Environment regulation case auxiliary analysis system based on large language model

Abstract

The invention relates to an environment law and regulation case auxiliary analysis system based on a large language model, and the system employs the large language model to carry out the case induction of a historical environment law violation judgment book, forms a case database, and supports the matching retrieval. The method comprises the following steps of: studying and judging an environment illegal scene by utilizing a large language model in combination with legal provisions to form a case summary description so as to retrieve historical cases; and on the basis of a big language model, integrating current case summary and historical judgment conditions, forming prediction analysis on illegal facts, judgment basis and punishment contents of the scene of the user, and assisting a production enterprise or an environment law enforcement officer to judge illegal consequences and punishment strength of a specific scene. Therefore, the method does not need professional personnel to carry out case teasing or data collection, the input requirement is low, the application difficulty is small, the output analysis result basis is more transparent, and the staff can conveniently check.

Description

Technical Field

[0001] The present invention relates to the field of environmental protection technology, and to application fields such as ecological environment law and regulation case query, supervision and law enforcement assistance, and in particular to an environmental law and regulation case auxiliary analysis system based on a large language model. Background Art

[0002] There are numerous laws, regulations and policy documents related to the ecological environment, which are constantly updated. It is difficult for relevant practitioners, especially environmental supervision and law enforcement personnel, corporate environmental protection related persons in charge and staff, to grasp the relevant regulations in real time and combine them with the actual situation. Especially for environmental law enforcement personnel, they often need to combine past cases to determine the judgment results of specific environmental violations, and the related retrieval and sorting work is time-consuming and labor-intensive.

[0003] The existing technology mainly assists in the judgment by obtaining the text related to the judgment case, determining the reference legal provisions associated with the text content of the case text, and finally determining the auxiliary judgment result of the case text based on the case text and the reference legal provisions. There are also data information such as obtaining labeled business data, such as basic information of market entities, historical inspection plan information, historical inspection task information, historical inspection information, and legal and regulatory information.

[0004] Based on the existing technology, the following problems are currently found: 1. High input requirements and high application difficulty: The current environmental law case auxiliary analysis technology mainly analyzes by inputting professional case texts or subject data, which requires relatively professional personnel to sort out the case or collect data, and the application scope is relatively small; 2. The basis of analysis results is not transparent: Some technical solutions use deep learning models such as transformer and seq2seq to build regulatory analysis models. Due to the inherent black box problem of deep learning, it is difficult for users to perceive the generation process of analysis results, which affects its application in serious and rigorous scenarios such as judicial practice. Summary of the invention

[0005] The purpose of the present invention is to provide an environmental regulation case auxiliary analysis system based on a large language model to solve the problems encountered in the above-mentioned background technology.

[0006] To achieve the above object, the technical solution of the present invention is as follows: An environmental law case auxiliary analysis system based on a large language model includes a reference knowledge base, a case information base, a case summary large model, and a judgment auxiliary analysis large model; the reference knowledge base uses the text of ecological environmental laws and regulations as an auxiliary basis, segments the text, sends the data to the large language model, and dynamically updates it; the case summary large model analyzes the scene description input by the user based on the large language model, combines the legal text matching results, and generates a professional case summary description result; the case information base includes environmental administrative penalty judgments, provides auxiliary references for subsequent environmental law enforcement based on the content of the administrative penalty judgments, and generates case information matching results based on user searches; the judgment auxiliary analysis large model inputs the case summary description results and the case information matching results into the large language model together, forming a predictive analysis of the illegal facts, judgment basis and penalty content of the scene described by the user.

[0007] In the reference knowledge base, when text segmentation is performed, the imported legal and regulatory text is segmented to match with more accurate text data; including for each text M 1 , by segmenting to form text fragments {M 1 1,M 1 2,...,M 1 n}.

[0008] Text segmentation is mainly performed in the following ways: ① Iterative segmentation: According to the set symbol priority, the text is segmented based on the symbol with the highest priority first, and then the segmented text is judged to see whether it reaches the length threshold. If it does not reach the threshold, it is further segmented with the symbol with the next priority, and so on, until the length of all text segments is below the threshold. ② Feature recognition and segmentation: Identify the "Article X" feature text format in the legal text, where X is a Chinese number of uncertain length; at the same time, determine whether the format text is at the very beginning of an article text based on the context, rather than a reference to other articles within the article. Based on the recognition results, segment the text according to each law and regulation; ③ Deep learning segmentation: segment text based on natural language deep learning algorithm.

[0009] In the above scheme, the segmented text fragments Perform quantitative analysis to form the corresponding text information vector ;in represents the i-th segment of text x, Represents the vector corresponding to the i-th segment of text x. Unstructured databases are used to store legal documents. The knowledge base mainly includes legal document information, text segment information associated with the document, and information vector tables associated with the text segments.

[0010] In the case summary model, when matching legal provisions, the same quantitative analysis method as text segmentation is used to vectorize the case description content q input by the user to form , and match them in the text segmentation knowledge base to find the legal provisions related to the case; Specifically, calculate and The vector distance : ; in, and are row vectors of the same dimension, represents the second-order norm, take The smallest n are used as reference inputs for the larger model.

[0011] According to the case summary big model, the legal provisions related to the case are searched, and the user inputs the case description text, and the big language model is used to integrate the reasoning ability to output a professional case summary description result Q.

[0012] In the case information database, based on the information integration capability of the large language model, the content of the environmental administrative penalty judgment is analyzed to extract the core key information; the illegal facts and the judgment basis are merged to form a case a i ,According to the quantitative analysis method, the case vector Vai is established.

[0013] As a preferred solution, the extracted key information is used to construct a case information database, which mainly includes the following steps: The core key information, including the parties, illegal facts, judgment basis and punishment content, is combined with the case vector as a record A i , construct case information database table A={A 1， A 2， ,...,A M}.

[0014] The case summary description result Q is used to establish the case summary description result vector V according to the quantitative analysis method. Q , the case summary description result vector V Q With the case vector Match and find the most relevant historical penalty cases; the specific calculation method is as follows: ; Pick The minimum n2 cases, matching cases form set A match , which is the case information matching result.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention uses a large language model to summarize the historical environmental violation judgments, forms a case database and supports matching retrieval; uses a large language model in combination with legal provisions to study and judge environmental violation scenarios, forms a case summary description to facilitate retrieval of historical cases; based on the large language model, the current case summary and historical precedents are combined to form a predictive analysis of the illegal facts, judgment basis and punishment content of the scenario described by the user, and assists production enterprises or environmental law enforcement personnel in judging the illegal consequences and punishment intensity of specific scenarios. Therefore, the present invention does not require professional personnel to sort out the case or collect data, has low input requirements, low application difficulty, and the output analysis results are more transparent, which is convenient for staff to check. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The disclosure of the present invention is described with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of the present invention. In the accompanying drawings, the same reference numerals are used to refer to the same components. Among them: Figure 1 It is a schematic diagram of the working principle of the present invention; Figure 2 This is a schematic diagram of data information extracted from case information in the present invention. DETAILED DESCRIPTION

[0017] In order to make the technical means, creative features, objectives and effects of the present invention easy to understand, the present invention is now further described in detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the relevant components of the present invention.

[0018] According to the technical solution of the present invention, without changing the essential spirit of the present invention, a person skilled in the art can propose a variety of interchangeable structural modes and implementation modes. Therefore, the following specific implementation modes and the accompanying drawings are only exemplary descriptions of the technical solution of the present invention, and should not be regarded as the entirety of the present invention or as a limitation or restriction to the technical solution of the present invention.

[0019] The technical solution of the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments.

[0020] like Figure 1 As shown, an environmental regulation case auxiliary analysis system based on a large language model includes a reference knowledge base, a case information base, a case summary large model, and a judgment auxiliary analysis large model.

[0021] 1. Establish a reference knowledge base: Build a reference knowledge base with ecological and environmental laws and regulations as an auxiliary basis, segment the legal and regulatory texts in the reference knowledge base, send the data to the large language model, and update it dynamically.

[0022] 1.1 Text segmentation: Segment the imported legal regulations text to perform more accurate matching. For each text M 1 , form text segments {M 1 1, M 1 2,..., M 1 n} through the following main methods: Iterative segmentation: According to the set symbol priorities, first segment the text based on the symbol with the highest priority, and judge whether the length of the segmented text reaches the length threshold. If the threshold is not reached, further segment it with the symbol of the next lower priority, and so on, until the length of all text segments is lower than the threshold. In the embodiment, the threshold is 50.

[0023] Feature recognition segmentation: Identify the "Article X" feature text format in legal provisions, where X is Chinese numerals of uncertain length, such as "one", "eleven", etc. At the same time, judge whether the text in this format is at the very beginning of a provision text according to the context, rather than a reference to other provisions within the provision. According to the recognition result, segment the text article by article; Deep learning segmentation, segment the text based on natural language deep learning algorithms. In the embodiment, a Chinese text segmentation model based on Bert is adopted.

[0024] 1.2 Quantitative analysis: Perform quantitative analysis on the segmented text segments to form corresponding text information vectors . Among them represents the i-th segment of text x, represents the vector corresponding to the i-th segment of text x. Quantitative analysis can adopt techniques such as word frequency statistics and deep learning-based word vector embedding; 1.3 Knowledge base establishment: Store legal regulations documents using an unstructured database. The knowledge base mainly includes regulatory document information, text segment information associated with the document, and an information vector table associated with the text segments.

[0025] 2. Case summary large model: Analyze the scenario description input by the user based on a large language model, and combine the legal provision matching results to generate a professional case summary description result.

[0026] 2.1 Legal provision matching: Adopt the same quantitative analysis method as in 1.2 to vectorize the case description content q input by the user to form , and perform matching in the knowledge base described in 1.2 to find legal provisions related to the case; specifically, calculate and vector distance : ; in, and are row vectors of the same dimension, represents the 2nd-order norm. The smallest n are used as reference inputs for the large model. In the embodiment, n is 3.

[0027] 2.2 Case summary analysis: Based on the search results for legal provisions related to the case as described in 2.1, and the case description text entered by the user, the large language model is used to integrate reasoning capabilities and output a professional case summary description result Q, including the case content, the content of the relevant laws and regulations violated, and other information; ; 3. Historical Penalty Information Case Library: The Historical Penalty Case Library includes environmental administrative penalty judgments, provides auxiliary references for subsequent environmental law enforcement based on the content of the administrative penalty judgments, and generates case information matching results based on user searches.

[0028] 3.1 Extraction of key case information: Based on the information integration capability of the large language model, analyze the content of the environmental administrative penalty judgment and extract the core key information, including the parties, illegal facts, judgment basis and penalty content. Figure 2 .

[0029] 3.2 Constructing a case information database: Constructing a case information database with the key information extracted in 3.1, mainly including the following steps: 1) Combine the illegal facts and the basis for judgment to form case a i , establish the case vector according to the quantitative analysis method described in 1.2 ; 2) Combine the key information described in 3.1, including the parties involved, illegal facts, judgment basis, and punishment content, with the case vector as a record A i , construct case information database table A={A 1， A 2， ,...,A M}; 3.3 Case information matching: The case summary description result Q described in 2.2 is used to establish the case summary description result vector V according to the quantitative analysis method described in 1.2 Q , the case summary description result vector V Q Same as the case vector described in 3.2 Match and find the most relevant historical penalty cases. Specifically: ; Pick The minimum number of n2 cases, in this embodiment, n2=3, the matching cases form a set A match , which is the case information matching result.

[0030] Judgment-assisted analysis model: Match the case summary description result Q described in 2.2 with the case information matching result A described in 3.3 match The information is input into the big language model together, and through the prompt project, the big language model is driven to form a predictive analysis of the illegal facts, judgment basis and punishment content of the scenario described by the user based on the above information and referring to past cases, so as to assist production enterprises or environmental law enforcement personnel to judge the illegal consequences and punishment intensity of specific scenarios.

[0031] ; ; The present invention has the following advantages over the prior art: 1. By establishing a knowledge base of ecological environmental laws and regulations, dynamic updating of laws and regulations is achieved, and query efficiency and accuracy are improved.

[0032] 2. Use a large language model to analyze the scenario description input by the user, and combine it with the legal text matching results to generate a professional case summary description, improving the accuracy and efficiency of case analysis and output.

[0033] 3. Build a case database of historical penalty information to provide auxiliary reference for environmental law enforcement and improve the pertinence and effectiveness of law enforcement.

[0034] 4. The case summary description results and the case information matching results are input into the large language model together, forming a predictive analysis of the illegal facts, judgment basis and punishment content of the scenario described by the user, assisting production enterprises or environmental law enforcement personnel to judge the illegal consequences and punishment intensity of specific scenarios, and improving the scientificity and fairness of decision-making.

[0035] The present invention uses a large language model to summarize the case of historical environmental violation judgments, form a case database and support matching retrieval; use a large language model in combination with legal provisions to study and judge environmental violation scenarios, form a case summary description to retrieve historical cases; based on the large language model, the current case summary and historical precedents are integrated to form a predictive analysis of the illegal facts, judgment basis and punishment content of the scene described by the user, and assist production enterprises or environmental law enforcement personnel to judge the illegal consequences and punishment intensity of specific scenes. Therefore, the present invention does not require professional personnel to sort out the case or collect data, and its input requirements are low, the application difficulty is small, and the output analysis results are more transparent, which is convenient for staff to check.

[0036] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. An environmental law case auxiliary analysis system based on a large language model, characterized in that: Including reference knowledge base, case information base, case summary model, and judgment auxiliary analysis model; The reference knowledge base uses the text of ecological environment laws and regulations as an auxiliary basis, segments the text, sends the data to the large language model, and dynamically updates it; The case summary model analyzes the scenario description input by the user based on the large language model, and generates a professional case summary description result in combination with the legal text matching result; The case information database includes environmental administrative penalty judgments, provides auxiliary references for subsequent environmental law enforcement based on the content of the administrative penalty judgments, and generates case information matching results based on user searches; The judgment-assisted analysis model inputs the case summary description results and the case information matching results into the large language model to form a predictive analysis of the illegal facts, judgment basis and punishment content of the scenario described by the user.

2. According to claim 1, the environmental law case auxiliary analysis system based on a large language model is characterized by: In the reference knowledge base, when text segmentation is performed, the imported legal and regulatory text is segmented to match with more accurate text data; including for each text M 1 , by segmenting to form text fragments {M 1 1,M 1 2,...,M 1 n}.

3. The environmental law case auxiliary analysis system based on a large language model according to claim 2 is characterized by: Text segmentation is mainly performed in the following ways: ① Iterative segmentation: According to the set symbol priority, the text is segmented based on the symbol with the highest priority first, and then the segmented text is judged to see whether it reaches the length threshold. If it does not reach the threshold, it is further segmented with the symbol with the next priority, and so on, until the length of all text segments is below the threshold. ② Feature recognition and segmentation: Identify the "Article X" feature text format in the legal text, where X is a Chinese number of uncertain length; at the same time, determine whether the format text is at the very beginning of an article text based on the context, rather than a reference to other articles within the article. Based on the recognition results, segment the text according to each law and regulation; ③ Deep learning segmentation: segment text based on natural language deep learning algorithm.

4. The environmental law case auxiliary analysis system based on a large language model according to claim 2 is characterized by: After segmentation, the text segment Perform quantitative analysis to form the corresponding text information vector ;in represents the i-th segment of text x, Represents the vector corresponding to the i-th segment of text x.

5. The environmental law case auxiliary analysis system based on a large language model according to claim 4 is characterized by: Unstructured databases are used to store legal and regulatory documents. The knowledge base mainly includes regulatory document information, text fragment information associated with the documents, and information vector tables associated with the text fragments.

6. The environmental law case auxiliary analysis system based on a large language model according to claim 2 is characterized by: In the case summary model, when matching legal provisions, the same quantitative analysis method as text segmentation is used to vectorize the case description content q input by the user to form , and match them in the text segmentation knowledge base to find the legal provisions related to the case; Specifically, calculate and The vector distance : ； in, and are row vectors of the same dimension, represents the second-order norm, take The smallest n are used as reference inputs for the larger model.

7. The environmental law case auxiliary analysis system based on a large language model according to claim 6 is characterized by: According to the case summary big model, the legal provisions related to the case are searched, and the user inputs the case description text, and the big language model is used to integrate the reasoning ability to output a professional case summary description result Q.

8. The environmental law case auxiliary analysis system based on a large language model according to claim 7 is characterized by: In the case information database, based on the information integration capability of the large language model, the content of the environmental administrative penalty judgment is analyzed to extract the core key information; Combine the illegal facts and the basis for judgment to form case a i , establish case vectors according to quantitative analysis methods .

9. The environmental law case auxiliary analysis system based on a large language model according to claim 8 is characterized by: The extracted key information is used to construct a case information database, which mainly includes the following steps: The core key information, including the parties, illegal facts, judgment basis and punishment content, is combined with the case vector as a record A i , construct case information database table A={A 1， A 2， ,...,A M }.

10. The environmental law case auxiliary analysis system based on a large language model according to claim 9 is characterized by: The case summary description result Q is used to establish the case summary description result vector V according to the quantitative analysis method. Q , the case summary description result vector V Q With the case vector Match and find the most relevant historical penalty cases; the specific calculation method is as follows: ； Pick The minimum n2 cases, matching cases form set A match , which is the case information matching result.

Images