field matching: For the extracted key fields, use rule matching or machine learning methods to make judgments. Rule matching can determine whether the key fields meet the requirements based on preset conditions, while machine learning methods can make more flexible risk judgments by training models and using historical data to improve the accuracy and adaptability of judgments; 3. Risk score calculation: Calculate the risk score of each key field based on the key field matching results. The scoring criteria can be set based on the severity of the risk factor, the probability of occurrence, etc. The higher the score, the greater the risk, which helps the bid evaluation experts identify high-risk key fields; 4. Risk label generation: Based on the risk score, set the risk threshold and generate warning labels for key fields that exceed the threshold. The setting of risk threshold can be adjusted according to the evaluation criteria of different projects to ensure timely warning of important risk points, so as to prompt the evaluation experts to pay attention and take necessary measures to avoid or reduce risks.

[0021] S7: Generate an early warning report based on the early warning prompts generated by S6, so that bid evaluation experts can quickly identify risk fields and locate and handle abnormalities during the bid evaluation process. The early warning report significantly improves the efficiency and accuracy of bid evaluation work and provides strong auxiliary support for bid evaluation experts.

[0022] A bid evaluation early warning system based on a large language model, comprising a text slicing module, a large language model module and a risk assessment 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 retrieval knowledge base; Large language model module, used to obtain all key field information; The risk assessment module performs risk detection on the review documents based on key field information and according to the preset risk assessment algorithm, adds warning information tags to all key field information, and generates a warning report. Example

[0023] In the bidding process of a large-scale construction project, bid evaluation experts need to extract and analyze key information from multiple bid documents to evaluate the compliance and risk level of each bid document. Through the method of the present invention, the system first establishes a review document library and stores all bid documents to be reviewed (mainly in PDF format) in the database. Then, the system uses OCR document parsing technology to convert the bid documents into text format, and performs in-depth analysis through a pre-trained text slice model to generate structured text slices, and stores these text slices as vectors to form a searchable knowledge base.

[0024] Next, the system retrieves text slices from the knowledge base based on preset key fields, such as "project duration", "budget amount", "quality assurance clause", etc., and screens out text content closely related to the key fields by comparing the similarity between the key fields and the text slices. Based on the screened text slices, the system generates prompt words and interacts with the large language model to accurately extract all information related to the key fields.

[0025] After extracting these key field information, the system uses the preset risk assessment algorithm to test it, for example, to determine whether the project duration is in line with reality, whether the budget is reasonable, whether the quality assurance clauses are complete, etc. For fields that pass the risk test, the system returns normally and adds a pass test label to them; for fields that fail the risk test, the system generates a warning prompt and adds a corresponding warning prompt label to the returned field information.

[0026] Finally, the system generates an early warning report based on the detection results of all key fields, listing in detail the fields with potential risks and corresponding prompt information.

[0027] By viewing the early warning report, the bid evaluation experts can quickly identify the risk information in the bid documents, and then locate and handle the abnormalities. For example, if the "project duration" of a bid document is marked as high risk, the experts can further analyze whether the document has a problem of too tight schedule.

[0028] The present invention creates a bid evaluation warning method and system that integrates text slicing technology, knowledge base retrieval function and large language model, which can effectively extract key information fields and timely warn of abnormal information. This invention aims to help bid evaluation experts quickly identify potential risks, assist them in completing bid evaluation work, achieve efficient bidding document information extraction, bid evaluation risk identification and warning, optimize the bid evaluation process, and significantly improve bid evaluation efficiency and accuracy.

Claims

1. A bid evaluation early warning method based on a large language model, characterized by: steps include: S1: Establish a review document library; All bid documents to be reviewed are stored in a database to achieve centralized management and quick retrieval of review documents; S2: Use document parsing technology and pre-trained text slicing models to automatically parse the documents in the review document library established in S1 to form a knowledge base for subsequent retrieval; Through document parsing technology, documents are converted into text format, and the text slicing model is used to perform in-depth analysis and structural processing on the text format documents; the processed text slices are then converted into vector form and stored in the document vector library to build a retrieval knowledge base; S3: Search the search knowledge base established in S2 through the preset key fields, and screen out text slices with high relevance by comparing the similarity between the key fields and the text slices; S4: Generate prompt words based on the key fields preset in S3, related text slices, and preset prompt words; S5: Based on the prompt words generated by S4 as the input field of the large language model, it interacts with the pre-trained large language model to accurately extract all key field information; S6: Based on the key field information obtained in S5 and the preset risk assessment algorithm, risk detection is performed on each key field information. For key field information that passes the detection, the key field information is returned normally and a pass detection label is added to the key field information; for key field information that fails the detection, an early warning prompt is given, and the key field information is returned with a corresponding early warning prompt information label added; S7: Generate an early warning report according to the early warning prompt generated in S6.

2. The bid evaluation early warning method based on a large language model according to claim 1 is characterized in that: The large language model is an open source large model that has been trained on the Internet.

3. The bid evaluation early warning method based on a large language model according to claim 1 is characterized in that: The risk assessment algorithm includes: 1) Preset risk factors: define risk factors based on bid evaluation criteria and specific project requirements; 2) Key field matching: Use rule matching or machine learning methods to make judgments based on the extracted key field information; 3) Risk score calculation: Based on the key field matching results, the risk score of each key field is calculated. The higher the score, the greater the risk level. This helps the bid evaluation experts identify high-risk key fields. 4) Risk label generation: Based on the risk score, set the risk threshold and generate warning labels for key fields that exceed the threshold.

4. The bid evaluation early warning method based on a large language model according to claim 3 is characterized by: The risk factors are based on an assessment of project goals, budget, and technical solutions, and cover all potential issues that may affect project success.

5. The bid evaluation early warning method based on a large language model according to claim 3 is characterized by: The rule matching determines whether the key fields meet the requirements through preset conditions; the machine learning method uses historical data to perform risk judgment through training models to improve the accuracy and adaptability of judgments.

6. The bid evaluation early warning method based on a large language model according to claim 3 is characterized by: The scoring criteria are set according to the severity and occurrence probability of the risk factors.

7. The bid evaluation early warning method based on a large language model according to claim 3 is characterized by: The risk threshold setting is adjusted according to the evaluation criteria of different projects to ensure timely identification and early warning of important risk points, so as to draw the attention of evaluation experts and take necessary measures to avoid or reduce risks.

8. A bid evaluation early warning system based on a large language model, characterized by: Includes text slicing module, large language model module and risk assessment 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 retrieval knowledge base; The large language model module is used to obtain all key field information; The risk assessment module performs risk detection on the review documents based on the key field information and according to a preset risk assessment algorithm, and adds warning prompt information tags to all key field information to generate a warning report.

Images