Method for accurately positioning clauses in contract question and answer
Through the big model technology of similarity calculation, combined with keyword semantic retrieval and vector similarity model, the problems of low information retrieval efficiency and inaccurate answers in the contract are solved, and the precise positioning and efficient information retrieval of contract questions and answers are realized, which improves information The accuracy of the search and the completeness of the answer.
Patent Information
- Application Number
- CN202411039747.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-31
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2044-07-31
AI Technical Summary
The prior art is difficult to accurately understand complex semantic and contextual information in contracts, resulting in low information retrieval efficiency, inaccurate answers, and difficult to quickly locate relevant information.
Using the big model technology of similarity calculation, by dividing the contract content into chapters, using keyword semantic search and vector similarity model, multiple rounds of dialogue citation chapters are carried out, and the similarity between blocks and answers, terms, and questions are calculated to accurately locate them. and answer generation.
It improves the accuracy and efficiency of information retrieval, enhances the completeness of question answers, ensures that important information is not omitted, reduces the risk of information omission, and provides more accurate and in-depth answers.
Smart Images

Figure CN118916463A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of information retrieval, and in particular to a method for accurately locating clauses in a contract question and answer session. Background Art
[0002] With the rapid development of information technology, the digitization of contracts and the massive growth of information have put forward higher requirements for the efficient management and accurate retrieval of the terms and contents in contracts. The main challenges faced by existing technologies include: Limitations of semantic understanding: Traditional systems often find it difficult to accurately understand the complex semantics and contextual information in documents, resulting in insufficient accuracy of the final answer and insufficient relevance of the retrieval results. For example, a contract usually contains general terms, special terms and annexes. Without combining all relevant chapters, it is difficult to draw accurate conclusions based on a single chapter or paragraph.
[0003] The efficiency of information retrieval: Faced with a huge set of terms, how to quickly locate relevant information is another challenge faced by existing technologies.
[0004] Accuracy of question answering: Existing systems may not provide comprehensive or accurate answers when answering complex questions.
[0005] Therefore, in response to the above problems, a method of accurately locating clauses in contract Q&A is needed. Summary of the invention
[0006] The purpose of the present invention is to provide a method for accurately locating clauses in contract questions and answers. The present invention uses a large model technology for similarity calculation to focus on information location and question answering in document sections. Through efficient information retrieval and processing capabilities, it provides users with more accurate and in-depth answer services.
[0007] The present invention is achieved in that:
[0008] The present invention provides a method for accurately locating clauses in a contract question and answer, which is specifically performed in the following steps:
[0009] S1: Divide the contents of the contract into multiple chapters according to the titles. For the first round of questions raised, directly combine them with keyword semantic search, conduct at least one round of screening, and answer the questions raised;
[0010] S2: searching the content of the answer to step S1 to see whether it is necessary to borrow other chapters that are not in the given chapters. If necessary, find the referenced chapters and add the referenced chapters in a multi-round dialogue;
[0011] S3: Divide the chapters obtained in step S1-step S2 into smaller granularity blocks, and calculate the similarity of these blocks with the final answer of the large model, similar terms and the questions input by the user through the vector similarity search model and summarize them, retain the blocks with higher final similarity, and number them one by one, and the granularity blocks are paragraphs; as shown in formula (1);
[0012] S final =α·S answer +β·S templates +γ·S question Formula (1)
[0013] in,
[0014] S final - The final similarity;
[0015] S answer - The similarity between the final answer generated by the small-size block and the large model;
[0016] S templates - Similarity between small granular blocks and similar terms input by the user;
[0017] S question - Similarity between small-scale chunks and the questions input by the user;
[0018] α, β, γ - are the weights of these similarities in the final similarity and are constants;
[0019] S4: The similarity calculation model is used to calculate the similarity between the numbered but unselected blocks in step S3 and the final answer of the large model. When it is found that there are blocks with higher similarity among the unselected blocks, the remaining blocks are re-encoded and step S3 is repeated. The similarity calculation model is specifically executed according to the following steps:
[0020] S 4.1 : First, pre-process the block data of the contract content, and then clean the data to remove noise data and outliers;
[0021] S 4.2 : Perform feature selection on the block data of the contract content, extract features that have an important impact on similarity calculation from the original data; and perform feature standardization: convert features of different dimensions to the same scale for comparison;
[0022] S 4.3 : Represent the selected features in vector form, with each feature corresponding to a dimension in the vector;
[0023] S 4.4: Select a similarity calculation method according to the specific application scenario and data characteristics, calculate the similarity, and apply the selected similarity calculation formula to calculate the similarity between two or more objects, such as formula (2);
[0024]
[0025] Among them, i·j represents the dot product of vectors i and j, and ││i││ and ││j││ represent the moduli of vectors i and j, respectively.
[0026] S5: The first round of questions raised in step S1 are accurately positioned again. Specifically, multiple rounds of dialogues are constructed by similarity calculation and renumbering of block contents. The final positioning is obtained by accurate similarity calculation based on the content of the dialogue, and the answer to the final positioning is output as information.
[0027] Furthermore, the present invention provides a computer-readable storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by a main controller, the method described in any one of the above is implemented.
[0028] Compared with the prior art, the present invention has the following beneficial effects:
[0029] 1. Improve the accuracy and efficiency of information retrieval.
[0030] 2. By combining information from multiple chapters, the inaccuracy caused by a single information source is avoided and the accuracy of information retrieval is improved.
[0031] 3. Enhance the completeness of problem answers.
[0032] 4. The system can effectively quote and combine information from multiple chapters to provide more complete and accurate answers and reduce the risk of missing information.
[0033] 5. Multiple modules work together to ensure that important information is not missed, thus improving the reliability of the system. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. It is understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.
[0035] Figure 1 is a flow chart of the method of the present invention;
[0036] Figure 2 It is a multi-round dialogue reference chapter diagram of the present invention;
[0037] Figure 3 It is the first round of dialogue call graph of the present invention;
[0038] Figure 4 It is a flow chart of finding a positioning block of the present invention. DETAILED DESCRIPTION
[0039] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the scope of protection of the present invention. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the invention claimed for protection, but is only for selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the scope of protection of the present invention.
[0040] See also Figure 1-4 The present invention provides a method for accurately locating clauses in a contract question and answer, which is specifically performed in the following steps:
[0041] Please follow the steps below:
[0042] S1: Divide the contents of the contract into multiple chapters according to the titles. For the first round of questions raised, directly combine them with keyword semantic search, conduct at least one round of screening, and answer the questions raised;
[0043] S2: searching the content of the answer to step S1 to see whether it is necessary to borrow other chapters that are not in the given chapters. If necessary, find the referenced chapters and add the referenced chapters in a multi-round dialogue;
[0044] S3: Divide the chapters obtained in step S1-step S2 into smaller granularity blocks, and calculate the similarity of these blocks with the final answer of the large model, similar terms and the questions input by the user through the vector similarity search model and summarize them, retain the blocks with higher final similarity, and number them one by one, and the granularity blocks are paragraphs; as shown in formula (1);
[0045] S final =α·S answer +β·S templates +γ·S question Formula (1)
[0046] in,
[0047] Sfinal - The final similarity;
[0048] S answer - The similarity between the final answer generated by the small-size block and the large model;
[0049] S templates - Similarity between small granular blocks and similar terms input by the user;
[0050] S question - Similarity between small-scale chunks and the questions input by the user;
[0051] α, β, γ - are the weights of these similarities in the final similarity and are constants;
[0052] S4: The similarity calculation model is used to calculate the similarity between the numbered but unselected blocks in step S3 and the final answer of the large model. When it is found that there are blocks with higher similarity among the unselected blocks, the remaining blocks are re-encoded and step S3 is repeated. The similarity calculation model is specifically executed according to the following steps:
[0053] S 4.1 : First, pre-process the block data of the contract content, and then clean the data to remove noise data and outliers;
[0054] S 4.2 : Perform feature selection on the block data of the contract content, extract features that have an important impact on similarity calculation from the original data; and perform feature standardization: convert features of different dimensions to the same scale for comparison;
[0055] S 4.3 : Represent the selected features in vector form, with each feature corresponding to a dimension in the vector;
[0056] S 4.4 : Select a similarity calculation method according to the specific application scenario and data characteristics, calculate the similarity, and apply the selected similarity calculation formula to calculate the similarity between two or more objects, such as formula (2);
[0057]
[0058] Among them, i·j represents the dot product of vectors i and j, and ││i││ and ││j││ represent the moduli of vectors i and j, respectively.
[0059] S5: The first round of questions raised in step S1 are accurately positioned again. Specifically, multiple rounds of dialogues are constructed by similarity calculation and renumbering of block contents. The final positioning is obtained by accurate similarity calculation based on the content of the dialogue, and the answer to the final positioning is output as information.
[0060] In this embodiment, the present invention provides a computer-readable storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by a main controller, the method described in any one of the above is implemented.
[0061] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention has various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. A method for accurately locating clauses in a contract question and answer, characterized by: Please follow the steps below: S1: Divide the contents of the contract into multiple chapters according to the titles. For the first round of questions raised, directly combine them with keyword semantic search, conduct at least one round of screening, and answer the questions raised; S2: searching the content of the answer to step S1 to see whether it is necessary to borrow other chapters that are not in the given chapters. If necessary, find the referenced chapters and add the referenced chapters in a multi-round dialogue; S3: Divide the chapters obtained in step S1-step S2 into smaller granularity blocks, and use the vector similarity search model to calculate the similarity of these blocks with the final answer of the large model, similar terms, and the questions input by the user, and summarize them, retain the blocks with higher final similarity, and number them one by one, as shown in formula (1); S final =α·S answer +β·S templates +γ·S question formula(1) in, S final - The final similarity; S answer - The similarity between the final answer generated by the small-size block and the large model; S templates - Similarity between small granular blocks and similar terms input by the user; S question - Similarity between small-scale chunks and the questions input by the user; α, β, γ - are the weights of these similarities in the final similarity and are constants; S4: using the similarity calculation model to calculate the similarity between the numbered but unselected blocks in step S3 and the final answer of the large model, and when it is found that there are blocks with high similarity among the unselected blocks, re-encode the remaining blocks and repeat step S3; S5: Re-precisely locate the first round of questions raised in step S1.
2. A method for accurately locating clauses in contract question and answer according to claim 1, characterized in that: In step S4, the similarity calculation model is specifically executed according to the following steps: S 4.1 : First, pre-process the block data of the contract content, and then clean the data to remove noise data and outliers; S 4.2 : Perform feature selection on the block data of the contract content and extract features that have an important impact on similarity calculation from the original data; And perform computational feature standardization: convert features of different dimensions to the same scale for comparison; S 4.3 : Represent the selected features in vector form, with each feature corresponding to a dimension in the vector; S 4.4 : Select a similarity calculation method according to the specific application scenario and data characteristics, calculate the similarity, and apply the selected similarity calculation formula to calculate the similarity between two or more objects, such as formula (2); Among them, i·j represents the dot product of vectors i and j, and ││i││ and ││j││ represent the moduli of vectors i and j, respectively.
3. According to claim 1, a method for accurately locating clauses in a contract question and answer, characterized in that: In step S3, the granularity block is a paragraph.
4. A method for accurately locating clauses in a contract question and answer according to claim 1, characterized in that: In step S5, by calculating similarity and renumbering the contents of the blocks, multiple rounds of dialogue are constructed, and the final positioning is obtained by accurately calculating the similarity based on the contents of the dialogue, and the answer to the final positioning is output as information.
5. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by the main controller, the method according to any one of claims 1 to 4 is implemented.
Citation Information
Patent Citations
Question and answer knowledge base using method and system based on large language model
CN118296104A
Intelligent self-service contract review method based on large language model
CN118379165A
Bone conduction earphone with noise reduction function
KR1020230172804A
