Arbitration trial thinking chain inference method and device based on large language model

Abstract

The invention discloses an arbitration trial thinking chain inference method and device based on a large language model, and relates to the technical field of labor arbitration intelligent trial. An information extraction module performs text analysis on an application file based on a natural language processing technology, and extracts a keyword data set; the information analysis module identifies and extracts the basic condition of the applicant in the on-job period based on the keyword data set of text analysis; the information analysis module sorts dispute items of the case; the information analysis module retrieves laws and regulations and judicial interpretation related to the case; a mapping relation between the arbitration requests and dispute items is established, reasonability and feasibility of the arbitration requests are analyzed, the arbitration requests are matched with the dispute items, an arbitration trial thinking chain is automatically generated, and the arbitration trial thinking chain comprises basic conditions of an applicant, the dispute items, the arbitration requests and corresponding facts, evidences and legal bases. The thinking mode of a gutta-percha tailor is simulated, and automation of the arbitration trial process is achieved. The arbitration efficiency is improved, and the fairness of the trial result is ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent trial of labor arbitration, and in particular to a method and device for inferring a thought chain of arbitration trial based on a large language model. Background Art

[0002] With the rapid growth in the number of labor dispute cases and the increasing complexity of case types, it has become an inevitable trend to use data and intelligent systems to establish a new model for the diversified resolution of labor disputes. This will help improve processing efficiency, reduce resource waste, and achieve more precise governance.

[0003] The construction of intelligent arbitration aims to reflect the advantages of convenience, efficiency and low cost. With the support of network technology and big data, the intelligent arbitration system can better serve arbitrators in handling cases and relevant departments in making decisions, and promote the construction of a diversified dispute resolution mechanism with labor and personnel dispute arbitration as the core.

[0004] Although intelligent arbitration is feasible, there are obstacles in its technical implementation. For example, labor and personnel dispute cases usually involve highly unstructured evidence materials, which poses a challenge to the degree of intelligent implementation.

[0005] Problems with coordination between systems: Currently, there is a lack of a coordinated case handling system between arbitration and litigation, resulting in a digital breakpoint between arbitration and judgment. In this case, information at the arbitration stage needs to be retrieved offline at the trial stage, affecting overall efficiency and accuracy.

[0006] The existing labor dispute resolution mechanism still relies on the traditional single-thread model, which is difficult to cope with the high incidence of labor disputes. Courts, arbitration, and judicial mediation belong to different systems, resulting in poor coordination among the three parties, which increases the time and financial burden on the parties.

[0007] In general, although intelligent labor arbitration has made certain progress in improving efficiency and convenience, it still faces many challenges and defects in technology implementation, system coordination and model innovation, and cannot realize intelligent arbitration based on the arbitration trial thinking chain. Summary of the invention

[0008] In order to solve the above-mentioned defects of intelligent labor arbitration, the present invention provides a method and device for inferring the thought chain of arbitration based on a large language model. The following technical solutions are adopted: The arbitration trial thought chain inference method based on the large language model includes the following steps: Step 1: Input the labor arbitration electronic application documents into the information input module; Step 2: The information extraction module performs text analysis on the application documents based on natural language processing technology to extract keyword data groups; Step 3: The information analysis module identifies and extracts the basic information of the applicant during his / her employment based on the keyword data group of the text analysis. The basic information of the applicant during his / her employment includes: the time of joining the company, the time of leaving the company, the salary structure, the position and the job responsibilities; Step 4: The information analysis module sorts out the disputed items of the case, uses a large language model to analyze the application documents to extract dispute keywords, and classifies the labor dispute items according to keyword clustering to form a list of disputed items; Step 5: The information analysis module searches for laws, regulations and judicial interpretations related to the case, uses knowledge graph technology to associate the disputed matters with legal provisions, compares the factual descriptions in the application documents, and matches appropriate evidence materials; Step 6: The information analysis module uses NLP technology to perform semantic analysis on the applicant's arbitration request and extract key elements of the arbitration request, such as the requested matter, amount, compensation method, etc.

[0009] Match the arbitration request with the disputed matters, establish a mapping relationship between the arbitration request and the disputed matters, analyze the rationality and feasibility of the arbitration request, match it with the disputed matters, and automatically generate an arbitration review thinking chain, which includes the basic information of the applicant, the disputed matters, the arbitration request and the corresponding facts, evidence and legal basis.

[0010] By adopting the above technical solution, the information input module can realize the input of the labor arbitration electronic application documents. The information input module can be in the form of an online information input page, which can realize the input of the labor arbitration electronic application documents and other relevant certification materials prepared by the applicant, and can also upload the relevant complaints and certification materials of the enterprise at the same time. The information analysis module can perform text analysis on the application documents based on natural language processing technology. The implementation form of the information analysis module can be a computer, smart chip or other device with autonomous computing capabilities, and can use natural language processing (NLP) technology to perform text analysis on the application documents. Identify and extract the following information: time of employment, time of leaving, salary composition structure, position, job responsibilities, etc., build a database of the basic information of the applicant, extract keyword data groups, and sort out the disputed matters of the case in the arbitration application documents, which are key steps. NLP technology can be used to analyze the application documents and extract dispute keywords. Dispute categories (such as labor remuneration, termination of labor contracts, etc.) can be divided according to keyword clustering.

[0011] The disputed matters are matched with relevant facts, evidence and legal provisions. The knowledge graph technology is used to associate the disputed matters with relevant legal provisions. The laws, regulations and judicial interpretations involved in the case can be automatically retrieved. The factual descriptions in the application documents are compared to match the appropriate evidence materials.

[0012] Analyze the applicant's arbitration request. Use NLP technology to perform semantic analysis on the applicant's arbitration request. Extract key elements of the arbitration request, such as the requested matter, amount, compensation method, etc.

[0013] Establish a mapping relationship between arbitration requests and dispute matters. Analyze the rationality and feasibility of arbitration requests and match them with dispute matters.

[0014] Automatically generate the arbitration review thinking chain, including the basic information of the applicant, the disputed matters, the arbitration request and the corresponding facts, and finally the corresponding evidence and legal basis.

[0015] Through the above intelligent design, the thinking mode of arbitrators can be simulated to realize the automation of arbitration hearing process. This method helps to improve arbitration efficiency and ensure the fairness of hearing results.

[0016] Optionally, after collecting the basic information of the applicant during employment in step 3, the data is also verified. The verification of the basic information data of the applicant during employment includes the following steps: The data verification module uses optical character recognition technology to scan and verify the supporting documents provided by the applicant and the employee file materials provided by the enterprise, and verifies the correctness of the identification of basic information during the employment period based on the recognition results.

[0017] By adopting the above technical solutions, data verification can extract the accuracy of information. Compare with the employee files provided by the enterprise. Use optical character recognition (OCR) technology to scan and verify relevant certification materials. The specific optical character recognition (OCR) technology is as follows: Adjust the brightness, contrast and resolution of the image to ensure that the text is legible. Use image processing technology to remove noise and spots generated during the scanning process. If the scanned image is tilted, it needs to be corrected through software. Use OCR software to analyze the pre-processed image and identify the text content in the image. Convert the text recognized by OCR into an editable text format, such as TXT or Word document. OCR software should be able to recognize the layout of the text and distinguish different parts such as the title, body, table, etc. Manually check whether the text output by OCR is consistent with the original scan to ensure the accuracy of the information. Correct the text that was incorrectly recognized by OCR, including spelling errors, formatting problems, etc. Classify the extracted text according to a certain structure, such as classifying pay slip information, overtime records, etc. Integrate the OCR-processed data into the arbitration hearing system and associate it with other information of the case.

[0018] Optional, categories of labor disputes include disputes over confirmation of labor relations, disputes over dismissal, resignation and resignation, disputes over working hours and rest vacations, disputes over social insurance and welfare, disputes over labor remuneration and work-related injury medical expenses, and disputes over economic compensation and damages.

[0019] Optionally, in step 2, the information extraction formula is: ; in represents the information extraction function, Indicates the input application documents, Indicates the nth keyword extracted.

[0020] Optionally, the applicant's basic information data during employment can be verified using the following information verification formula: ; Where V represents the information verification function, represents the keyword to be verified, D represents the employee file provided by the enterprise, Indicates the verification result of the mth keyword.

[0021] Optionally, the formula for identifying disputed items in step 4 is: ; R is the dispute item identification function, T is the text content, represents the kth disputed item.

[0022] Optionally, in step 5, the evidence is associated with the legal provisions using the following formula: ; Where A represents the correlation function, represents the kth disputed item, F represents the fact set, L represents the legal provisions set, represents the pth fact associated with the disputed item, Represents the pth legal provision associated with the disputed item.

[0023] Optionally, in step 6, the formula for matching the arbitration request with the disputed matter is as follows: ; M represents the matching function between arbitration request and dispute matters, Indicates the elements of the arbitration request, Indicates disputed items. Indicates the matching result between the arbitration request and the disputed matters.

[0024] Optionally, in step 4, the keyword extraction formula of the large language model is: ; Where K represents the keyword extraction function, T represents the preprocessed text, represents the model parameters, Indicates the nth keyword; The keyword clustering formula is: ; C represents the clustering function, K represents the extracted keyword set, represents the clustering algorithm parameters, represents the mth cluster category; The formula for generating the list of dispute matters is: ; L represents the dispute list generation function, C represents the clustering result, Represents the generated m-th dispute list.

[0025] By adopting the above technical solution, the labor arbitration application documents are preprocessed to obtain the cleaned text. The keyword extraction model (such as BERT) is used to extract the keywords in the text. The clustering algorithm (such as K-means) is applied to cluster the keywords to determine the category of the dispute. Based on the clustering results, a list of dispute items is generated.

[0026] Before model training, a large amount of labeled data is required to train keyword extraction and clustering models.

[0027] The performance of the model depends on the accuracy of keyword extraction and clustering, so the model parameters need to be continuously optimized.

[0028] Labor dispute cases are complex and ever-changing, and the model may need to be updated regularly to adapt to new types of disputes.

[0029] Through the above architecture and formula, a large language model can be implemented that can process labor arbitration application documents, effectively extract dispute keywords and classify dispute matters.

[0030] An arbitration trial thinking chain inference device based on a large language model comprises an information input module, an information extraction module, an information analysis module and a result display module, wherein the information input module is used to input labor arbitration electronic application documents and scanned certification materials, and the information extraction module is connected to the information extraction module in communication, and runs the information extraction program designed in step 2 of the arbitration trial thinking chain inference method based on a large language model, and analyzes the labor arbitration electronic application documents to obtain an extracted keyword data group; The information analysis module is communicated with the information input module and the information extraction module respectively, and adopts the analysis program designed in steps 2 to 6 of the arbitration trial thinking chain inference method based on the large language model. The analysis program is run to obtain the arbitration trial thinking chain analysis results. The result display module communicates and interacts with the information analysis module to analyze the results and display the arbitration trial thinking chain analysis results.

[0031] By adopting the above technical solution, the information input module can be a networked computer, the information extraction module and the information analysis module are computer-based extraction or analysis programs, and the result display module is a display.

[0032] In summary, the present invention includes at least one of the following beneficial technical effects: The present invention can provide an arbitration trial thinking chain inference method and device based on a large language model. The information input module can realize the input of the labor arbitration electronic application document. The information analysis module can perform text analysis on the application document based on natural language processing technology, and use natural language processing (NLP) technology to perform text analysis on the application document. Identify and extract the following information: time of employment, time of resignation, salary composition structure, position, job responsibilities, etc., build a basic information database of the applicant, extract keyword data groups, and sort out the disputed matters of the case in the arbitration application document. It is a key step. Use NLP technology to analyze the application document and extract dispute keywords. According to keyword clustering, the disputed matters can be classified into categories (such as labor remuneration, termination of labor contract, etc.). The disputed matters correspond to relevant facts, evidence, and legal provisions. Use knowledge graph technology to associate the disputed matters with relevant legal provisions. Automatic retrieval of laws, regulations, judicial interpretations, etc. involved in the case can be realized. Compare the factual descriptions in the application document and match appropriate evidence materials. Use NLP technology to perform semantic analysis on the applicant's arbitration request. Extract key elements of the arbitration request, such as the requested matters, amount, compensation method, etc.

[0033] Establish a mapping relationship between arbitration requests and disputed matters. Analyze the rationality and feasibility of arbitration requests and match them with disputed matters. Automatically generate arbitration trial thinking chain, including the basic situation of the applicant, disputed matters, arbitration requests and corresponding facts, and finally corresponding evidence and legal basis.

[0034] Through the above intelligent design, the thinking mode of arbitrators can be simulated to realize the automation of arbitration hearing process. This method helps to improve arbitration efficiency and ensure the fairness of hearing results. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is a flow chart of the arbitration trial thinking chain inference method based on the large language model of the present invention. DETAILED DESCRIPTION

[0036] The present invention is further described in detail below in conjunction with the accompanying drawings.

[0037] The embodiment of the present invention discloses a method and device for inferring arbitration trial thinking chains based on a large language model.

[0038] Reference Figure 1, the arbitration trial thinking chain inference method based on the large language model includes the following steps: Step 1: Input the labor arbitration electronic application documents into the information input module; Step 2: The information extraction module performs text analysis on the application documents based on natural language processing technology to extract keyword data groups; Step 3: The information analysis module identifies and extracts the basic information of the applicant during his / her employment based on the keyword data group of the text analysis. The basic information of the applicant during his / her employment includes: the time of joining the company, the time of leaving the company, the salary structure, the position and the job responsibilities; Step 4: The information analysis module sorts out the disputed items of the case, uses a large language model to analyze the application documents to extract dispute keywords, and classifies the labor dispute items according to keyword clustering to form a list of disputed items; Step 5: The information analysis module searches for laws, regulations and judicial interpretations related to the case, uses knowledge graph technology to associate the disputed matters with legal provisions, compares the factual descriptions in the application documents, and matches appropriate evidence materials; Step 6: The information analysis module uses NLP technology to perform semantic analysis on the applicant's arbitration request and extract key elements of the arbitration request, such as the requested matter, amount, compensation method, etc.

[0039] Match the arbitration request with the disputed matters, establish a mapping relationship between the arbitration request and the disputed matters, analyze the rationality and feasibility of the arbitration request, match it with the disputed matters, and automatically generate an arbitration review thinking chain, which includes the basic information of the applicant, the disputed matters, the arbitration request and the corresponding facts, evidence and legal basis.

[0040] The information entry module can realize the entry of labor arbitration electronic application documents. The information entry module can be in the form of an online information entry page, which can realize the entry of the labor arbitration electronic application documents and other relevant certification materials prepared by the applicant, and can also upload the relevant complaints and certification materials of the enterprise at the same time. The information analysis module can perform text analysis on the application documents based on natural language processing technology. The implementation form of the information analysis module can be a computer, smart chip or other device with autonomous computing capabilities, and can use natural language processing (NLP) technology to perform text analysis on the application documents. Identify and extract the following information: time of employment, time of leaving, salary composition structure, position, job responsibilities, etc., build a database of the applicant's basic situation, extract keyword data groups, and sort out the disputed matters of the case in the arbitration application documents. It is a key step. NLP technology can be used to analyze the application documents and extract dispute keywords. Dispute categories (such as labor remuneration, termination of labor contracts, etc.) can be divided according to keyword clustering.

[0041] The disputed matters are matched with relevant facts, evidence and legal provisions. The knowledge graph technology is used to associate the disputed matters with relevant legal provisions. The laws, regulations and judicial interpretations involved in the case can be automatically retrieved. The factual descriptions in the application documents are compared to match the appropriate evidence materials.

[0042] Analyze the applicant's arbitration request. Use NLP technology to perform semantic analysis on the applicant's arbitration request. Extract key elements of the arbitration request, such as the requested matter, amount, compensation method, etc.

[0043] Establish a mapping relationship between arbitration requests and dispute matters. Analyze the rationality and feasibility of arbitration requests and match them with dispute matters.

[0044] Automatically generate the arbitration review thinking chain, including the basic information of the applicant, the disputed matters, the arbitration request and the corresponding facts, and finally the corresponding evidence and legal basis.

[0045] Through the above intelligent design, the thinking mode of arbitrators can be simulated to realize the automation of arbitration hearing process. This method helps to improve arbitration efficiency and ensure the fairness of hearing results.

[0046] Step 3: After collecting the basic information of the applicant during his / her employment, the data is verified. The verification of the basic information data of the applicant during his / her employment includes the following steps: The data verification module uses optical character recognition technology to scan and verify the supporting documents provided by the applicant and the employee file materials provided by the enterprise, and verifies the correctness of the identification of basic information during the employment period based on the recognition results.

[0047] Data verification can extract the accuracy of information. Compare with the employee files provided by the company. Use optical character recognition (OCR) technology to scan and verify relevant certification materials. The specific optical character recognition (OCR) technology is as follows: Adjust the brightness, contrast and resolution of the image to ensure that the text is legible. Use image processing technology to remove noise and spots generated during the scanning process. If the scanned image is tilted, it needs to be corrected through software. Use OCR software to analyze the pre-processed image and identify the text content in the image. Convert the text recognized by OCR into an editable text format, such as TXT or Word document. OCR software should be able to recognize the layout of the text and distinguish different parts such as the title, body, table, etc. Manually check whether the text output by OCR is consistent with the original scan to ensure the accuracy of the information. Correct the text that was incorrectly recognized by OCR, including spelling errors, formatting problems, etc. Classify the extracted text according to a certain structure, such as classifying pay slip information, overtime records, etc. Integrate the OCR-processed data into the arbitration hearing system and associate it with other information of the case.

[0048] Categories of labor disputes include disputes over confirmation of labor relations, disputes over dismissal, resignation and resignation, disputes over working hours and rest vacations, disputes over social insurance and welfare, disputes over labor remuneration and work-related injury medical expenses, and disputes over economic compensation and damages.

[0049] In step 2, the information extraction formula is: ; in represents the information extraction function, Indicates the input application documents, Indicates the nth keyword extracted.

[0050] The applicant's basic information data during employment is verified using the following information verification formula: ; Where V represents the information verification function, represents the keyword to be verified, D represents the employee file provided by the enterprise, Indicates the verification result of the mth keyword.

[0051] The formula for identifying disputed items in step 4 is: ; R is the dispute item identification function, T is the text content, represents the kth disputed item.

[0052] In step 5, the evidence is associated with the legal provisions using the following formula: ; Where A represents the correlation function, represents the kth disputed item, F represents the set of facts, L represents the set of legal provisions, represents the pth fact associated with the disputed item, Represents the pth legal provision associated with the disputed item.

[0053] In step 6, the formula for matching the arbitration request with the disputed matters is as follows: ; M represents the matching function between arbitration request and dispute matters, Indicates the elements of the arbitration request, Indicates disputed items. Indicates the matching result between the arbitration request and the disputed matters.

[0054] In step 4, the keyword extraction formula of the large language model is: ; Where K represents the keyword extraction function, T represents the preprocessed text, represents the model parameters, Indicates the nth keyword; The keyword clustering formula is: ; C represents the clustering function, K represents the extracted keyword set, represents the clustering algorithm parameters, represents the mth cluster category; The formula for generating the list of dispute matters is: ; L represents the dispute list generation function, C represents the clustering result, Represents the generated m-th dispute list.

[0055] Preprocess the labor arbitration application documents to obtain the cleaned text. Use a keyword extraction model (such as BERT) to extract keywords from the text. Apply a clustering algorithm (such as K-means) to cluster the keywords and determine the category of the dispute. Generate a list of dispute items based on the clustering results.

[0056] Before model training, a large amount of labeled data is required to train keyword extraction and clustering models.

[0057] The performance of the model depends on the accuracy of keyword extraction and clustering, so the model parameters need to be continuously optimized.

[0058] Labor dispute cases are complex and ever-changing, and the model may need to be updated regularly to adapt to new types of disputes.

[0059] Through the above architecture and formula, a large language model can be implemented that can process labor arbitration application documents, effectively extract dispute keywords and classify dispute matters.

[0060] An arbitration trial thinking chain inference device based on a large language model comprises an information input module, an information extraction module, an information analysis module and a result display module, wherein the information input module is used to input labor arbitration electronic application documents and scanned certification materials, and the information extraction module is connected to the information extraction module in communication, and runs the information extraction program designed in step 2 of the arbitration trial thinking chain inference method based on a large language model, and analyzes the labor arbitration electronic application documents to obtain an extracted keyword data group; The information analysis module is communicated with the information input module and the information extraction module respectively, and adopts the analysis program designed in steps 2 to 6 of the arbitration trial thinking chain inference method based on the large language model. The analysis program is run to obtain the arbitration trial thinking chain analysis results. The result display module communicates and interacts with the information analysis module to analyze the results and display the arbitration trial thinking chain analysis results.

[0061] The information input module can be a networked computer, the information extraction module and the information analysis module are computer-based extraction or analysis programs, and the result display module is a display.

[0062] The following specific embodiments are used to illustrate the implementation principle of the arbitration trial thinking chain inference method and device based on the large language model of the present invention: The applicant Zhang San joined the respondent, a certain technology company, on X / X / 2018 as a software engineer. Zhang San resigned on X / X / 2024. Zhang San claimed that the company did not pay overtime pay as agreed and did not pay the year-end bonus when he resigned.

[0063] Summary of Arbitration Request Documents Zhang San started working on X / X / 2018.

[0064] Zhang San's resignation date is X / X / 2024.

[0065] Zhang San’s salary structure consists of basic salary + performance bonus + overtime pay.

[0066] Zhang San often worked overtime during his tenure, but the company did not pay him overtime pay.

[0067] The company stipulates that the year-end bonus is 10% of the employee's annual salary, but Zhang San did not receive the year-end bonus when he left.

[0068] Part I: Summarize the applicant’s basic information during employment; ; in represents the information extraction function, Indicates the input application documents, Indicates the nth keyword extracted.

[0069] in: =\text{"May 1, 2018"}; b1="2018 X month X day" (date of joining the company) =\text{"2024 X month X day"}; = "October 31, 2023" (resignation date) =\text{"basic salary + performance bonus + overtime pay"}; = "basic salary + performance bonus + overtime pay" (salary structure); Part II: Sorting out the controversial issues in the case Core formula application ; Where A represents the correlation function, represents the kth disputed item, F represents the set of facts, L represents the set of legal provisions, represents the pth fact associated with the disputed item, Represents the pth legal provision associated with the disputed item.

[0070] in: =\text{"Unpaid overtime"}d1="Unpaid overtime" =\text{"Zhang San's overtime record"}f1="Zhang San's overtime record" \text{"Overtime application form, attendance record"}e1="Overtime application form, attendance record" =\text{"Labor Law Article 41: If an employer arranges overtime work, it shall pay overtime pay"}l1="Labor Law Article 41: If an employer arranges overtime work, it shall pay overtime pay" =\text{"Year-end bonus not paid"}d2="Year-end bonus not paid" =\text{"Company's year-end bonus payment policy"}f2="Company's year-end bonus payment policy" \text{"Company rules and regulations, labor contract"}e2="Company rules and regulations, labor contract" =\text{"Labor Contract Law Article 47: The employer shall pay labor remuneration in accordance with the agreement"}l2="Labor Contract Law Article 47: The employer shall pay labor remuneration in accordance with the agreement" Part III: Automated Analysis of the Claimant’s Arbitration Request Preprocess the labor arbitration application documents to obtain the cleaned text. Use a keyword extraction model (such as BERT) to extract keywords from the text. Apply a clustering algorithm (such as K-means) to cluster the keywords and determine the category of the dispute. Generate a list of dispute items based on the clustering results.

[0071] The final dispute items are: the corresponding arbitration requests are: "request for payment of overtime pay" and "request for payment of year-end bonus", and the corresponding dispute matters are "failure to pay overtime pay" and "failure to pay year-end bonus"; Through the above method, we obtained the following arbitration trial thinking chain: The basic information of Zhang San is provided as follows: Zhang San’s joining time is X month X day, 2018.

[0072] Zhang San's resignation date is X / X / 2024.

[0073] Zhang San’s salary structure consists of basic salary + performance bonus + overtime pay.

[0074] Zhang San often worked overtime during his tenure, but the company did not pay him overtime pay.

[0075] The company stipulates that the year-end bonus is 10% of the employee's annual salary, but Zhang San did not receive the year-end bonus when he left.

[0076] The disputed matters were: “failure to pay overtime wages” and “failure to pay year-end bonuses”; The arbitration requests are: "request for payment of overtime pay" and "request for payment of year-end bonus"; The evidence is: Zhang San's overtime records, overtime application form, attendance records, and the company's year-end bonus distribution policy; The legal basis is: Article 41 of the Labor Law and Article 47 of the Labor Contract Law.

[0077] The above are all preferred embodiments of the present invention, and are not intended to limit the protection scope of the present invention. Therefore, any equivalent changes made based on the structure, shape, and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. The arbitration trial thought chain inference method based on the large language model is characterized by: The following steps are involved: Step 1: Input the labor arbitration electronic application documents into the information input module; Step 2: The information extraction module performs text analysis on the application documents based on natural language processing technology to extract keyword data groups; Step 3: The information analysis module identifies and extracts the basic information of the applicant during his / her employment based on the keyword data group of the text analysis. The basic information of the applicant during his / her employment includes: the time of joining the company, the time of leaving the company, the salary structure, the position and the job responsibilities; Step 4: The information analysis module sorts out the disputed items of the case, uses a large language model to analyze the application documents to extract dispute keywords, and classifies the labor dispute items according to keyword clustering to form a list of disputed items; Step 5: The information analysis module searches for laws, regulations and judicial interpretations related to the case, uses knowledge graph technology to associate the disputed matters with legal provisions, compares the factual descriptions in the application documents, and matches appropriate evidence materials; Step 6: The information analysis module uses NLP technology to perform semantic analysis on the applicant's arbitration request and extract key elements of the arbitration request; Match the arbitration request with the disputed matters, establish a mapping relationship between the arbitration request and the disputed matters, analyze the rationality and feasibility of the arbitration request, match it with the disputed matters, and automatically generate an arbitration review thinking chain, which includes the basic information of the applicant, the disputed matters, the arbitration request and the corresponding facts, evidence and legal basis.

2. The arbitration trial thought chain inference method based on a large language model according to claim 1 is characterized by: Step 3: After collecting the basic information of the applicant during his / her employment, the data is verified. The verification of the basic information data of the applicant during his / her employment includes the following steps: The data verification module uses optical character recognition technology to scan and verify the supporting documents provided by the applicant and the employee file materials provided by the enterprise, and verifies the correctness of the identification of basic information during the employment period based on the recognition results.

3. The arbitration trial thought chain inference method based on a large language model according to claim 2 is characterized by: Categories of labor disputes include disputes over confirmation of labor relations, disputes over dismissal, resignation and resignation, disputes over working hours and rest vacations, disputes over social insurance and welfare, disputes over labor remuneration and work-related injury medical expenses, and disputes over economic compensation and damages.

4. The arbitration trial thought chain inference method based on a large language model according to claim 3 is characterized by: In step 2, the information extraction formula is: ； in represents the information extraction function, Indicates the input application documents, Indicates the nth keyword extracted.

5. The arbitration trial thought chain inference method based on a large language model according to claim 4 is characterized by: The applicant's basic information data during employment is verified using the following information verification formula: ； Where V represents the information verification function, represents the keyword to be verified, D represents the employee file provided by the enterprise, Indicates the verification result of the mth keyword.

6. The arbitration trial thought chain inference method based on a large language model according to claim 5 is characterized by: The formula for identifying disputed items in step 4 is: ； R is the dispute item identification function, T is the text content, represents the kth disputed item.

7. The arbitration trial thought chain inference method based on a large language model according to claim 6 is characterized by: In step 5, the evidence is associated with the legal provisions using the following formula: ； Where A represents the correlation function, represents the kth disputed item, F represents the set of facts, L represents the set of legal provisions, represents the pth fact associated with the disputed item, Represents the pth legal provision associated with the disputed item.

8. The arbitration trial thought chain inference method based on a large language model according to claim 7 is characterized by: In step 6, the formula for matching the arbitration request with the disputed matters is as follows: ； M represents the matching function between arbitration request and dispute matters, Indicates the elements of the arbitration request, Indicates disputed items. Indicates the matching result between the arbitration request and the disputed matters.

9. The arbitration trial thought chain inference method based on a large language model according to claim 8 is characterized by: In step 4, the keyword extraction formula of the large language model is: ； Where K represents the keyword extraction function, T represents the preprocessed text, represents the model parameters, Indicates the nth keyword; The keyword clustering formula is: ； C represents the clustering function, K represents the extracted keyword set, represents the clustering algorithm parameters, represents the mth cluster category; The formula for generating the list of dispute matters is: ； L represents the dispute list generation function, C represents the clustering result, Represents the generated m-th dispute list.

10. An arbitration trial thinking chain inference device based on a large language model, characterized by: It includes an information input module, an information extraction module, an information analysis module and a result display module. The information input module is used to input the labor arbitration electronic application documents and scan the supporting materials. The information extraction module is connected to the information extraction module in communication. The information extraction program designed in step 2 of the arbitration trial thinking chain inference method based on the large language model described in claim 9 is run to analyze the labor arbitration electronic application documents to obtain an extracted keyword data group; The information analysis module is communicated with the information input module and the information extraction module respectively, and adopts the analysis program designed in steps 2 to 6 of the arbitration trial thinking chain inference method based on the large language model described in claim 9, and runs the analysis program to obtain the arbitration trial thinking chain analysis results. The result display module communicates and interacts with the information analysis module to analyze the results and displays the arbitration trial thinking chain analysis results.

Images