Bid invitation purchasing bid evaluation method and system based on big language model technology

Abstract

The invention provides a bid invitation purchasing bid evaluation method and system based on a large language model technology. The method comprises the following steps: S1, obtaining a bid invitation file and a bid document; s2, processing the bid invitation file based on an LLM large language model, and extracting a corresponding bid invitation term feature set; s3, processing the bidding file based on an LLM large language model, extracting bidding content entities based on the bidding file, and aligning the bidding content entities to corresponding bidding terms; s4, performing hard index analysis based on the bid invitation terms and the corresponding bidding content entities to obtain a hard index analysis result; s5, when the hard index analysis result passes, further performing matching analysis on the bid invitation terms and the corresponding bidding content entities based on different domain analysis engines to obtain domain analysis results corresponding to the bid invitation terms; and further performing consistency analysis according to the domain analysis result to obtain a supplier bidding matching degree comprehensive analysis result. The bid evaluation efficiency and the intelligent level of bid invitation purchase bid evaluation can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of large language model, and particularly relates to a bidding and purchasing evaluation method and system based on large language model technology. BACKGROUND

[0002] The scientificity and rationality of the bidding and purchasing strategy of an enterprise have an important influence on its economic benefits and high-quality development. At present, bidding and purchasing evaluation refers to the process of auditing, scoring, sorting and selecting the best bidder by multiple evaluation experts on the basis of the bidding documents, which is one of the important steps in the bidding and purchasing process.

[0003] Electronic bidding is a way of using the Internet and digital technology to manage, publish and process the bidding and purchasing process. With the increasing maturity of new generation information technology, emerging technologies such as big data, artificial intelligence and cloud computing have been widely applied in various industries. As one of the emerging technologies, the large language model can not only provide strong technical support for electronic bidding, but also provide new ideas and methods for each process of electronic bidding.

[0004] However, the current electronic bidding technical solution is usually in the stage of publishing and information collection, but for the specific evaluation process, the traditional evaluation method still adopts the artificial evaluation mode, which leads to the situation of insufficient objectivity and intelligence level, and also needs to consume a large amount of manpower, affecting the evaluation efficiency. SUMMARY

[0005] In view of the above technical problems, the present application aims to provide a bidding and purchasing evaluation method and system based on large language model technology.

[0006] The purpose of the present application is achieved by the following technical solution: In a first aspect, the present application provides a bidding and purchasing evaluation method based on large language model technology, comprising the following steps: S1 obtaining the bidding documents and the bidding documents; S2 processing the bidding documents based on the LLM large language model, and extracting the corresponding bidding term feature set, wherein the bidding term feature set comprises one or more bidding terms; S3 processing the bidding documents based on the LLM large language model, extracting the bidding content entity based on the bidding documents, and aligning the bidding content entity to the corresponding bidding term; S4 performing hard index analysis based on the bidding terms and the corresponding bidding content entity, and obtaining the hard index analysis result; S5 When the hard index analysis result passes, further matching analysis is performed on the bidding terms and the corresponding bidding content entity based on different field analysis engines, to obtain a field analysis result corresponding to the bidding terms; further consistency analysis is performed according to the field analysis result, to obtain a comprehensive analysis result of the supplier bidding matching degree.

[0007] Preferably, the method further comprises: S6 Based on the bidding matching degree analysis result of each supplier, the N suppliers with the highest bidding matching degree form a supplier candidate list.

[0008] Preferably, step S1 comprises: Obtaining a bidding document provided by a bidding party, wherein the bidding document comprises a technical specification and a procurement clause document, etc. Obtaining a bidding document provided by a bidding party, wherein the bidding document comprises a technical specification and a procurement clause document, etc.

[0009] The document format of the bidding document and the bidding document comprises PDF, DOCX, XLSX, JPEG, etc.

[0010] Preferably, step S2 comprises: Based on the LLM large language model, first, the unstructured data of the bidding document is extracted through the analysis tool, including the original text, paragraph mark and table, etc., and the unstructured data is converted into a processable text sequence; The text sequence is processed by the semantic understanding module for keyword matching or context analysis, to obtain the bidding term content; The structured bidding term information is outputted in combination with the term rule template and the fine-tuning model, wherein the bidding term information comprises the term item and the term content.

[0011] Preferably, step S3 comprises: Based on the LLM large language model, first, the unstructured data of the bidding document is extracted through the analysis tool, including the original text, paragraph mark and table, etc., and the unstructured data is converted into a processable text sequence; The text sequence obtained is subjected to key entity extraction based on the field fine-tuning semantic analysis model, to obtain the bidding content entity, wherein the bidding content entity comprises the entity category and the entity content; According to the obtained bidding content entity and the bidding term, the semantic similarity is calculated, and the bidding content entity and the bidding term are associated according to the semantic similarity, to obtain an alignment set.

[0012] Preferably, step S5 comprises: According to the set field, the corresponding field analysis engine is called respectively, wherein the set field includes a technical field, a financial field and a legal field, and the field analysis engine includes a technical field analysis engine, a financial field analysis engine and a legal field analysis engine; Based on the scoring model arranged in the field analysis engine, the scoring model is used for matching analysis on the bidding terms and the corresponding bidding content entity, and a corresponding field score is obtained, wherein the field score includes a technical field score, a financial field score and a legal field score; According to the field score, a consistency analysis is performed to obtain a comprehensive analysis result of the supplier bidding matching degree.

[0013] In a second aspect, the present application further provides a bidding and procurement evaluation system based on a large language model technology, which comprises a processor, wherein the processor is used for executing the bidding and procurement evaluation method based on the large language model technology as described in any one of the embodiments of the first aspect.

[0014] The bidding and procurement evaluation method and system based on the large language model technology can use the bidding document of the bidding party and the bidding document of the supplier as the basis, extract the corresponding bidding terms for the bidding document, and sequentially perform entity extraction on the bidding document based on the extracted bidding terms, so as to align the bidding content entity in the bidding document to the corresponding bidding terms, thereby improving the efficiency and intelligent level of the standardized processing of the bidding and tender data. Further, the hard index analysis on the aligned bidding terms and bidding content entity can analyze the index-based terms of the supplier, thereby screening the suppliers meeting the standard requirements. For the suppliers meeting the basic requirements, the multi-field analysis scheme is further used to analyze the non-index-based terms or the bidding content above the standard, analyze the adaptability of the supplier conditions to the terms from different field angles, and obtain a comprehensive analysis result, which can help to improve the objectivity and accuracy of the quantitative evaluation of the non-standardized bidding content entity based on the large language model, thereby improving the intelligent level and efficiency of the selection of the suppliers. BRIEF DESCRIPTION OF DRAWINGS

[0015] The present application is further described by using the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present application. Other drawings can be obtained by those skilled in the art without creative labor on the premise that the following drawings are not limited.

[0016] Figure 1 A flowchart of the bidding and procurement evaluation method based on the large language model technology shown in the embodiments; Figure 2 A logic framework diagram of the bidding and procurement evaluation method based on the large language model technology shown in the embodiments. DETAILED DESCRIPTION

[0017] The application is further described in combination with the following application scenarios.

[0018] Reference is made to Figure 1 , Figure 2 which shows a tender evaluation method based on large language model technology, comprising the following steps: S1 obtaining the tender file and the bid file; S2 processing the tender file based on the LLM large language model, extracting the corresponding tender clause feature set, wherein the tender clause feature set includes one or more tender clauses; S3 processing the bid file based on the LLM large language model, extracting the bid content entity based on the bid file, and aligning the bid content entity to the corresponding tender clause; S4 performing hard indicator analysis based on the tender clause and the corresponding bid content entity, obtaining the hard indicator analysis result; S5 when the hard indicator analysis result is passed, further performing matching analysis on the tender clause and the corresponding bid content entity based on different field analysis engines, obtaining the field analysis result corresponding to the tender clause; further performing consistency analysis according to the field analysis result, obtaining the comprehensive analysis result of the supplier bid matching degree.

[0019] Preferably, the method further comprises: S6 generating a supplier candidate list based on the bid matching degree analysis result of each supplier, the N suppliers with the highest bid matching degree.

[0020] The above embodiments of the application take the tender file of the tendering party and the bid file of the supplier as the basis, extract the corresponding tender clauses for the tender file, and sequentially extract the entities from the bid file as the basis, align the bid content entities contained in the bid file to the corresponding tender clauses, thereby improving the efficiency and intelligent level of the standardized processing of the bid and tender data. Further, the hard indicator analysis based on the aligned tender clauses and bid content entities can analyze the supplier's analysis of the index-based clauses, thereby screening out suppliers that meet the standard requirements; for suppliers that meet the basic requirements, the multi-field analysis scheme is further proposed based on the non-index-based clauses or the bid content above the standard, the adaptability of the supplier conditions to the clauses is analyzed from different field angles, and the comprehensive analysis result is obtained, which can help to improve the objectivity and accuracy of the quantitative evaluation of the non-standardized bid content entity content based on the large language model, thereby improving the intelligent level of the selection of suppliers.

[0021] Among them, large language models have excellent natural language processing capabilities. These models have undergone large-scale pre-training and learned the grammar, structure and semantics of the language from massive bidding and procurement data. With the help of large language models, bidding and procurement companies can more conveniently enter supplier data, combine these large amounts of data information, and establish a more scientific and effective bidding and tendering scoring index system to select the best suppliers, which is of great value to improving the electronic bidding and tendering management level of bidding and procurement companies.

[0022] Preferably, step S1 includes: Obtain the bidding documents provided by the tenderer, including technical specifications and procurement terms and conditions; Obtain the bidding documents provided by the supplier, which include the bidding letter, qualification certificate, financial statements, construction plan, quotation sheet, etc.

[0023] The document formats of bidding documents and tender documents include PDF, DOCX, XLSX, JPEG, etc.

[0024] During the bidding process, bidders and suppliers submit documents in various formats (such as PDF contracts, scanned documents, and Excel spreadsheets). The traditional method involves manually reading these documents and recording key information, which is time-consuming and error-prone. This invention automatically extracts structured text from bidding and tender documents based on a large language model, further extracting key information from them. This helps improve the efficiency and intelligence of document data processing.

[0025] Preferably, step S2 includes: Based on the LLM large language model, we first extract the unstructured data of the bidding documents, including original text, paragraph marks, and tables, through parsing tools, and convert the unstructured data into a processable text sequence; The semantic understanding module performs keyword matching or context analysis on the text sequence to obtain the content of the bidding terms; Combining the clause rule template and the fine-tuning model, structured bidding clause information is output, where the bidding clause information includes clause items and clause content.

[0026] In one scenario, the LLM large language model uses built-in OCR or PDF parsing tools to extract the raw text of bidding documents, identify structures such as paragraphs and tables, and convert unstructured data into processable text sequences. Using a self-attention mechanism and positional encoding technology, it performs keyword matching (such as "bid bond") and context analysis (such as determining the relevance of clauses). Combining bidding clause sentence structure rules with a domain-specific fine-tuning model, it outputs structured JSON results.

[0027] Based on the large language model combined with domain fine-tuning technology, intelligent tender clause extraction can be performed on the tender document, and the intelligent level of key clause information acquisition is improved.

[0028] Preferably, step S3 comprises: Based on the LLM large language model, first, the unstructured data of the tender document is extracted through the parsing tool, including original text, paragraph mark, and table, etc., and the unstructured data is converted into a processable text sequence; The obtained text sequence is subjected to key entity extraction based on the domain fine-tuned semantic parsing model, and the tender content entity is obtained, wherein the tender content entity includes entity category and entity content; According to the obtained tender content entity and the tender clause, semantic similarity calculation is performed, the tender content entity and the tender clause are associated according to the semantic similarity, and an alignment set is obtained.

[0029] In one scenario, the domain fine-tuned semantic understanding model specifically comprises: The Llama-based model is taken as a basis, a large amount of bidding and tendering file data and associated policy and regulation files are taken as training data to train a domain fine-tuned semantic parsing engine; the target text is extracted from the text sequence based on paragraph mark division or logical paragraph division, etc., wherein in the process of entity recognition on the target text, the key entity extraction is completed through the key entity recognition function set by the parsing engine; wherein the key entity recognition function is:

[0030] Wherein, represents the probability that the target text belongs to entity i, represents the trainable weight matrix of entity i, represents the 768-dimensional text feature vector obtained after the target text is processed based on the LLM text encoder; represents the bias term corresponding to entity i; represents that variable j belongs to entity set ; represents the trainable weight matrix of entity j, represents the bias term corresponding to entity j; According to the probability that the target text belongs to each entity, the entity corresponding to the maximum probability is selected as the key entity of the target text.

[0031] Wherein, the preset key entities include registered capital, ISO certification, historical litigation, technical standard, service scheme, and product parameter, etc.

[0032] In one scenario, the LLM large language model extracts the original text of the bidding document based on the built-in OCR or PDF parsing tool, identifies the structure of the paragraphs, tables, etc., and converts the unstructured data into a processable text sequence; the LLM semantic parsing engine based on domain fine-tuning extracts entities from the text sequence to obtain the entity category and corresponding entity content of the text sequence, and according to the obtained entity category, it is associated to the corresponding bidding clause item based on semantic similarity, and the corresponding bid text content and bidding clause text content are aligned to obtain an alignment set.

[0033] Further processing of the bid file based on the large language model can first extract entity content from the bid file, and further align the question content to the corresponding bidding clause based on the extracted entity content, realizing cross-file content matching of the bidding file and the bid file, and realizing automatic extraction of the supplier's response content to the bidding clause, laying a foundation for subsequent targeted matching analysis of the bidding clause and the corresponding bid content entity.

[0034] Preferably, step S4 comprises: Hard indicator analysis is performed on the bidding clause and the corresponding bid content entity, wherein the hard indicator analysis function used is:

[0035] wherein, represents the hard indicator deviation factor, represents the total number of bidding clauses, and variable k represents the corresponding kth bidding clause; represents the weight of the kth bidding clause, represents the clause attribute factor of the kth bidding clause, wherein when the kth bidding clause is a conditional clause, , otherwise when the kth bidding clause is a non-conditional clause, ; represents the BERT vector of the kth bidding clause content ; represents the BERT vector of the bid entity content corresponding to the kth bidding clause, represents the judgment function, when and the cosine similarity is greater than the set threshold , , otherwise when , ; According to the obtained hard indicator deviation factor , when is less than the set deviation threshold , the hard indicator analysis result is passed; otherwise when Greater than or equal to the set deviation threshold , the hard indicator analysis result is failed.

[0036] In one scenario, set the threshold , or set .

[0037] In one scenario, the weight of the bidding terms When the corresponding clause is a breach of contract clause or a standard regulation clause, The larger the value of , for example ; When the corresponding clause is a general clause, The smaller the value of , for example .

[0038] In one scenario, based on the hard indicator deviation factor The value range is , among which Indicates that it fully complies with the hard indicators. When, set the deviation threshold .

[0039] In one scenario, the tender documents for procurement or technical services contain conditional clauses (generally mandatory minimum standards that must be met) including independent legal person status (requiring bidders to be legal persons or other organizations with independent civil liability capabilities, and branches bidding must provide a letter of authorization and business license from the head office); technical team staffing (specifying the number or qualifications of experts involved in technical services); specific performance requirements (requiring bidders to provide two similar project performances completed in the past three years (e.g., hospital project experience is required for medical equipment procurement), but not limited to specific administrative regions or industries); proof of financial capacity (requiring bidders to provide audited financial statements or bank credit certificates showing that they have been in good financial condition and have not suffered significant losses in the past three years); technical qualification certification (specifying that the bid products must pass national mandatory certification (e.g., 3C certification) or industry-specific standards (e.g., ISO 13485 Medical Device Quality Management System); no-affiliation restrictions (organizations providing design and consulting services in the early stages of the bidding project are prohibited from participating in the bidding to prevent conflicts of interest); hardware equipment requirements (for example, the minimum server parameters must meet); material parameter requirements (for example, the supplied materials must meet the corresponding standards and parameter requirements), etc. Non-conditional clauses include optional accessory brands (the scoring terms require the brand of hardware, locks, and other accessories to be indicated, but bidders are allowed to choose other brands (providing proof of equivalent quality)); recommended technical parameters (allowing bidders to provide technical solutions "equivalent" to the parameters recommended in the bidding documents), etc.

[0040] The conditional clause acquisition can be calibrated in advance or calibrated adaptively according to historical standard data.

[0041] In a scenario, when the hard indicator analysis result of the supplier is not passed, the bid evaluation result of the supplier is directly obtained as not passed; and when the hard indicator analysis result of the supplier is passed, the analysis of the subsequent step is further entered.

[0042] In the above embodiment of the present application, when the hard indicator analysis is performed, the tender clause content and the corresponding tender entity content are converted into mathematical vectors, and whether the supplier meets the standard is intelligently analyzed according to the similarity of the vectors, so as to review the hard indicators such as the qualification of the supplier, which helps to improve the efficiency of the supplier review.

[0043] Considering that when the non-conditional review or the review of the part above the standard is performed, there is usually no unified standard, so that when the review of such content is performed, it usually needs to rely on the subjective evaluation of human beings, and the traditional expert review method also needs to gather experts in different fields to participate and review together, which leads to great difficulty in coordination and high labor cost. Although some technologies in the prior art are deep learning models to simulate the review of non-conditional indicators by experts, in the actual application process, since the model based on training cannot cover the consideration based on different situations in the actual situation, the objective and accuracy of the review result are insufficient.

[0044] In view of the above situation, the present application further particularly provides a technical scheme of objectively and professionally analyzing the supplier tender response content from different field perspectives based on the built multi-field analysis engine, so as to improve the objectivity and accuracy of the review of the non-indicator tender clause response content based on the large language model.

[0045] Preferably, the step S5 comprises: According to the set field, the corresponding field analysis engine is called respectively, wherein the set field comprises a technical field, a financial field and a legal field, and the field analysis engine comprises a technical field analysis engine, a financial field analysis engine and a legal field analysis engine; Based on the scoring model set in the field analysis engine, the scoring model is used to perform matching analysis on the tender clause and the corresponding tender content entity, to obtain the corresponding field score, wherein the field score comprises a technical field score, a financial field score and a legal field score; According to the field score, the consistency analysis is performed to obtain the supplier tender matching degree comprehensive analysis result.

[0046] The technical field analysis engine focuses on the technical feasibility of the bidding content entity of different bidding terms. The technical field analysis engine is trained based on product performance evaluation data, historical technical scheme evaluation report data, and corresponding evaluation results as a training data set, so as to score the technical field of the bidding content entity of the bidding term; The financial field analysis engine focuses on the economic / cost effect of the bidding content entity of different bidding terms. The technical field analysis engine is trained based on product cost benefit data, enterprise financial analysis data, and corresponding evaluation results as a training data set, so as to score the financial field of the bidding content entity of the bidding term; The legal field analysis engine focuses on the risk of violation of the bidding content entity of different bidding terms. The technical field analysis engine is trained based on standard regulations, historical judicial documents, and penalty case data, and corresponding evaluation results as a training data set, so as to score the legal field of the bidding content entity of the bidding term; The value range of the field score is 0-10; In one scenario, the field analysis engine is an LLM model with an analysis module. Different field analysis engines are equipped with LoRA adapters corresponding to the field. The LoRA adapter component is used to fine-tune the LLM large language model in the field, so that the LLM large language model can respond professionally to the corresponding field, and thus obtain the analysis and evaluation results of the corresponding field.

[0047] Preferably, the consistency analysis is performed according to the field score, specifically including: Obtaining a set of field scores of different field analysis engines on the bidding terms and the corresponding bidding content entities The set of field scores includes at least ; respectively represent the technical field score, the financial field score, and the legal field score of the current kth bidding term; The consistency quantification factor is calculated based on the field score, and the consistency quantification factor calculation function used is:

[0048] wherein, represents the consistency quantification factor, and the variable represents an element belonging to the field score set , wherein represents the total number of field scores in the field score set , wherein represents the average value of the field scores in the field score set , wherein represents the standard deviation of the field scores obtained from the historical analysis data for the kth bid term; represents the set maximum score difference value; According to the obtained consistency quantification factor , record the matching degree score corresponding to the kth bid term as the average value of the current set of field scores ; otherwise, when is greater than or equal to the set threshold value , mark the current kth bid term as a conflict, and record the matching degree score corresponding to the kth bid term as the minimum value of the current set of field scores . According to the matching degree scores of the bid terms, obtain the comprehensive analysis result of the supplier's bid matching degree.

[0049] In one scenario, the set maximum score difference value .

[0050] In one scenario, according to the matching degree scores of the bid terms, generate a comprehensive analysis report of the supplier's bid, wherein the comprehensive analysis report contains the matching degree scores of the supplier for different bid terms, as well as the corresponding conflict marks; wherein the conflict marks can enable professionals to review the conflict situation and thereby re-determine the corresponding matching degree scores. The comprehensive analysis report also contains the comprehensive matching degree score of the supplier, wherein the comprehensive matching degree score is obtained by weighted averaging, averaging, superimposing, or weighted superimposing, etc. of the matching degree scores for different bid terms.

[0051] ​​​​The above-mentioned embodiments of the present application simulate the perspectives of experts in different fields to evaluate the adaptability of suppliers to the bidding terms in different fields, train analysis models in different fields to perform matching analysis on the bidding content based on the perspectives of the professional fields, and complete field evaluation. Further, according to the evaluation results in different fields, consistency analysis and quantification are performed to extract the consistency and conflict characteristics of the evaluation results in different fields. When the consistency of the evaluation results based on the knowledge in different fields meets the standard, it is considered that the matching analysis results based on the knowledge in different fields have high reliability, and further, the final comprehensive evaluation result is obtained based on the field score. When the consistency analysis result does not meet the standard, it is considered that the matching analysis results in different fields have conflicts (at this time, a single result cannot reflect the objectivity of the analysis result), and further, the analysis result is marked for conflict to finally determine the matching evaluation result through expert analysis. By simulating the joint review method of experts in different fields in the above-mentioned manner, on the one hand, the objectivity and accuracy of the review can be improved, and on the other hand, the efficiency of the review can be improved.

[0052] The comprehensive matching degree score of the supplier for different bidding terms is obtained by setting rules, which can adapt to the needs of different bidding terms and adapt to the use in different bidding scenes.

[0053] In a scenario, for the comprehensive analysis result of the matching degree of the supplier bidding, only the part marked as conflict needs to be further reviewed by a multi-field expert meeting, which effectively reduces the human cost and efficiency of the review.

[0054] The present application also provides a bidding and procurement evaluation method based on a large language model technology, which comprises a processor, wherein the processor is used to execute the bidding and procurement evaluation method based on a large language model technology as shown in the above Figure 1 The specific embodiments corresponding to each step of the bidding and procurement evaluation method based on a large language model technology are not repeated in the present application.

[0055] It should be noted that each functional unit / module in each embodiment of the present application can be integrated in one processing unit / module, or each unit / module can exist physically, or two or more units / modules can be integrated in one unit / module. The above-mentioned integrated unit / module can be realized in the form of hardware or in the form of a software functional unit / module.

[0056] Those skilled in the art can clearly understand from the description of the above embodiments that the embodiments described herein can be implemented in hardware, software, firmware, middleware, code or any appropriate combination thereof. For hardware implementation, the processor can be implemented in one or more of the following units: application specific integrated circuit (ASIC), digital signal processor (DSP), digital signal processing device (DSPD), programmable logic device (PLD), field programmable gate array (FPGA), processor, controller, microcontroller, microprocessor, other electronic units designed to implement the functions described herein, or a combination thereof. For software implementation, part or all of the processes of the embodiments can be instructed by a computer program to relevant hardware. When implemented, the above program can be stored in a computer readable medium or transmitted as one or more instructions or codes on a computer readable medium. The computer readable medium includes computer storage medium and communication medium, wherein the communication medium includes any medium facilitating the transmission of computer programs from one place to another. The storage medium can be any available medium accessible by a computer. The computer readable medium can include, but is not limited to, RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage medium or other magnetic storage device, or any other medium capable of carrying or storing desired program codes in the form of instructions or data structures and accessible by a computer.

[0057] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, and are not limited to the scope of the present application. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should analyze that the technical solutions of the present application can be modified or replaced by equivalents without departing from the essence and scope of the present application.

Claims

1. A bidding and procurement evaluation method based on large language model technology, characterized in that: The steps include: S1 obtains the bidding documents and tender documents; S2 processes the bidding documents based on the LLM large language model and extracts the corresponding bidding clause feature set, where the bidding clause feature set includes one or more bidding clauses; S3 processes the bidding documents based on the LLM large language model, extracts bidding content entities based on the bidding documents, and aligns the bidding content entities with the corresponding bidding terms; S4 performs hard indicator analysis based on the bidding terms and corresponding bidding content entities to obtain hard indicator analysis results; S5 When the hard indicator analysis results are passed, the bidding terms and corresponding bidding content entities are further matched and analyzed based on different field analysis engines to obtain the field analysis results corresponding to the bidding terms; further consistency analysis is performed based on the field analysis results to obtain the comprehensive analysis results of the supplier bid matching degree.

2. The bidding and procurement evaluation method based on large language model technology according to claim 1 is characterized in that: The method further includes: S6 generates a supplier candidate list based on the bid matching analysis results of each supplier and the N suppliers with the highest bid matching degrees.

3. The bidding and procurement evaluation method based on large language model technology according to claim 1 is characterized in that: Step S1 includes: Obtain the bidding documents provided by the tenderer, including technical specifications and procurement terms; Obtain bidding documents provided by suppliers, including bidding letters, qualification certificates, financial statements, construction plans, and quotation sheets; The document formats of bidding documents and tender documents include PDF, DOCX, XLSX, and JPEG.

4. The bidding and procurement evaluation method based on large language model technology according to claim 1 is characterized in that: Step S2 includes: Based on the LLM large language model, we first use parsing tools to extract the unstructured data of the bidding documents, including raw text, paragraph tags, and tables, and convert the unstructured data into a processable text sequence; The semantic understanding module performs keyword matching or context analysis on the text sequence to obtain the content of the bidding terms; Combining the clause rule template and the fine-tuning model, structured bidding clause information is output, where the bidding clause information includes clause items and clause content.

5. The bidding and procurement evaluation method based on large language model technology according to claim 4 is characterized in that: Step S3 includes: Based on the LLM large language model, the unstructured data of the bidding documents, including original text, paragraph tags, and tables, is first extracted through parsing tools, and the unstructured data is converted into a processable text sequence; The domain-fine-tuned semantic parsing model is used to extract key entities from the obtained text sequence to obtain bidding content entities, where the bidding content entities include entity categories and entity content; Semantic similarity is calculated based on the obtained bidding content entities and bidding terms, and the bidding content entities and bidding terms are associated based on the semantic similarity to obtain an aligned set.

6. The bidding and procurement evaluation method based on large language model technology according to claim 5 is characterized in that: The semantic understanding model based on domain fine-tuning specifically includes: Based on the Llama model, a domain-tuned semantic parsing engine is trained using a large amount of bidding document data and related policy and regulatory documents. The target text is extracted from the text sequence based on paragraph mark division or logical paragraph division. During entity recognition of the target text, key entity extraction is completed through the key entity recognition function set by the parsing engine. The key entity recognition function is: in, represents the probability that the target text belongs to entity i, represents the trainable weight matrix of entity i, Represents the 768-dimensional text feature vector obtained after processing the target text based on the LLM text encoder; represents the bias term corresponding to entity i; Indicates that variable j belongs to the entity set ; represents the trainable weight matrix of entity j, represents the bias term corresponding to entity j; According to the probability that the target text belongs to each entity, the entity corresponding to the maximum probability is selected as the key entity of the target text.

7. The bidding and procurement evaluation method based on large language model technology according to claim 5 is characterized in that: Step S4 includes: A hard indicator analysis is performed on the bidding terms and the corresponding bidding content entities, and the hard indicator analysis function used is: in, represents the hard indicator deviation factor, represents the total number of tender clauses, and the variable k represents the corresponding k-th tender clause; represents the weight of the k-th bidding clause, represents the clause attribute factor of the kth tender clause, where when the kth tender clause is a conditional clause, , otherwise when the kth bidding clause is a non-conditional clause, ; Indicates the content of the k-th bidding terms BERT vectors; The BERT vector representing the bidding entity content corresponding to the k-th bidding clause, Represents the judgment function, when and Cosine similarity of Greater than the set threshold hour, , otherwise when hour, ; According to the obtained hard index deviation factor ,when Less than the set deviation threshold When the hard index analysis result is passed; otherwise, when Greater than or equal to the set deviation threshold , the hard indicator analysis result is failed.

8. The bidding and procurement evaluation method based on large language model technology according to claim 7 is characterized in that: Step S5 includes: According to the set fields, the corresponding field parsing engines are called respectively. The set fields include technical field, financial field and legal field. The field parsing engines include technical field parsing engine, financial field parsing engine and legal field parsing engine. Based on the scoring model set in the domain analysis engine, the bidding terms and corresponding bidding content entities are matched and analyzed by the scoring model to obtain the corresponding domain scores, where the domain scores include technical domain scores, financial domain scores, and legal domain scores; A consistency analysis is performed based on the field scores to obtain a comprehensive analysis result of the supplier bid matching degree.

9. The bidding and procurement evaluation method based on large language model technology according to claim 8 is characterized in that: The consistency analysis was conducted based on the domain scores, including: Get the domain score set of different domain parsing engines for bidding terms and corresponding bidding content entities , where the domain score set At least include ; They represent the technical field score, financial field score, and legal field score for the current k-th bidding clause respectively; The consistency quantification factor is calculated based on the domain score, and the consistency quantification factor calculation function used is: in, Represents the consistency quantification factor, variable Represented as belonging to the domain score set The elements in Represents a domain score set The total number of domain ratings in Represents a domain score set The average score of each field, represents the standard deviation of the historical domain scores for the k-th tender clause, which is obtained based on the standard deviation of the domain scores obtained from the historical analysis data for the k-th tender clause; Indicates the maximum score difference value set; According to the obtained consistency quantification factor ,when Less than the set threshold When the matching score of the k-th bidding clause is recorded, it is the current domain score set Average score for each area Otherwise, when Greater than or equal to the set threshold , the current k-th bidding clause is marked as conflicting, and the matching score of the current record corresponding to the k-th bidding clause is recorded as the current domain score set Minimum score for each field ; The comprehensive analysis results of supplier bid matching are obtained based on the matching scores of each bidding clause.

10. A bidding and procurement evaluation system based on large language model technology, characterized by: It includes a processor, wherein the processor is used to execute the bidding and procurement evaluation method based on large language model technology as described in any one of claims 1 to 9.

Images