LLM dynamic context resolution method for ambiguous terms of bid invitation document
By combining an LLM large model with external information to dynamically resolve ambiguous clauses, the problems of time-consuming manual resolution and poor applicability of traditional NLP tools in existing technologies are solved, achieving high-precision and high-efficiency parsing of ambiguous clauses in tender documents.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-01-28
- Publication Date
- 2026-04-07
AI Technical Summary
When dealing with ambiguous clauses in bidding documents, existing technologies rely on time-consuming manual resolution, which is easily affected by subjective factors. Traditional NLP tools have poor applicability and cannot dynamically adjust the interpretation based on the project background, resulting in incomplete resolution of ambiguities.
The algorithm employs a large LLM model combined with a pre-built database of historical similar tender documents, industry standards, and a high-frequency ambiguous trigger word database in the bidding field. Through ambiguous clause detection, project feature extraction, and semantic similarity calculation, it performs ambiguity resolution reasoning based on Prompt input and outputs accurate interpretations.
It achieves high-precision and high-efficiency parsing of ambiguous clauses in bidding documents, completely eliminating ambiguity and meeting the requirements of high precision and high efficiency.
Smart Images
Figure CN121809409A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of Natural Language Processing (NLP), and in particular to an LLM dynamic context resolution method for ambiguous clauses in tender documents. Background Technology
[0002] As the core basis for bidding activities, the accuracy of the terms in the tender documents directly affects the quality of bid responses and the compliance of the bidding process. Ambiguous clauses are common in tender documents, typically manifested as vague wording, missing definitions, and unclear boundaries. For example, vague wording like "similar project experience" lacks clarity regarding project scale, amount, and completion time, leading to ambiguity. Missing definitions, such as "high-quality supplier," are unclear because they don't provide specific evaluation indicators (e.g., qualification level, performance record). Unclear boundaries, such as "circumstances for extension of the construction period," lack specific triggering conditions (e.g., weather, policy changes), leading to ambiguity.
[0003] In existing technologies, ambiguous clauses in bidding documents are resolved through the following methods. One method is manual resolution by professionals based on their experience. However, manual resolution is time-consuming and susceptible to subjective factors, resulting in a high error rate (≥15%) when processing in batches. Another method is to use traditional NLP tools to identify and interpret ambiguities based solely on preset keywords (such as "similar" and "high-quality"). This static rule approach has significant limitations, as it is based only on the current document and does not thoroughly resolve ambiguities. It also cannot dynamically adjust the interpretation based on the project context (such as "municipal engineering" vs. "equipment procurement"), leading to poor applicability.
[0004] Therefore, it is essential to provide an LLM dynamic context resolution method for ambiguous clauses in tender documents to overcome the shortcomings of existing technologies. Summary of the Invention
[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide an LLM dynamic context resolution method for ambiguous clauses in tender documents. This method can associate external information such as historical tender cases and industry standards to achieve high-precision and high-efficiency parsing of ambiguous clauses in tender documents.
[0006] The objective of this invention is achieved through the following technical measures.
[0007] A dynamic context resolution method based on LLM (Library Management Model) is provided for resolving ambiguous clauses in tender documents, which involves the following steps: S1, inspect the contents of the tender documents, identify ambiguous clauses, and mark the types of ambiguous clauses; S2, extract the project features from the bidding documents and perform a dual-database search in a pre-built historical similar bidding document database and an industry standard database; S3 uses an LLM large model to perform ambiguity resolution reasoning and outputs the interpretation of ambiguous clauses.
[0008] Preferably, the above-mentioned LLM dynamic context resolution method for ambiguous clauses in tender documents, S1 specifically includes: S11, the content of the tender document is split into paragraphs, and ambiguous trigger words are detected for each paragraph; the ambiguous trigger words detected can be based on a high-frequency ambiguous trigger word library in the tendering field.
[0009] S12, when a paragraph contains ambiguous trigger words, the paragraph is split into phrases according to punctuation marks, and the clauses containing keywords are selected from the phrases; S13, perform semantic similarity calculation between the text of the clause containing the keyword and the unambiguous standard expression library. Set the cosine similarity threshold to <0.6. If it is lower than the threshold, it is judged as an ambiguous clause, and the "ambiguous clause type" of the ambiguous clause is obtained.
[0010] Preferably, the above-mentioned LLM dynamic context resolution method for ambiguous clauses in tender documents, S2 specifically includes: S21, extract the project characteristics of the current tender document, specifically by parsing the document cover and project overview section using LLM to obtain "project type", "budget size" and "industry". S22, based on the "project type" and "budget size" obtained in S21, perform historical case matching in the bidding document database to obtain the corresponding "ambiguous clauses - clarification" data; S23. Based on the "industry" obtained in S21 and the "ambiguous clause type" obtained in S13, match them in the industry standard library to obtain the corresponding industry standard requirements.
[0011] Preferably, in the above-mentioned LLM dynamic context resolution method for ambiguous clauses in tender documents, S3 specifically combines the original text of the ambiguous clause, project characteristics, and contextual support information, and inputs it into the LLM large model through Prompt for scenario-based reasoning, outputting an accurate interpretation.
[0012] Preferably, the LLM dynamic context resolution method for ambiguous clauses in tender documents described above adopts a three-part Prompt design consisting of "instruction + context + constraint".
[0013] Preferably, in the above-mentioned LLM dynamic context resolution method for ambiguous clauses in tender documents, the Prompt statement specifically refers to: "Based on the following information, please resolve the ambiguous clauses in the tender documents into unambiguous statements, and clarify the specific standards for the vague statements (such as scale, time, indicators)." # Original text of ambiguous clauses # Project Features Project type is Budget size The industry it belongs to is .
[0014] # Historical Cases # Industry Standards Constraints: The interpretation must conform to industry practices, match the current project budget and type, and be concise and clear.
[0015] # Output Format Related project characteristics: Refer to the context: Generate interpretation: ".
[0016] Preferably, the LLM dynamic context resolution method for ambiguous clauses in tender documents described above uses the Qwen-32B large model or the Qwen2.5-32B, Qwen3-302B-A3B, or Qwen3-32B large model.
[0017] Furthermore, the aforementioned LLM dynamic context resolution method for ambiguous clauses in tender documents, which pre-constructs a library of similar historical tender documents, is implemented through the following steps: A1. Collect ambiguous clauses in publicly tendered documents from relevant industries for the past 3-10 years; A2, each tender document is split into multiple data entries based on ambiguous clauses, and then verified; The data format is: Ambiguous clause (original text): Official clarification and interpretation: Basic project information: A3, label the "Project Type", "Budget Scale" and "Industry" in the tender documents to obtain tag data; A4. Store the ambiguous clause data from step 2) and the label data from step 3) together in the tender document database for later use.
[0018] Furthermore, the aforementioned LLM dynamic context resolution method for ambiguous clauses in tender documents utilizes a pre-built industry standard library through the following steps: B1 contains the necessary industry bidding standards and local bidding rules; B2, mark the "industry" and "ambiguous clause type" in the bidding standards; B3 stores the bidding standard content, "industry", and "ambiguous clause types" data together in the industry standard database for future use.
[0019] Furthermore, the aforementioned LLM dynamic context resolution method for ambiguous clauses in bidding documents, which pre-constructs a high-frequency ambiguity triggering thesaurus and an unambiguous standard expression library in the bidding field, is implemented through the following steps: C1, by analyzing frequently used ambiguous trigger words in the bidding and tendering field, a high-frequency ambiguous trigger word database for the bidding and tendering field is obtained; including three types: Vague qualifiers: similar, related, high-quality, appropriate, reasonable; Missing definition words: qualified, up to standard, meet requirements, of the same scale; Vaguely defined terms: in principle, depending on the situation, when necessary, relevant parties; C2, collect unambiguous standard expression phrases corresponding to high-frequency ambiguous trigger words, and mark the "ambiguous clause type"; C3 uses the Sentence-BERT model to extract feature vectors from the unambiguous standard expression phrases obtained in step C2, and stores them using Postgres. The storage format is phrase text, ambiguous clause type, and phrase vector, resulting in an unambiguous standard expression library.
[0020] This invention presents an LLM dynamic context resolution method for ambiguous clauses in tender documents. By accurately identifying ambiguous clauses, it can associate them with external information such as historical tender cases and industry standards, achieving high-precision and high-efficiency parsing of ambiguous clauses in tender documents and thoroughly eliminating ambiguity. It provides a complete solution encompassing "accurate ambiguity identification - dynamic context expansion - scenario-based reasoning," meeting the high-precision and high-efficiency requirements of tender document parsing. Attached Figure Description
[0021] The invention will be further described with reference to the accompanying drawings, but the contents of the drawings do not constitute any limitation on the invention.
[0022] Figure 1 This is a flowchart of an LLM dynamic context resolution method for ambiguous clauses in tender documents according to the present invention. Detailed Implementation
[0023] The present invention will be further described in conjunction with the following embodiments.
[0024] Example 1 A dynamic context resolution method for LLM (Limited Language Management) for ambiguous clauses in tender documents, such as... Figure 1 As shown, the procedure is as follows.
[0025] The method in this embodiment utilizes a pre-built database of historical similar tender documents, an industry standard database, and a database of high-frequency ambiguous trigger words in the bidding and tendering field.
[0026] The construction of a historical database of similar tender documents is carried out through the following steps: A1. Collect ambiguous terms and conditions from publicly tendered documents in relevant industries over the past 3-10 years. For example, you can collect ambiguous terms and conditions from publicly tendered documents in various industries (construction, equipment procurement, service outsourcing, etc.) over the past 5 years.
[0027] A2: Each tender document is split into multiple data entries based on ambiguous clauses, and then verified.
[0028] The data format is: Ambiguous clause (original text): Official clarification and interpretation: Basic project information: A3, label the “Project Type”, “Budget Scale” and “Industry” of the tender documents to obtain label data.
[0029] A4. Store the ambiguous clause data from step 2) and the label data from step 3) together in the tender document database for later use.
[0030] The pre-built industry standard library is constructed through the following steps: B1 contains the necessary industry bidding standards (such as the "Administrative Measures for Bidding on Construction of Housing and Municipal Infrastructure Projects") and local bidding rules; B2, mark the "industry" and "ambiguous clause type" in the bidding standards; B3 stores the bidding standard content, "industry", and "ambiguous clause types" data together in the industry standard database for future use.
[0031] The pre-built high-frequency ambiguity trigger term library and unambiguous standard expression library in the bidding field are constructed through the following steps: C1, by analyzing frequently used ambiguous trigger words in the bidding and tendering field, a high-frequency ambiguous trigger word database for the bidding and tendering field is obtained; including three types: Vague qualifiers: similar, related, high-quality, appropriate, reasonable; Missing definition words: qualified, up to standard, meet requirements, of the same scale; Vaguely defined terms: in principle, depending on the situation, when necessary, relevant parties; C2 collects unambiguous standard phrases corresponding to high-frequency ambiguous trigger words, such as "must have experience in municipal road projects with a single contract amount of ≥5 million in the past 3 years", and marks the "ambiguous clause type"; C3 uses the Sentence-BERT model to extract feature vectors from the unambiguous standard expression phrases obtained in step C2, and stores them using Postgres. The storage format is phrase text, ambiguous clause type, and phrase vector, resulting in an unambiguous standard expression library.
[0032] This embodiment presents an LLM dynamic context resolution method for ambiguous clauses in tender documents, such as... Figure 1 As shown, the specific steps are as follows: S1, inspect the contents of the tender documents, identify ambiguous clauses, and mark the types of ambiguous clauses; S2, extract the project features from the bidding documents and perform a dual-database search in a pre-built historical similar bidding document database and an industry standard database; S3 uses an LLM large model to perform ambiguity resolution reasoning and outputs the interpretation of ambiguous clauses.
[0033] Specifically, S1 includes: S11: The content of the tender document is broken down into paragraphs, and ambiguous trigger words are detected for each paragraph. The ambiguous trigger words used in the detection can be based on a high-frequency ambiguous trigger word database in the tendering field.
[0034] S12, when a paragraph contains ambiguous trigger words, the paragraph is split into phrases according to punctuation marks, and the clauses containing keywords are selected from the phrases; S13, calculate the semantic similarity between the clause text containing the keywords and the unambiguous standard expression library. The cosine similarity threshold is set to <0.6. If it is lower than the threshold, it is judged as an ambiguous clause, and the "ambiguous clause type" of the ambiguous clause is obtained.
[0035] The unambiguous standard expression library is a text feature library that stores unambiguous expressions. Each high-frequency trigger word corresponds to one unambiguous expression. For example, if the clause text containing the keyword "requires experience in similar projects," the corresponding unambiguous standard expression in the library would be "requires experience in municipal road projects with a single contract amount ≥ 5 million in the past 3 years." During storage, Sentence-BERT is first used to extract features from the unambiguous expressions, and these features are then normalized and stored in the Postgres vector library. During retrieval, a similarity search is performed between the clause text with extracted features and the text features from the unambiguous standard expression library. If the top-1 similarity is below 0.6, the clause is considered ambiguous.
[0036] After identifying ambiguous clauses and their types, a dual-database search is performed using both the tender document database and the industry standards database. S2 specifically includes: S21. Extract the project characteristics of the current tender document. Specifically, use LLM to parse the document cover and project overview section to obtain (e.g., "110kV substation construction"), budget scale (e.g., "≥20 million"), and industry (e.g., "power").
[0037] S22. Based on the "project type" and "budget size" obtained in S21, perform historical case matching in the tender document database to obtain the corresponding "ambiguous clauses - clarification" data.
[0038] For example, based on the "project type" and "budget size" obtained from S21, a precise match can be made in the tender document database (such as searching historical cases of "power industry + substation projects with a budget of over 20 million") to obtain the corresponding "ambiguous clauses - clarification" data.
[0039] S23. Based on the "industry" obtained in S21 and the "ambiguous clause type" obtained in S13, match them in the industry standard library to obtain the corresponding industry standard requirements.
[0040] For example: Based on the "industry" obtained in S21 and the "ambiguous clause type" obtained in S13, accurately match in the industry standard library (e.g., "power industry - project experience ambiguity") to obtain the corresponding industry standard requirements (e.g., "similar experience in power projects must include the construction of substations of the same voltage level").
[0041] S3 specifically combines the original text of ambiguous clauses, project characteristics, and supporting contextual information, inputting it via a Prompt to the LLM large model for scenario-based reasoning, and outputting a precise interpretation. The LLM large model is specifically the Qwen-32B large model or Qwen2.5-32B, Qwen3-302B-A3B, or Qwen3-32B large model.
[0042] The Prompt project design adopts a three-part Prompt structure of "instruction + context + constraint".
[0043] For example: "Based on the following information, please resolve the ambiguous clauses in the tender documents into unambiguous statements, and clarify the specific standards for the vague statements (such as scale, time, indicators)." # Original text of ambiguous clauses Bidders must have experience in similar projects. # Project Features The project is a 110kV substation construction project with a budget of 25 million yuan, belonging to the power industry.
[0044] # Historical Cases In 2023, a 110kV substation project clarified that "similar project experience refers to the construction experience of 110kV and above substations with a single contract amount of ≥20 million RMB in the past 3 years." # Industry Standards The "Guidelines for the Management of Bidding for Power Engineering Construction Projects" require that "experience in similar projects must match the project's voltage level and contract amount." Constraints: The interpretation must conform to power industry practices, match the current project budget and type, and be concise and clear.
[0045] # Output Format Related project characteristics: Refer to the context: Generate interpretation: ".
[0046] This invention presents an LLM dynamic context resolution method for ambiguous clauses in tender documents. By accurately identifying ambiguous clauses, it can associate them with external information such as historical tender cases and industry standards, achieving high-precision and high-efficiency parsing of ambiguous clauses in tender documents and thoroughly eliminating ambiguity. It provides a complete solution encompassing "accurate ambiguity identification - dynamic context expansion - scenario-based reasoning," meeting the high-precision and high-efficiency requirements of tender document parsing.
[0047] Example 2 I. This embodiment uses the resolution of ambiguous clauses in bidding documents for power industry substation construction projects based on Qwen-32B to illustrate the LLM dynamic context resolution method for ambiguous clauses in bidding documents of the present invention.
[0048] II. Implementation Scenario Parameters Application Scenario: Analysis of Tender Documents for 110kV Substation Construction Project (Industry: Power; Project Type: Infrastructure Construction; Budget: 25 million RMB) Input file: Bidding documents for the construction of 110kV substations in a certain province in 2024 (including 6 core chapters such as "Bidder Qualification Requirements" and "Project Experience Recognition", totaling 89 clauses). Hardware configuration: CPU Intel Xeon 8375C (32 cores), GPU NVIDIA A100 (40GB), RAM 128GB Software environment: Python 3.9, PyTorch 2.0, Sentence-BERT (version 2.2.2), and a finely tuned Qwen-32B model (training data: 50,000 pairs of "ambiguous clauses - standard interpretations"). III. Dual-database basic data Historical similar bidding document database: 320 bidding documents for 110kV and above substation construction projects in the power industry in the past 5 years were screened, and 1200 sets of ambiguous clauses and official clarification data in the categories of "project experience requirements" and "qualification certification" were extracted. The documents were then categorized and indexed by "project type (110kV / 220kV) - budget scale (10 million - 30 million / over 30 million) - industry".
[0049] Industry Standards Library: Includes 8 core standards such as the "Guidelines for Bidding Management of Power Engineering Construction Projects" and the "Measures for the Bidding Management of Housing Construction and Municipal Infrastructure Engineering Construction". It stores 286 clauses in the categories of "Project Experience" and "Qualification Requirements" in a structured manner according to "Industry" and "Type of Ambiguous Clauses".
[0050] IV. Specific Implementation Steps 1) Identification of Ambiguous Clauses Keyword matching: 15 clauses containing trigger words such as "similar", "qualified" and "in principle" were selected from the bidding documents. Among them, 8 clauses such as "the bidder must have similar project experience" and "a qualified test report must be provided" were entered into the semantic judgment stage.
[0051] Semantic similarity calculation: The cosine similarity between the 8 clauses and the unambiguous standard expression library is calculated using the Sentence-BERT model. Among them, 6 clauses with a similarity of <0.6 (such as "similar project experience" with a similarity of 0.42) are identified as ambiguous clauses and the retrieved tags are "project experience" and "qualification certification".
[0052] Sentence-BERT is suitable for handling sentence-level tasks, such as obtaining a sentence's vector representation and calculating text semantic similarity. Inputting two texts into the Sentence-BERT model yields their corresponding feature representations, and then calculating the cosine similarity yields a value between -1 and 1, which is used as the similarity score. The specific method for calculating the cosine similarity between a sentence and an unambiguous standard representation library using the Sentence-BERT model is common knowledge in the field and will not be elaborated upon here.
[0053] 2) Dynamic Context Extension Project Feature Extraction: By parsing the "Project Overview" section of the tender document using LLM, the core features were obtained: 110kV substation construction, budget of 25 million yuan, and power industry.
[0054] Dual-database search: Search the historical database for relevant cases of "110kV substation + 20-30 million budget" to obtain the top 10 highly relevant clarification data; search the industry standards database for "power industry + project experience ambiguity" to obtain 3 industry standard requirements.
[0055] Information filtering: Sort by "matching degree 60%+ and timeliness 40%", and select three core contextual supports, including a clarification case of a 110kV substation project in 2023 and relevant clauses of the "Guidelines for Bidding Management of Power Engineering Construction Projects".
[0056] 3) LLM Ambiguity Resolution Reasoning Prompt Construction: Generate a Prompt in the format of "instructions + context + constraints", with explicit requirements such as "clearly specifying the time, amount, and voltage level standards of the project experience" and constraints such as "conforming to power industry practices and matching the budget of 25 million".
[0057] Model reasoning: Input the Prompt into the Qwen-32B model, associate the project budget of 25 million with the model, set the amount standard to ≥20 million, combine the "last 3 years" time requirement of historical cases with the "voltage level matching" requirement of industry standards, and generate unambiguous interpretations.
[0058] Output results: Related project characteristics: Budget 25 million → Determine amount standard ≥ 20 million; Project type 110kV substation → Determine voltage level standard 110kV and above. Referring to the context: combining the "last 3 years" time standard of historical cases with the "voltage level + amount" dual-dimensional requirements of industry standards; Generate interpretation: The final output is "Bidders must have experience in the construction of 110kV and above substations with a single contract amount of ≥20 million RMB in the past 3 years", and the basis for the interpretation is indicated (e.g., "Based on a historical case of a 110kV substation project in 2023 and the 'Guidelines for Bidding Management of Power Engineering Construction Projects'").
[0059] V. Verification of Implementation Results Accuracy of identification: All 6 ambiguous clauses were accurately identified, with no omissions or misjudgments.
[0060] Resolution accuracy: The unambiguous output was reviewed by three bidding experts in the power industry and all met industry practice and the actual needs of the project.
[0061] Processing efficiency: The entire processing time for a single tender document (89 clauses) is 120 seconds, which is 20 times more efficient than manual processing (average 4 hours).
[0062] This invention presents an LLM dynamic context resolution method for ambiguous clauses in tender documents. By accurately identifying ambiguous clauses, it can associate them with external information such as historical tender cases and industry standards, achieving high-precision and high-efficiency parsing of ambiguous clauses in tender documents and thoroughly eliminating ambiguity. It provides a complete solution encompassing "accurate ambiguity identification - dynamic context expansion - scenario-based reasoning," meeting the high-precision and high-efficiency requirements of tender document parsing.
[0063] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the essence and scope of the technical solutions of the present invention.
Claims
1. A dynamic context resolution method for LLM (Limited Least Context Management) for ambiguous clauses in tender documents, characterized in that, This can be done through the following steps: S1, inspect the contents of the tender documents, identify ambiguous clauses, and mark the types of ambiguous clauses; S2, extract the project features from the bidding documents and perform a dual-database search in a pre-built historical similar bidding document database and an industry standard database; S3 uses an LLM large model to perform ambiguity resolution reasoning and outputs the interpretation of ambiguous clauses.
2. The LLM dynamic context resolution method for ambiguous clauses in tender documents according to claim 1, characterized in that, S1 specifically includes: S11, the content of the tender document is split into paragraphs, and ambiguous trigger words are detected in each paragraph; S12, when a paragraph contains ambiguous trigger words, the paragraph is split into phrases according to punctuation marks, and the clauses containing keywords are selected from the phrases; S13, perform semantic similarity calculation between the text of the clause containing the keyword and the unambiguous standard expression library. Set the cosine similarity threshold to <0.
6. If it is lower than the threshold, it is judged as an ambiguous clause, and the "ambiguous clause type" of the ambiguous clause is obtained.
3. The LLM dynamic context resolution method for ambiguous clauses in tender documents according to claim 2, characterized in that, S2 specifically includes: S21. Extract the project characteristics of the current tender document. Specifically, use LLM to parse the document cover and project overview section to obtain "project type", "budget size" and "industry". S22, based on the "project type" and "budget size" obtained in S21, perform historical case matching in the bidding document database to obtain the corresponding "ambiguous clauses - clarification" data; S23. Based on the "industry" obtained in S21 and the "ambiguous clause type" obtained in S13, match them in the industry standard library to obtain the corresponding industry standard requirements.
4. The LLM dynamic context resolution method for ambiguous clauses in tender documents according to claim 3, characterized in that, S3 specifically combines the original text of ambiguous clauses, project characteristics, and supporting contextual information, inputs it into the LLM large model via Prompt for scenario-based reasoning, and outputs accurate interpretations.
5. The LLM dynamic context resolution method for ambiguous clauses in tender documents according to claim 4, characterized in that, The Prompt project design adopts a three-part Prompt structure of "instructions + context + constraints".
6. The LLM dynamic context resolution method for ambiguous clauses in tender documents according to claim 5, characterized in that, The Prompt statement is specifically: "Based on the following information, please resolve the ambiguous clauses in the tender documents into unambiguous statements, and clarify the specific standards for ambiguous statements;" # Original text of ambiguous clauses # Project Features Project type is Budget size The industry it belongs to is ; # Historical Cases # Industry Standards Constraints: The interpretation must conform to industry practices, match the current project budget and type, and be concise and clear; # Output Format Related project characteristics: Refer to the context: Generate interpretation: ”。 7. The LLM dynamic context resolution method for ambiguous clauses in tender documents according to any one of claims 1 to 6, characterized in that, The LLM large model specifically refers to the Qwen-32B large model or the Qwen2.5-32B, Qwen3-302B-A3B, and Qwen3-32B large models.
8. The LLM dynamic context resolution method for ambiguous clauses in tender documents according to claim 7, characterized in that, The pre-built historical similar tender document library is constructed through the following steps: A1. Collect ambiguous clauses in publicly tendered documents from relevant industries that are needed in the past 3-10 years; A2, each tender document is split into multiple data entries based on ambiguous clauses, and then verified; The data format is: Ambiguous clause (original text): Official clarification and interpretation: Basic project information: A3, label the "Project Type", "Budget Scale", and "Industry" in the tender documents to obtain tag data; A4. Store the ambiguous clause data from step 2) and the label data from step 3) together in the tender document database for later use.
9. The LLM dynamic context resolution method for ambiguous clauses in tender documents according to claim 8, characterized in that, The pre-built industry standard library is constructed through the following steps: B1 contains the necessary industry bidding standards and local bidding rules; B2, mark the "industry" and "type of ambiguous clauses" in the bidding standards; B3 stores the bidding standard content, "industry", and "ambiguous clause types" data together in the industry standard database for future use.
10. The LLM dynamic context resolution method for ambiguous clauses in tender documents according to claim 7, characterized in that, The pre-built high-frequency ambiguity trigger term library and unambiguous standard expression library in the bidding field are constructed through the following steps: C1, by analyzing frequently used ambiguous trigger words in the bidding and tendering field, a high-frequency ambiguous trigger word database for the bidding and tendering field is obtained; including three types: Vague qualifiers: similar, related, high-quality, appropriate, reasonable; Missing definition words: qualified, up to standard, meet requirements, of the same scale; Vaguely defined terms: in principle, depending on the situation, when necessary, relevant parties; C2 collects unambiguous standard expression phrases corresponding to high-frequency ambiguous trigger words and marks the "ambiguous clause type"; C3 uses the Sentence-BERT model to extract feature vectors from the unambiguous standard expression phrases obtained in step C2, and stores them using Postgres. The storage format is phrase text, ambiguous clause type, and phrase vector, resulting in an unambiguous standard expression library.