A collaborative legal supply and demand matching platform integrating legal NLP and implementation method

Abstract

The application relates to the technical field of artificial intelligence processing of legal information, and particularly discloses a collaborative legal supply-demand matching platform fusing legal NLP and an implementation method, which automatically deconstructs case materials through a legal exclusive NLP module, extracts core elements, and automatically generates legal documents, converts the traditional time-consuming and labor-consuming repetitive work into an automatic process, accurately identifies legal entities and relationships through semantic understanding and logical deduction of a knowledge graph, automatically matches a case cause, calls the latest regulations and pushes similar cases, and through a twelve-dimensional weighted scoring model and a dynamic activity degree addition algorithm, automatically scans legal logic loopholes and practice risks through intelligent compliance review, and chains and stores the hash operation log of collaborative revision as evidence. The application significantly improves the matching efficiency and quality between lawyers and cases, and improves the intelligence and accuracy of supply-demand matching.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence processing technology for legal information, and specifically discloses a collaborative legal supply and demand matching platform and implementation method that integrates legal NLP. Background Technology

[0002] The rapid development of artificial intelligence and big data, along with the continuous advancement of the rule of law and the ongoing improvement and updating of the legal system, have led to increasingly complex and stringent compliance requirements for the business activities of various market entities. This has promoted the intelligent and information-based development of legal services and further fostered the formation of smart legal services.

[0003] While the smart legal services industry has achieved some success, it still suffers from four major technological shortcomings and lacks interfaces for next-generation computing power. First, legal semantic analysis is superficial. Existing systems only support text organization and legal provision retrieval, lacking a dedicated legal NLP architecture. This prevents in-depth case analysis, extraction of legal relationships, and verification of evidence chains. The sorting of points of contention and the drafting of court documents rely entirely on manual operation, resulting in low efficiency, difficulty in standardizing case handling, and poor adaptability to practical scenarios. Second, collaborative management and compliance systems are lacking. The industry lacks standardized hierarchical access control, full-process operation tracking, and accountability mechanisms. Document content cannot undergo automated compliance verification, easily leading to professional compliance risks. Online case data and offline case file data are fragmented, and confidential case data lacks hierarchical isolation and protection mechanisms, posing significant information leakage and security risks. Third, the supply and demand matching model for legal services is crude and inefficient. Industry resources are scattered, and there is a significant information asymmetry between supply and demand. Existing matching algorithms are simplistic and lack accuracy. The order placement, performance supervision, and service evaluation systems are incomplete, hindering the industry's ability to achieve large-scale, standardized operation and development. Fourth, the technological barriers are weak and the computing power expansion capability is insufficient. The core computing algorithms of similar products have not been solidified and locked, making them easy for competitors to imitate, replace and circumvent, resulting in insufficient core competitiveness of the products. The existing algorithm system is only compatible with classical computing architecture and has not been designed with quantum computing adaptation interfaces, making it impossible to connect with the next generation of computing power technologies in the future, thus limiting the long-term iterative upgrade capability of the system. Summary of the Invention

[0004] The purpose of this invention is to solve the problem that existing general natural language processing technologies cannot deeply analyze legal text content and the accuracy of identifying legal relationships in cases does not meet the requirements of practical case handling.

[0005] According to one aspect of the present invention, a collaborative legal supply and demand matching implementation method integrating legal NLP is provided, comprising the following steps: Step 1: By receiving legal service requests and case information, obtain the original materials of legal cases, perform image correction and format standardization preprocessing on the original materials, and store the standardized data into the case database. Step 2: Deconstruct the case content using the legal NLP semantic parsing module to extract the core elements of the case, including the parties involved, the facts of the case, the evidence, and the rights and responsibilities. Step 3: Perform vector cosine similarity comparison between the case facts parsed in Step 2 and the case information in the pre-stored case information database, automatically determine the type of the case and recommend three matching case information, and retrieve the applicable legal provisions from the legal database based on the case information and case facts. Step 4: Based on the case subject parsed in Step 2 and the cause of action matched in Step 3, extract the points of contention from the original materials. The points of contention include substantive disputes and procedural disputes. Use hard coding to solidify four fixed weight values ​​to unify and normalize the various indicators of the case into a score in the range of zero to one, so as to sort the points of contention from high to low scores. Step 5: Based on the dynamic document template library that distinguishes between case causes of action, trial location, trial level, and agent status, the template content is automatically filled in by combining the analysis results of Step 2, the applicable legal provisions of Step 3, and the disputed issues sorted in Step 4, and the corresponding trial materials are generated. Step 6: Based on the difficulty of the case and the field of business, intelligently match the corresponding level of practitioners and assign exclusive business operation permissions; Step 7: Based on the legal service needs in Step 1 and the corresponding level of practitioners in Step 6, calculate the comprehensive score through similarity, and match the lawyers with the top preset number of comprehensive scores as matching lawyers to supply and demand, forming an electronic service order; Step 8: Sequence the fulfillment nodes of the electronic service order are broken down into standard operation nodes and bound to execution time limits. The execution status of each node is monitored in real time. When the dwell time of any node exceeds the execution time limit, an early warning notification is automatically generated. When the upload operation of the deliverable is completed, a compliance review is carried out. The deliverable that passes the review is automatically pushed to the user end, and the settlement and payment process is triggered after the user confirms receipt.

[0006] The principle and effect of this basic scheme are as follows: By receiving and processing raw materials, breaking down case details, extracting relevant elements and information, matching the cause of action, automatically pushing applicable legal provisions, scoring and ranking multiple points of contention, automatically generating court hearing materials, assigning corresponding permissions, and automatically generating orders through transactions, the system can deeply analyze legal text content and accurately identify legal relationships in cases to meet practical case-handling needs.

[0007] Furthermore, in step 6, revision information of the court hearing materials generated in step 5 is obtained from at least two users with different permission levels. The permission levels include three levels: lead lawyer, co-lead lawyer, and trainee lawyer. The revision information is recorded in real time at the character-level difference granularity to generate a hash operation log containing fields such as the reviser's identifier, operation timestamp, operation IP address, text content before and after revision, and revision type. The hash operation log is then stored in the blockchain evidence storage network.

[0008] Furthermore, in step 3, after each retrieval of the applicable legal provisions, it automatically determines whether there is a corresponding updated or repealed judicial interpretation. If there is an updated or repealed judicial interpretation, the judicial interpretation information of the applicable legal provisions is changed, and the user of the legal service is inferred to determine whether to actively push the judicial interpretation based on their level of legal knowledge. The level of knowledge includes proficiency, comprehension, and unfamiliarity. When the user is proficient, the judicial interpretation is not pushed; when the user is comprehension, the applicable legal provisions and their corresponding judicial interpretations are actively pushed; when the user is unfamiliar, the judicial interpretation is pushed, and similar precedents are obtained. The beneficial effects are: by judging the judicial interpretations corresponding to the matched applicable legal provisions, and pushing different information according to the user's level of knowledge, accurate information can be pushed without consuming additional system resources during use, ensuring that the user can understand the legal information as much as possible.

[0009] Furthermore, in step 3, the estimation of the current user's level of legal information knowledge is based on a comprehensive score of the current user's age group, job type, education level, and whether they have engaged in legal activities, and the comprehensive score is used to determine the level of legal information knowledge. The beneficial effect is that by comprehensively scoring the current user's awareness of legal information from multiple perspectives, the accuracy of determining the level of awareness is improved.

[0010] Furthermore, in step 3, after the applicable legal provision is retrieved, the corresponding frequency statistics are performed. A frequency judgment is performed every preset period. When the frequency is less than the threshold, a judicial interpretation update or repeal judgment for the legal provision is actively performed. When the frequency is greater than or equal to the threshold, the judicial interpretation update or repeal judgment is performed after the legal provision is retrieved. The beneficial effects are: by conducting separate judicial interpretation updates and repeal judgments for some legal provisions with low application frequency, the timeliness and accuracy of judicial interpretation update and repeal information extraction are ensured, and information gaps are avoided.

[0011] Furthermore, it also includes step 9, which involves scoring services based on service professionalism, service attitude, processing efficiency, and case completion results. The professional level and order allocation weight are dynamically adjusted based on the service score. During the dynamic adjustment process, when service scores are the same, the service personnel with service scores higher than the set value are selected to determine whether their case progress has any pending cases. If so, the end time of the pending cases is estimated, and the service score of the service personnel is increased based on the end time.

[0012] The beneficial effects are: by judging the end time of pending cases, the service score can be dynamically adjusted according to the real-time situation, thereby improving timeliness and accuracy.

[0013] According to another invention of the present invention, a collaborative legal supply and demand matching platform integrating legal NLP is provided, comprising: The case material collection module is used to collect the original materials of legal cases and preprocess the original materials to obtain unstructured text. The original materials include materials from legal service needs, materials from various interface terminals and data in various formats. The legal NLP semantic parsing module is used to extract core elements of case subjects, case facts, case evidence, and rights and responsibilities from unstructured text; The automatic case cause matching and legal retrieval module is used to compare the parsed case facts with the case cause information in the pre-stored case cause database using vector cosine similarity, automatically determine the type of the case and recommend three types of matching case causes, and retrieve applicable legal provisions from the legal database based on the case cause and case fact elements. The fixed-weight dispute focus sorting module is used to extract the dispute focus in a case and use hard-coded fixed weight values ​​to normalize the various indicators of the case into a score between zero and one, so as to sort the dispute focus according to the score from high to low. The multi-role court hearing document intelligent generation module is used to build a dynamic document template library that distinguishes the cause of action, trial location, trial level, and agent status. It automatically fills in the template content based on NLP analysis results, points of contention, and applicable legal provisions to generate court hearing documents for the corresponding role. The lawyer hierarchical access control module is used to classify practitioners into four levels based on the RBAC role-based access control architecture: specially appointed senior legal experts, star-rated practicing lawyers, ordinary practicing lawyers, and on-site legal specialists. The collaborative revision full-process traceability module is used to record account information, operation time, device IP address, and content before and after modification in real time, and automatically generate an independent document version snapshot for each content adjustment; The intelligent compliance review module for case handling content is used to identify document problems through a triple verification mode of keyword retrieval, text semantic comparison, and legal logic in a compliance review feature library with pre-set legal service violation semantics. The Level 3 Case Data Security Protection Module is used to divide case information into three storage levels according to the sensitivity of the information; The cross-system forced data synchronization module is used to perform incremental real-time synchronization with a preset duration, combined with a full data verification operation mode at a preset time every day. It uses a private encryption protocol to transmit data and a unified standard message format. The lawyer qualification management module is used to connect to the official practice information verification interface to conduct real-name authentication of the identity of the lawyer, verification of the authenticity of the practice certificate, and filing of practice information. The legal service demand collection module is used to collect business demands, extract information such as case type, service area, dispute claims, service budget, urgency of handling, and lawyer selection preferences, generate 20-dimensional structured demand tags, and uniformly archive and store them in a preset demand database. The supply and demand two-way matching module is used to calculate a comprehensive score through twelve weighted scoring features. The weighted scoring features include professional field fit, years of practice, number of similar cases handled, case win probability, user reputation rating, service area matching, order response speed, cost budget fit, case urgency, lawyer workload, platform star rating, and historical cooperation evaluation. Lawyers with the highest comprehensive scores among the top preset number are matched with lawyers. The full-scenario online legal service integration module is used to integrate all business portals such as online legal consultation, legal document drafting, contract review and revision, case process handling, corporate legal counsel, dispute mediation, and court appearance assistance. It opens service access to the outside world through multiple channels such as computer web pages, mobile mini programs, and mobile APPs, and all business operations are recorded and archived online. The service process visualization and control module is used to retain text and voice communication records, business performance vouchers, case handling nodes, and progress change snapshots; The standardized order fulfillment module is used to automatically generate electronic service orders based on the supply and demand matching results. Both parties confirm the order's validity online with electronic signatures, and the business nodes are verified and accepted online after completion. The service rating and industry order management module is used to score services based on service professionalism, service attitude, processing efficiency, and case completion results, and dynamically adjust the professional level and order allocation weight based on the service rating.

[0014] Furthermore, the four fixed weight values ​​include the weight value of the amount involved, the weight value of the fault liability division, the weight value of the judgment tendency of similar cases, and the weight value of the priority of court debate, which are respectively set to 0.25, 0.3, 0.25, and 0.2.

[0015] Furthermore, it also includes a quantum dual-mode algorithm scheduling module, which is natively adapted to the classical computer operation mode, while standardizing and reserving extended interfaces for quantum parallel operation, quantum semantic retrieval, quantum key encryption, and quantum feature matching, and can be seamlessly connected to quantum computing hardware devices to achieve system computing power upgrades. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 A flowchart illustrating a collaborative legal supply and demand matching implementation method integrating legal NLP, as proposed in an embodiment of this application, is shown. Figure 2 This illustration shows a schematic diagram of a collaborative legal supply and demand matching platform and implementation method that integrates legal NLP, as proposed in an embodiment of this application. Detailed Implementation

[0018] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0019] Example 1

[0020] A collaborative legal supply and demand matching platform integrating legal NLP includes a case material collection module, a legal NLP semantic parsing module, an automatic case cause matching and legal regulation retrieval module, a fixed-weight dispute focus sorting module, a multi-role court hearing material intelligent generation module, a lawyer hierarchical permission control module, a collaborative revision full-process traceability module, a case handling content intelligent compliance review module, a three-level case data security protection module, a cross-system mandatory data synchronization module, a lawyer onboarding qualification control module, a legal service demand collection module, a supply and demand two-way matching module, a full-scenario online legal service integration module, a service full-process visualization control module, a standardized order fulfillment module, a service rating and industry order control module, and a quantum dual-mode algorithm scheduling module.

[0021] The case material collection module is configured to collect raw materials of legal cases, and preprocess the raw materials to obtain unstructured text. The raw materials include materials derived from legal service demands, and materials from multiple interface terminals and data in multiple formats. The interface terminals include a PC end, a mobile end (mini-program / APP), an API interface, etc., and the formats of the materials may be images, PDF, Word, Excel, audio (court hearing recording), video (evidence video), etc.

[0022] For the raw materials in image / PDF format, the preprocessing may call an existing OCR (Optical Character Recognition) engine to identify text regions in the image and convert the text regions into plain text characters. For example, plain text characters "I have borrowed a total of 100,000 RMB from Li ..." are identified from a photo of an IOU.

[0023] For the raw materials in audio / video format, the preprocessing may call an existing speech recognition engine to transcribe human voice dialogues into a text script with timestamps.

[0024] For the raw materials in Word / Excel format, the preprocessing directly extracts the internal text content thereof.

[0025] The legal NLP semantic parsing module is configured to extract core elements of case subjects, case facts, case evidence, rights and responsibilities from the unstructured text. The case subjects include parties, lawyers, judges, case numbers, disputed subject matters, competent courts, litigation claims, etc. The case facts include contractual relationships, tort relationships, marriage relationships, labor relationships, etc., and parse the corresponding timeline, capital flow, and obligation performance status.

[0026] The automatic cause of action matching and regulation retrieval module is configured to perform vector cosine similarity comparison between the parsed case facts and the cause of action information in a pre-stored cause of action library, automatically determine the type to which the case belongs and recommend three types of matched causes of action, and retrieve applicable legal provisions from a regulation library according to the causes of action and the case fact elements. The cause of action library includes legal provisions of civil law, commercial law, etc. The matching accuracy rate is not less than 95%, the system is connected to the official legal provision and judgment case database in real time, deletes invalid and abolished legal provisions, and synchronously retrieves applicable regulations of the trial jurisdiction and reference cases of similar effective judgments.

[0027] The fixed-weight dispute focus ranking module extracts the disputed issues in a case and uses hard-coded fixed weight values ​​to normalize all indicators of the case into scores ranging from zero to one, ranking the disputed issues from highest to lowest score. Utilizing text summarization and graph neural networks, it extracts the 3-7 most likely disputed issues in a case, such as "whether a breach of contract exists" and "whether the liquidated damages are excessive." The four fixed weight values ​​include weights for the amount involved, fault liability, the tendency of judgments in similar cases, and the priority of courtroom arguments, with values ​​of 0.25, 0.3, 0.25, and 0.2 respectively. A comprehensive score is calculated using the formula S = Factual Score × 0.25 + Liability Score × 0.30 + Precedent Score × 0.25 + Courtroom Score × 0.20, and the disputed issues are ranked from highest to lowest score.

[0028] The multi-role intelligent court hearing document generation module is used to build a dynamic document template library that distinguishes between case causes of action, trial location, trial level, and agent status. It automatically fills in the template content with NLP analysis results, points of contention, and applicable legal provisions to generate court hearing documents for the corresponding role. For example, for the plaintiff, it generates court hearing documents such as the complaint, evidence list, draft cross-examination opinions, and trial outline (including a question list); for the defendant, it generates court hearing documents such as the answer, counterclaim template, and key points of evidence rebuttal; and for the judge / assistant, it generates court hearing element table, court hearing report, and a list of matters to be verified.

[0029] The lawyer tiered access control module, based on the RBAC (Role-Based Access Control) architecture, categorizes practitioners into four levels: Senior Legal Experts, Star-Rated Lawyers, Regular Lawyers, and On-Site Legal Specialists. For example, different levels of practitioners have permissions limited to viewing only the evidence they are responsible for. Senior Legal Experts have full permissions for content editing, practice optimization, content review, and final drafting; Star-Rated Lawyers have editing and optimization and content review permissions; Regular Lawyers can only edit and modify initial drafts; and On-Site Legal Specialists are limited to browsing documents and annotating content. The system is bound to a JWT identity token, has a preset fixed permission matrix, and intercepts operation requests that fail permission verification. Permission configuration parameters cannot be modified in the backend.

[0030] The collaborative revision full-process traceability module records account information, operation time, device IP address, and content before and after modification in real time. It automatically generates an independent document version snapshot for each content adjustment and uses the existing SHA-256 hash algorithm combined with digital signature technology to verify data integrity. All operation records are permanently saved and cannot be deleted or tampered with. Logs can be retrieved by case number, operator, and operation time to accurately trace business responsibility.

[0031] The intelligent compliance review module for case handling content is used to identify document problems through a triple verification mode of keyword search, text semantic comparison, and legal logic by using a compliance review feature library with pre-set legal service violation semantics. The compliance review feature library has more than 2,000 pre-built legal service violation semantics to automatically identify illegal expressions, logical loopholes, and non-compliant litigation claims, classify risk levels, and mark the location of problems.

[0032] The three-tier case data security protection module categorizes case information into three storage levels based on its sensitivity. For example, customer ID cards, addresses, and asset information are classified as private / privacy level; case handling strategies and unpublished litigation materials are classified as internal confidential level; and publicly available cases and platform promotional content are classified as public business level. These three types of data are stored in independent databases, and cross-level data retrieval requires a multi-level, dual-person approval process. Private / privacy level data is encrypted using the national standard SM4 algorithm with masking for data anonymization, while internal confidential data is encrypted using the AES-256 algorithm. The keys are periodically rotated and updated, and all data access operation logs are retained for at least three years.

[0033] The cross-system forced data synchronization module is used for incremental real-time synchronization with a preset duration, combined with a full data verification operation mode at a preset time every day. It uses a private encryption protocol to transmit data, with a unified standard message format. The preset duration and preset time can be adjusted according to actual needs, such as a preset duration of five minutes and a preset time of 0:00 AM or 1:00 AM. The private encryption protocol uses encryption content defined according to actual needs. For example, the private encryption protocol encrypts data by adding additional identical data at a set location in the data to disrupt the actual viewing of the data.

[0034] The lawyer onboarding qualification management module is used to connect to the official practice information verification interface to conduct real-name authentication of onboarding personnel, verification of the authenticity of practice certificates, and filing of practice information, such as including qualification resumes, case handling cases, service reputation, star rating information, and setting up a blacklist mechanism for violations. Personnel with illegal or irregular behavior will be permanently restricted from onboarding, such as verifying names, ID numbers, lawyer practice certificate numbers, and live faces.

[0035] The legal service demand collection module is used to collect business demands, extract information such as case type, service area, dispute claims, service budget, urgency of handling, and lawyer selection preferences, generate 20-dimensional structured demand tags, and uniformly archive and store them in a preset demand database. It also collects business demands through form entry and intelligent dialogue modes.

[0036] The supply and demand two-way matching module is used to calculate a comprehensive score through twelve-dimensional weighted scoring features. These features include professional field fit, years of practice, number of similar cases handled, case win probability, user reputation rating, service region matching, order response speed, cost budget fit, case urgency, lawyer workload, platform star rating, and historical cooperation evaluation. Lawyers with the highest comprehensive scores among the top preset number are selected as matching lawyers. The preset number is set according to actual needs, for example, 5.

[0037] The integrated online legal service module integrates all business portals, including online legal consultation, legal document drafting, contract review and revision, case process handling, corporate legal counsel, dispute mediation, and court appearance assistance. It provides access to services through multiple channels, including computer web pages, mobile mini-programs, and mobile apps, and all business operations are recorded and archived online.

[0038] The end-to-end service visualization and management module is used to retain text and voice communication records, business performance certificates, case handling milestones, and progress change snapshots. For example, it allows querying the entire process of business records based on order number, user information, and lawyer information. Alarms are automatically triggered when business processing exceeds preset time limits, ensuring that services proceed on schedule.

[0039] The standardized order fulfillment module is used to automatically generate electronic service orders based on the supply and demand matching results. Both parties confirm the order's validity through online electronic signatures, and the business nodes are verified and accepted online upon completion.

[0040] The service rating and industry order management module is used to score services based on service professionalism, service attitude, processing efficiency, and case completion results, and dynamically adjust the professional level and order allocation weight based on the service score.

[0041] It also includes a quantum dual-mode algorithm scheduling module, which is natively adapted to the classical computer operation mode, while standardizing and reserving extended interfaces for quantum parallel computing, quantum semantic retrieval, quantum key encryption, and quantum feature matching, which can be seamlessly connected to quantum computing hardware devices to upgrade the system's computing power.

[0042] The system in this embodiment receives and processes raw materials, then breaks down the case details, extracts relevant elements and information, matches the cause of action, automatically pushes applicable legal provisions, scores and ranks multiple points of contention, automatically generates court hearing materials, assigns corresponding permissions, automatically stamps and generates orders for transactions, and retains all historical data to avoid data overwriting and corruption. Through a hash signature evidence storage mechanism, it ensures that the operation records are authentic, complete, and unalterable. With twelve evaluation feature dimensions, it optimizes the scoring weight ratio, improves resource matching fit, and deeply analyzes the legal text content. The system has high accuracy in identifying legal relationships in cases, thus meeting the needs of practical case handling.

[0043] Example 2

[0044] A collaborative legal supply and demand matching method integrating legal NLP, based on a collaborative legal supply and demand matching platform integrating legal NLP as described in Example 1, such as... Figure 1 As shown, it includes the following steps: Step 1 involves receiving legal service requests and case information, obtaining the original materials for the legal cases, and performing image correction and format standardization preprocessing on the original materials. The standardized data is then stored in the case database. Original materials are obtained through interfaces such as PC, mobile (mini-program / APP), and API interfaces. Preprocessing includes noise reduction, tilt correction, and layout segmentation for image data.

[0045] Step 2: The case content is broken down using a legal NLP semantic parsing module to extract core elements such as the case subjects, facts, evidence, and rights and responsibilities. Figure 2 As shown, NLP semantic parsing uses the existing BERT-BiLSTM-CRF deep learning architecture for legal named entity recognition. The BERT encoding layer generates dynamic vectors with contextual semantics for each character, accurately distinguishing the meaning of "owe money" in different sentences as "owes someone money" or "is owed money by someone." The BiLSTM sequence modeling layer captures long-distance dependencies, understanding logical relationships across paragraphs such as "the plaintiff claims...the defendant argues...". The CRF label decoding layer ensures that the entity label sequence conforms to legal logic, preventing the label "lawyer's name" from immediately following the word labeled "plaintiff." Models such as RoBERTa-wwm-GRU-Attention-VAT are used to extract entities, and graph neural networks (GNNs) are used to handle complex dependencies between entities. Lightweight classifiers specifically designed for the legal domain, such as ReLeVAnT, are employed. For multi-label, multi-level classification, a multi-model fusion method such as ELMo-BERT is used. During extraction, an existing bidirectional matching entity relationship extraction method based on causal discovery is adopted. The extracted content includes: case subject information such as plaintiff, defendant, third party, etc.; case facts such as signing time, performance status, description of breach of contract or tort event, etc.; legal relationships such as contractual relationship, tort relationship, marital relationship, etc.; and litigation claims such as compensation amount, confirmation of rights, termination of contract, etc.

[0046] Step 3: Perform a vector cosine similarity comparison between the case facts parsed in Step 2 and the case information in the pre-stored case information database. Automatically determine the case type and recommend three matching case information categories. Retrieve applicable legal provisions from the legal database based on the case information and factual elements. The cosine similarity comparison calculation formula is: ; in, To obtain the final vector for the current case, a hybrid coding strategy is employed, which encodes elements from different modalities separately and then weights and fuses them to obtain the final vector. This involves iterating through all the cause vectors in the cause library.

[0047] The hybrid encoding strategy is as follows: For each factual element (e.g., "contract signing date 2023-01-01", "subject matter is a house", "defendant failed to pay on time"), a pre-trained legal BERT model (e.g., Lawformer, Legal-RoBERTa) is used to encode it separately, obtaining a vector representation of each element; the structured factual text output from step 2 (i.e., the case summary after removing unstructured noise) is input into the legal BERT model, and the vector of the [CLS] token is taken as the global semantic representation; for numerical features such as the amount in dispute and the time difference (e.g., the number of days until the statute of limitations expires), Min-Max normalization is performed, and then mapped to the same dimension as the text vector through a linear layer to form a numerical feature vector. The three vectors obtained by the hybrid encoding strategy are summed by attention weighting, and the weighting coefficients are obtained through training using a labeled case classification dataset, for example, by using MLP to learn the optimal fusion weights. Step 4: Based on the case subject analyzed in Step 2 and the cause of action matched in Step 3, extract the points of contention from the original materials. The points of contention include substantive disputes and procedural disputes. Use hard-coded fixed weight values ​​to normalize the various indicators of the case into a score between zero and one, and sort the points of contention according to the scores from high to low. The fixed weight values ​​include the weight value of the amount involved, the weight value of the fault liability division, the weight value of the judgment tendency of similar cases, and the weight value of the priority of court debate, which are respectively set to 0.25, 0.3, 0.25, and 0.2.

[0048] Based on the cause of action, such as "private lending dispute," all legal provisions tagged with this cause of action are quickly selected from the legal database to form a candidate legal set. Matching is then calculated from this candidate legal set, comparing the constituent elements of the legal provisions (such as "lenders and borrowers," "payment of loan," and "agreed interest") with the factual elements of the case extracted in step 2. A matching network based on a pre-trained model (such as BERT-Siamese) is used to match and score the "legal element sentences and case fact sentences." For elements involving numerical values ​​such as amount and time, such as "loan amount exceeding 100,000 yuan," logical judgment is directly performed against the numerical values ​​in the case facts before calculating the score.

[0049] Step 5: Based on the dynamic document template library that distinguishes between case causes of action, trial location, trial level, and agent status, the template content is automatically filled in by combining the analysis results from Step 2, the applicable legal provisions from Step 3, and the disputed issues sorted from Step 4, generating trial materials for the corresponding roles.

[0050] Step 6: Based on the difficulty of the case and the business field, intelligently match practitioners of corresponding levels and assign exclusive business operation permissions. Obtain revision information of the court hearing materials generated in Step 5 from at least two users with different permission levels. The permission levels include three levels: lead lawyer, co-lead lawyer, and trainee lawyer. Record the revision information in real time with character-level differences to generate a hash operation log containing fields such as the reviser's identifier, operation timestamp, operation IP address, text content before and after revision, and revision type. Store the hash operation log in the blockchain evidence storage network.

[0051] Step 7: Based on the legal service needs in Step 1 and the corresponding level of practitioners in Step 6, a comprehensive score is calculated using similarity. Lawyers with the highest comprehensive scores (within a predetermined number) are matched for supply and demand, generating electronic service orders. The comprehensive score is weighted by factors such as professional field relevance, years of practice, number of similar cases handled, case win probability, user reputation rating, service region matching, order response speed, budget fit, case urgency, lawyer workload, platform star rating, and historical cooperation evaluations. Data such as "number of similar cases handled" and "years of practice" are calculated using existing Min-Max normalization, mapping the results to a range of 0-1. Data based on evaluations, such as "user reputation rating" and "platform star rating," are themselves scores and can be used directly or fine-tuned. Data based on matching degree, such as "professional field fit" and "service region fit," are directly assigned scores based on the degree of matching, such as perfect match = 1, mismatch = 0. Data based on cost, such as "fee budget fit," refers to the "lowest price priority method," where the closer the budget is to or higher than the lawyer's quoted price, the higher the score. Data based on status, such as "lawyer workload," should have a higher score if the workload is lower. The weights for weighted calculations are pre-set and stored, for example, the weight of professional field fit is 20%, the weight of case win probability is 15%, and the weight of user reputation rating is 15%.

[0052] Step 8 involves serializing and breaking down the fulfillment nodes of electronic service orders into standard operation nodes and binding them with execution time limits. For example, different node templates are set for "first-instance civil litigation representation," "contract review," and "lawyer's letter delivery." The node sequence for "first-instance civil litigation representation" is as follows: order acceptance, needs confirmation, case analysis and strategy formulation, finalization of the complaint / response, online case filing / respondence, pre-trial evidence exchange, court hearing, post-trial submission of legal arguments, and judgment / case closure. The deadline for the "finalization of the complaint" node is 48 hours after signing the contract, and the deadline for the "online case filing" node is 24 hours after finalization. The execution status of each node is monitored in real time. An early warning notification is automatically generated when any node's dwell time exceeds the execution time limit. This notification can be sent via SMS to a mobile device. Upon completion of the upload of deliverables, a compliance review is conducted. The compliance review is set according to actual needs, such as format and completeness checks, legal logic and professional risk scanning, and final confirmation of conflicts of interest. Results that pass the review are automatically pushed to the user, and the settlement and payment process is triggered after the user confirms receipt.

[0053] Step 9: Service scores are calculated based on four dimensions: service professionalism, service attitude, processing efficiency, and case completion results. The service score is obtained through a weighted calculation of these four dimensions. The weights are pre-set according to their importance; for example, service professionalism might have a weight of 45%, service attitude 25%, processing efficiency 25%, and case completion results 5%. The actual scores for each dimension are set according to actual needs. The professional level and order allocation weights are dynamically adjusted based on the service score. A higher service score corresponds to a higher professional level and a higher order allocation weight. During adjustments, for every 10-point increase in the service score, the corresponding professional level increases by 0.2-0.4, and the corresponding order allocation weight increases by 0.3-0.5. During the dynamic adjustment process, when service scores are the same, professionals with service scores higher than the set value are selected to determine if they have any pending cases. If so, the end time of these pending cases is estimated. The estimated end time is calculated based on a fixed proportion of the similarity in historical cases, and the professional's service score is increased based on the end time.

[0054] This embodiment's method automates the deconstruction of case materials using a legally-specific NLP module, extracts core elements, and automatically generates legal documents. This transforms traditionally time-consuming and repetitive tasks into an automated process, improving the efficiency of legal text processing and document generation. For example, the generation efficiency of court documents and complaints can be increased by more than 30%. Furthermore, through semantic understanding and logical deduction using knowledge graphs, it accurately identifies legal entities and relationships. Simultaneously, by automatically matching case causes, retrieving the latest regulations, and pushing similar precedents, it provides solid and comprehensive data support for legal decision-making, enhancing the accuracy and comprehensiveness of legal analysis. Through a twelve-dimensional weighted scoring model and a dynamic activity-based algorithm, it overcomes the limitations of traditional static scoring, upgrading from "being able to handle cases" to "precise and efficient case handling," significantly improving the efficiency and quality of matchmaking between lawyers and cases, and enhancing the intelligence and precision of supply and demand matching. Through intelligent compliance review, it automatically scans for legal logic loopholes and professional risks, and stores collaboratively revised hash operation logs on the blockchain, achieving immutability and traceability of operation records. This provides reliable protection for data security and judicial appraisal, ensuring the compliance and security of the case handling process.

[0055] Example 3

[0056] A collaborative legal supply and demand matching implementation method integrating legal NLP differs from Embodiment 2 in that, in step 3, after each retrieval of applicable legal provisions, it automatically determines whether there is a corresponding updated or repealed judicial interpretation. The determination of updated or repealed judicial interpretations is made by directly obtaining official information. If there is an updated or repealed judicial interpretation, the judicial interpretation information of the applicable legal provisions is changed, that is, the corresponding repealed information is deleted and the newly updated information replaces the previous information. Furthermore, based on the current user of the legal service, it is inferred whether to actively push judicial interpretations to them. The level of knowledge includes proficiency, comprehension, and unfamiliarity. When it is proficient, judicial interpretations are not pushed; when it is comprehension, applicable legal provisions and their corresponding judicial interpretations are actively pushed; when it is unfamiliar, judicial interpretations are pushed, and similar precedents are obtained.

[0057] The estimation of the current user's legal information awareness is based on a comprehensive score considering the user's age group, job type, education level, and whether they have engaged in legal activities. A fixed base score is set for each age group, and a corresponding fixed base score is pre-set for each job type. Education level is pre-set with higher education levels receiving higher pre-set scores, and those who have engaged in legal activities receive higher scores than those who have not. The comprehensive score is calculated by summing the scores of each factor, and the degree of legal information awareness is determined based on this comprehensive score. Proficiency, comprehension, and unfamiliarity correspond to different comprehensive score ranges.

[0058] After the applicable legal provisions are retrieved, the frequency of retrieval of the corresponding legal provisions is statistically analyzed. A frequency judgment is performed every preset period. When the frequency is less than the threshold, the threshold is set according to actual needs, and a judgment on the judicial interpretation update or repeal of the legal provision is actively performed. When the frequency is greater than or equal to the threshold, the judgment on the judicial interpretation update or repeal is performed after the legal provision is retrieved.

[0059] By judging the judicial interpretations corresponding to the matched applicable legal provisions and pushing different information based on the user's level of knowledge, accurate information can be pushed without consuming additional system resources during use, ensuring that the user can understand the legal information as much as possible. Furthermore, by comprehensively scoring the user's level of legal knowledge from multiple aspects, the accuracy of determining the level of knowledge is improved.

[0060] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any indirect modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A collaborative legal supply and demand matching implementation method integrating legal NLP, characterized in that, Includes the following steps: Step 1: By receiving legal service requests and case information, obtain the original materials of legal cases, perform image correction and format standardization preprocessing on the original materials, and store the standardized data into the case database. Step 2: Deconstruct the case content using the legal NLP semantic parsing module to extract the core elements of the case, including the parties involved, the facts of the case, the evidence, and the rights and responsibilities. Step 3: Perform vector cosine similarity comparison between the case facts parsed in Step 2 and the case information in the pre-stored case information database, automatically determine the type of the case and recommend three matching case information, and retrieve the applicable legal provisions from the legal database based on the case information and case facts. Step 4: Based on the case subject parsed in Step 2 and the cause of action matched in Step 3, extract the points of contention from the original materials. The points of contention include substantive disputes and procedural disputes. Use hard coding to solidify four fixed weight values ​​to unify and normalize the various indicators of the case into a score in the range of zero to one, so as to sort the points of contention from high to low scores. Step 5: Based on the dynamic document template library that distinguishes between case causes of action, trial location, trial level, and agent status, the template content is automatically filled in by combining the analysis results of Step 2, the applicable legal provisions of Step 3, and the disputed issues sorted in Step 4, and the corresponding trial materials are generated. Step 6: Based on the difficulty of the case and the field of business, intelligently match the corresponding level of practitioners and assign exclusive business operation permissions; Step 7: Based on the legal service needs in Step 1 and the corresponding level of practitioners in Step 6, calculate the comprehensive score through similarity, and match the lawyers with the top preset number of comprehensive scores as matching lawyers to supply and demand, forming an electronic service order; Step 8: Sequence the fulfillment nodes of the electronic service order are decomposed into standard operation nodes and bound with execution time limits. The execution status of each node is monitored in real time. When the dwell time of any node exceeds the execution time limit, an early warning notification is automatically generated. When the upload of deliverables is completed, a compliance review is conducted. Deliverables that pass the review are automatically pushed to the user's end, and the settlement and payment process is triggered after the user confirms receipt.

2. The collaborative legal supply and demand matching implementation method integrating legal NLP as described in claim 1, characterized in that, In step 6, revision information of the court hearing materials generated in step 5 is obtained from at least two users with different permission levels. The permission levels include three levels: lead lawyer, co-lead lawyer, and trainee lawyer. The revision information is recorded in real time at the character-level difference granularity, generating a hash operation log containing fields such as the reviser identifier, operation timestamp, operation IP address, text content before and after revision, and revision type. The hash operation log is then stored in the blockchain evidence storage network.

3. The collaborative legal supply and demand matching implementation method integrating legal NLP as described in claim 1, characterized in that, In step 3, after each retrieval of the applicable legal provisions, it automatically determines whether there is a corresponding updated or repealed judicial interpretation. If there is an updated or repealed judicial interpretation, the judicial interpretation information of the applicable legal provisions is changed, and the current user of the legal service is inferred to determine whether to actively push judicial interpretations. The level of knowledge includes proficiency, understanding, and unfamiliarity. When it is proficient, no judicial interpretation is pushed; when it is understanding, the applicable legal provisions and their corresponding judicial interpretations are actively pushed; when it is unfamiliar, the judicial interpretation is pushed, and similar precedents are obtained.

4. The collaborative legal supply and demand matching implementation method integrating legal NLP as described in claim 1, characterized in that, In step 3, the estimation of the current user's level of legal information knowledge is based on a comprehensive score of the current user's age group, job type, education level, and whether they have engaged in legal activities. The comprehensive score is used to determine the level of legal information knowledge.

5. The collaborative legal supply and demand matching implementation method integrating legal NLP as described in claim 1, characterized in that, In step 3, after the applicable legal provision is retrieved, the corresponding frequency statistics are performed. The frequency is judged once every preset period. When the frequency is less than the threshold, the judicial interpretation update or repeal judgment of the legal provision is actively performed. When the frequency is greater than or equal to the threshold, the judicial interpretation update or repeal judgment is performed after the legal provision is retrieved.

6. The collaborative legal supply and demand matching implementation method integrating legal NLP as described in claim 1, characterized in that, It also includes step 9, which scores services based on service professionalism, service attitude, processing efficiency, and case completion results. The professional level and order allocation weight are dynamically adjusted based on the service score. During the dynamic adjustment process, when the service scores are the same, the service personnel with service scores higher than the set value are selected to determine whether they have any pending cases. If so, the end time of the pending cases is estimated, and the service score of the service personnel is increased based on the end time.

7. A collaborative legal supply and demand matching platform integrating legal NLP, characterized in that, include: The case material collection module is used to collect the original materials of legal cases and preprocess the original materials to obtain unstructured text. The original materials include materials from legal service needs, materials from various interface terminals and data in various formats. The legal NLP semantic parsing module is used to extract core elements of case subjects, case facts, case evidence, and rights and responsibilities from unstructured text; The automatic case cause matching and legal retrieval module is used to compare the parsed case facts with the case cause information in the pre-stored case cause database using vector cosine similarity, automatically determine the type of the case and recommend three types of matching case causes, and retrieve applicable legal provisions from the legal database based on the case cause and case fact elements. The fixed-weight dispute focus sorting module is used to extract the dispute focus in a case and use hard-coded fixed weight values ​​to normalize the various indicators of the case into a score between zero and one, so as to sort the dispute focus according to the score from high to low. The multi-role court hearing document intelligent generation module is used to build a dynamic document template library that distinguishes the cause of action, trial location, trial level, and agent status. It automatically fills in the template content based on NLP analysis results, points of contention, and applicable legal provisions to generate court hearing documents for the corresponding role. The lawyer hierarchical access control module is used to classify practitioners into four levels based on the RBAC role-based access control architecture: specially appointed senior legal experts, star-rated practicing lawyers, ordinary practicing lawyers, and on-site legal specialists. The collaborative revision full-process traceability module is used to record account information, operation time, device IP address, and content before and after modification in real time, and automatically generate an independent document version snapshot for each content adjustment; The intelligent compliance review module for case handling content is used to identify document problems through a triple verification mode of keyword retrieval, text semantic comparison, and legal logic in a compliance review feature library with pre-set legal service violation semantics. The Level 3 Case Data Security Protection Module is used to divide case information into three storage levels according to the sensitivity of the information; The cross-system forced data synchronization module is used to perform incremental real-time synchronization with a preset duration, combined with a full data verification operation mode at a preset time every day. It uses a private encryption protocol to transmit data and a unified standard message format. The lawyer qualification management module is used to connect to the official practice information verification interface to conduct real-name authentication of the identity of the lawyer, verification of the authenticity of the practice certificate, and filing of practice information. The legal service demand collection module is used to collect business demands, extract information such as case type, service area, dispute claims, service budget, urgency of handling, and lawyer selection preferences, generate 20-dimensional structured demand tags, and uniformly archive and store them in a preset demand database. The supply and demand two-way matching module is used to calculate a comprehensive score through twelve weighted scoring features. The weighted scoring features include professional field fit, years of practice, number of similar cases handled, case win probability, user reputation rating, service area matching, order response speed, cost budget fit, case urgency, lawyer workload, platform star rating, and historical cooperation evaluation. Lawyers with the highest comprehensive scores among the top preset number are matched with lawyers. The full-scenario online legal service integration module is used to integrate all business portals such as online legal consultation, legal document drafting, contract review and revision, case process handling, corporate legal counsel, dispute mediation, and court appearance assistance. It opens service access to the outside world through multiple channels such as computer web pages, mobile mini programs, and mobile APPs, and all business operations are recorded and archived online. The service process visualization and control module is used to retain text and voice communication records, business performance vouchers, case handling nodes, and progress change snapshots; The standardized order fulfillment module is used to automatically generate electronic service orders based on the supply and demand matching results. Both parties confirm the order's validity online with electronic signatures, and the business nodes are verified and accepted online after completion. The service rating and industry order management module is used to score services based on service professionalism, service attitude, processing efficiency, and case completion results, and dynamically adjust the professional level and order allocation weight based on the service rating.

8. A collaborative legal supply and demand matching platform integrating legal NLP as described in claim 7, characterized in that, The four fixed weight values ​​include the weight value of the amount involved, the weight value of the fault liability division, the weight value of the judgment tendency of similar cases, and the weight value of the priority of court debate, which are respectively set to 0.25, 0.3, 0.25, and 0.

2.

9. A collaborative legal supply and demand matching platform integrating legal NLP as described in claim 1, characterized in that, It also includes a quantum dual-mode algorithm scheduling module, which is natively adapted to the classical computer operation mode, while standardizing and reserving extended interfaces for quantum parallel operation, quantum semantic retrieval, quantum key encryption, and quantum feature matching, and can be seamlessly connected to quantum computing hardware devices to upgrade the system's computing power.

Images