Whole-process judgment auxiliary legal semantic system supporting fact semantic extraction and dispute focusing

Abstract

The invention discloses a whole-process judgment auxiliary legal semantic system supporting fact semantic extraction and dispute focusing. The system is constructed based on a DIKWP mesh cognitive model and comprises a data acquisition and preprocessing module, an information extraction and semantic association module, an intelligent analysis and dispute focus recognition module, a judgment intention association and priority ranking module, a white-box reasoning chain recording and visualization module and a Bug detection and consistency verification module. The system realizes layer-by-layer abstraction and feedback optimization from original data to judgment intentions by performing multi-level semantic analysis on case materials, and can accurately identify case dispute focuses. The method is characterized in that a consciousness relativistic mechanism is introduced to realize multi-role semantic alignment, a white-box interpretable inference chain tracing function is realized, and semantic contradictions can be automatically detected to realize logic self-correction. According to the method, the depth of factual understanding of complex cases and the accuracy of dispute focus extraction are effectively improved, and the method can be widely applied to the judgment auxiliary whole process of civil and criminal cases and the like.

Description

Technical Field

[0001] This invention relates to the fields of judicial artificial intelligence and natural language processing technology, and in particular to a legal semantic analysis system for assisting in the entire process of adjudication. Specifically, this invention is a legal semantic system based on a multi-layered cognitive semantic model, capable of deep semantic understanding of case materials, accurate extraction of factual elements, intelligent identification and focusing of points of contention, and possessing white-box interpretability and logical self-verification capabilities. Background Technology

[0002] With the development of smart court construction, various regions have explored AI-assisted judicial adjudication systems, attempting to cover the entire process from case filing and trial to judgment document generation. Existing intelligent adjudication assistance systems typically focus on case element analysis, achieving functions such as fact analysis, dispute identification, and legal application. However, in cases with complex facts and intertwined legal relationships, existing systems struggle to accurately understand and intelligently extract the facts. The main bottleneck lies in the highly diverse and context-dependent nature of facts described in natural language; traditional methods based on keywords or element extraction struggle to accurately grasp key points and understand deeper meanings. For example, regarding the circumstance of "voluntary surrender," a simple statement is easily identified as self-surrender, but if described as "accompanied by parents to the criminal investigation team" or "ran to the police station after being scolded by his sister," traditional systems may overlook this important sentencing element. Similarly, in robbery cases, different statements may involve implicit semantics such as "purpose of entry," and existing systems often only extract generalized charge labels, failing to identify more subtle differences in details. These limitations lead to inaccurate extraction of disputed points, increasing the burden on judges in analyzing the case.

[0003] Furthermore, different judicial roles (judges, lawyers, and parties involved) may have differing understandings of the same factual statement. Information asymmetry and contextual differences often lead to cognitive biases and misunderstandings. Existing technologies lack mechanisms to identify and eliminate such semantic misunderstandings, often failing to effectively align the statements of all parties, resulting in deviations in the positioning of the focus of the dispute. Simultaneously, AI black-box models lack interpretability in judicial contexts, making it impossible for judges to trace the factual analysis process provided by the AI, reducing the credibility of the system's suggestions. Especially in cases of conflicting evidence or logical inconsistencies, existing systems often fail to automatically detect and alert to these "bugs," requiring manual intervention and impacting trial efficiency. Therefore, a new type of legal semantic analysis system is urgently needed, capable of multi-layered analysis of case materials based on a more advanced cognitive semantic model, progressively abstracting from the data layer to the intent layer, achieving a deep understanding of the facts and accurate extraction of the focus of the dispute. Simultaneously, the system should possess a white-box visualization function of the reasoning chain, capable of self-detecting semantic contradictions ("bugs") and performing logical consistency checks. The relative differences in awareness among different judicial participants should also be considered, aligning the parties' understanding of the semantics of the facts through mechanisms to reduce misunderstandings and achieve semantic uniformity. Summary of the Invention

[0004] This invention provides a legal semantic system module that supports factual semantic extraction and dispute focus. Based on the DIKWP (Data-Information-Knowledge-Intelligence-Intent) network cognitive model, it performs hierarchical semantic analysis of case materials, solving problems such as difficulty in understanding complex case facts, inaccurate extraction of disputed points, and inexplicable reasoning in existing technologies. The system extracts information from raw data, integrates knowledge and intelligent judgment, and ultimately constructs a semantic path guided by the judge's intent, enabling AI to "read" case materials and identify the core disputes between the parties. Compared with traditional methods, this invention has the following beneficial effects:

[0005] Full Semantic Hierarchical Understanding: Employing the DIKWP five-layer semantic model, the process of acquiring case facts is divided into multiple layers: data, information, knowledge, wisdom, and intent, forming a network-like multi-layered cognitive system that enables bidirectional feedback and iteration of semantics. For example, key information is extracted from original evidence data, then elevated to the legal knowledge layer for associative reasoning, followed by determining which facts are disputed at the wisdom layer, and finally, at the intent layer, selecting key facts based on the judicial objectives. This hierarchical structure ensures that the system's understanding of case facts goes beyond surface text matching, incorporating legal common sense and judicial purposes, thus enhancing its ability to process complex semantics.

[0006] Precise Extraction of Points of Contention: The system compares the similarities and differences in the facts presented by each party and identifies points of contention through a semantic alignment algorithm. For example, differences between the facts asserted by the plaintiff and those admitted by the defendant will be automatically extracted as points of contention and clearly identified. Utilizing the theory of relativistic consciousness, the system acknowledges the differences in understanding of the same information by different cognitive subjects; therefore, this invention introduces a relative consciousness alignment mechanism to identify and measure differences and misunderstandings in the statements of each party in the semantic space. For detected semantic inconsistencies, the system resolves misunderstandings by supplementing context and unifying terminology definitions, ensuring that the extraction of points of contention is based on a semantic consensus among all parties.

[0007] White-box Explainability and Bug Verification: This invention incorporates a white-box reasoning chain tracing module, capable of recording and visualizing the entire process of case fact extraction and dispute identification. The reasoning path behind each conclusion (such as a fact being determined to be a point of contention) is traceable, including the evidence involved, the rules adopted, and semantic connections. This allows judges to examine the AI's reasoning chain to ensure it conforms to legal logic. Simultaneously, the system introduces an error detection mechanism guided by the theory of conscious "bugs": it automatically identifies contradictions or omissions during multi-layered semantic parsing, treating them as cognitive bugs and promptly marking them. For example, if evidence A contradicts evidence B on a key fact, the system will mark it as a potential bug and request review from a higher level (the intelligence / intent layer) or human confirmation through a metacognitive loop, thereby correcting the internal inconsistency. This mechanism is equivalent to integrating a "debugging" function within the AI, enabling the system to self-correct and ensuring logical consistency in its output.

[0008] Multi-role Semantic Alignment and Interpretation: Considering the different focuses of judges, lawyers, and parties in understanding the facts, this invention provides a relative consciousness expression and alignment mechanism. When extracting points of contention, the system considers the legal facts relevant to the judge, the facts favorable to the lawyer's side, and the facts directly relevant to the party, establishing corresponding weights or perspectives for each role in the semantic model. By introducing human feedback calibration at the intelligence and intent layers, the system can adaptively adjust its framework for understanding the facts. The final output list of points of contention includes multi-perspective explanations, such as annotations indicating "judge's focus" and "party's questions," to help all parties reach a consensus on the system's analysis results and improve acceptance. This multi-subject alignment mechanism originates from the concept in the theory of consciousness relativity that "every consciousness has its own coordinate system," eliminating biases caused by the relativity of consciousness by emphasizing alignment.

[0009] In summary, this invention integrates the deep semantic understanding capabilities of artificial intelligence with the needs of judicial practice through a combined hardware and software approach. On the software side, it integrates the DIKWP multi-layer semantic parsing algorithm, a knowledge graph reasoning engine, and a white-box visualization interface. On the hardware side, it can be deployed on a court's dedicated network server or a judicial AI workstation equipped with an Artificial Consciousness Processing Unit (ACPU) to ensure data security and processing efficiency. This invention is expected to significantly improve the efficiency and accuracy of courts in analyzing complex case facts, laying a high-quality foundation for subsequent legal application and judgment generation, and possesses promising industrial application prospects and promotional value.

[0010] To achieve the above objectives, this invention provides a fact semantic extraction and dispute focusing system based on the DIKWP semantic mesh model, the structure of which includes, but is not limited to, the following components:

[0011] 1. Data Acquisition and Preprocessing Module: This module receives and imports multi-source case materials, including indictments, answers, witness testimonies, and scanned copies of evidence. It performs format conversion, noise filtering, and structured preprocessing on the raw data, transforming unstructured text and images into parsable digital information. For example, it performs OCR recognition on scanned paper case files, transcribes speech transcripts into text, and categorizes and labels the data according to case roles (judge, plaintiff, defendant, witness) for subsequent semantic analysis.

[0012] 2. Information Extraction and Semantic Association Module: This module corresponds to the information and knowledge layers of the DIKWP model. First, using natural language processing technology and a legal knowledge base, it extracts basic factual elements of the case (such as people, time, place, behavior, and outcome) from the preprocessed data, forming a structured information set. Then, based on a legal knowledge graph and rule base, it semantically associates the extracted information with relevant legal concepts and regulations, inferring implicit relationships and background knowledge. For example, it associates the fact of a "loan contract" with the concept of a loan relationship in contract law, and matches the behavior of "breaking into a house at night" with the aggravating circumstances of robbery in criminal law. This module also uses a probabilistic verification mechanism to evaluate the degree of matching between the extracted information and legal knowledge, ensuring the accuracy of semantic association. When ambiguity or uncertain associations occur, the module marks the content requiring further judgment for subsequent processing.

[0013] 3. Intelligent Analysis and Dispute Focus Identification Module: Corresponding to the intelligent layer function of the DIKWP model. This module integrates the aforementioned structured information and knowledge associations, uses reasoning algorithms to analyze the factual context of the case, uncovers differences and contradictions between the parties' statements and evidentiary materials, and identifies the dispute focus. Specifically, the system constructs a semantic graph of the case facts, linking and comparing the descriptions of the same events by all parties. If the plaintiff and defendant's descriptions of an event are inconsistent, or if the evidence and statements contradict each other, the event is identified as a potential dispute focus, and a corresponding problem description is generated. For example, if the system detects that the plaintiff claims "the defendant has not repaid the loan," while the defendant claims "partial repayment," the system extracts "whether the defendant has repaid part of the loan" as the disputed issue. A misunderstanding detection mechanism based on understanding theory is introduced during the identification process: the system calculates the similarity and differences between the parties' statements in the semantic space. If it finds that a key term or concept is used inconsistently by both parties (e.g., different understandings of the word "profit"), it is considered a potential misunderstanding. The system will automatically query the legal terminology database or case database to clarify the definition, or prompt the judge to investigate this ambiguity, thereby improving the accuracy and fairness of dispute focus identification.

[0014] 4. Judicial Intent Association and Priority Ranking Module: Corresponding to the intent layer function of the DIKWP model. In judicial scenarios, "intent" is reflected in the judgment's objectives and legal value considerations, such as resolving disputes, fairness and justice, and the balance between punishment and education. This module associates and prioritizes multiple identified points of contention based on the judgment's intent. For example, in civil cases, the judgment's intent may focus on clarifying the rights and obligations of the parties; therefore, disputes directly related to liability determination (such as whether the contract is valid or whether there is a breach of contract) have higher priority. In criminal cases, disputes directly related to conviction and sentencing (such as whether the conduct constitutes a crime or whether there are aggravating circumstances) have higher priority. The module assesses and scores the importance of the points of contention through pre-set judicial policies and value criteria, selecting issues that require focused adjudication. This process also applies the human-machine collaborative closed-loop concept guided by the theory of consciousness relativity, meaning the system can accept feedback from judges to adjust the priority of points of contention, ensuring that the ranking results conform to human trial experience and the actual needs of the case. For example, judges can view the preliminary list of points of contention and manually adjust the weights; the system then recalculates the ranking to achieve human-machine collaborative intent alignment.

[0015] 5. White-box Reasoning Chain Recording and Visualization Module: Throughout the aforementioned analysis stages, this module records the system's internal reasoning process and presents it to the user in a graphical interface. For each extracted factual information and each identified point of contention, the system provides a traceable explanation of its source and the basis for its reasoning. For example, regarding the point of contention "whether the defendant has repaid part of the loan," the system interface can display: the plaintiff's statement, the defendant's statement, the comparison results of relevant evidence (such as bank statements), and the system's reasons for determining the dispute. The reasoning chain is presented in the form of a flowchart or knowledge graph, with nodes representing facts or evidence, lines representing reasoning or related operations, and the rules or algorithms used labeled. Judges can gradually unfold the process to understand the details of each step of the system's "thinking" to verify its rationality. This interpretable design responds to the judiciary's requirement for transparency in the AI ​​decision-making process, making AI a true "white-box" assistant to judges, rather than an unquestionable black box.

[0016] 6. Bug Detection and Consistency Verification Module: An embedded monitoring unit within the system workflow, this module captures anomalies or contradictions in the analysis process in real time. When semantic conflicts, logical loops, or missing evidence occur, this module marks them as awareness "bugs" and initiates corresponding processing strategies. For example, if two pieces of evidence from different sources inconsistently describe the same key fact, the system will highlight the conflict with a warning marker in the reasoning chain and add a "Contradictory Evidence Requires Review" prompt to the list of disputed points. If a disputed point is found to be unrelated to any legal element (indicating it may be irrelevant or a pseudo-dispute), the system will indicate a possible analytical bias, suggesting the judge ignore it or allow the system to relearn and adjust parameters. The bug detection module also collects this conflict information as a data source for continuous optimization of the AI ​​model. As the awareness bug theory reveals: without bugs, there is no need for higher levels of awareness; conversely, one of the important roles of awareness is to discover and correct these bugs. By introducing a "self-debugging" mechanism into the algorithm, the system significantly improves the robustness of the analysis and the reliability of the conclusions.

[0017] The modules of this invention can be deployed on general-purpose computing devices via software or accelerated by dedicated hardware. Preferably, the system is deployed on a local area network server within the court, and secure encryption is used to ensure the confidentiality of case data. Where feasible, a dedicated AI chip (such as the ACPU specific to the DIKWP model) can be used to improve the efficiency of semantic reasoning and knowledge retrieval. The entire system is designed with a modular architecture, making it easy to expand or tailor to different application scenarios. For example, on a mediation platform, only the fact extraction and dispute identification modules can be deployed to assist mediators in quickly grasping the key points of the dispute; on a third-party legal technology platform, the modules of this invention can be integrated with user consultation needs to provide lawyers with case key point extraction services. Through this scalable architecture combining software and hardware, this invention can be flexibly applied to court trial operations, online dispute resolution, and legal consultation scenarios, and has broad industrialization prospects. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the overall architecture of the fact semantic extraction and dispute focusing system.

[0019] Figure 2 This is a flowchart illustrating the system's process for fact extraction and dispute identification in civil cases.

[0020] Figure 3 A schematic diagram of a visual example interface for the points of contention generated by the system. Detailed Implementation

[0021] Example 1: Semantic Extraction of Facts and Identification of Points of Contention in Civil Lending Disputes. In a civil lending dispute case, Plaintiff A claims that Defendant B has failed to repay a loan, while B argues that he has partially repaid the loan and objects to the interest calculation. After inputting the relevant materials of the case (including the complaint, answer, IOU, bank statements, etc.) into the system of this invention, the system operates according to the following steps:

[0022] 1. Data Layer Processing: The system uses a data acquisition and preprocessing module to convert scanned loan agreements and bank statement images into text data, and imports the complaint and answer into the text analysis engine. The imported text undergoes word segmentation and entity recognition to extract basic data such as loan amount, loan date, and repayment records. At this stage, the system detects that both the plaintiff's and defendant's texts contain keywords such as "loan" and "repayment," initially identifying the case as a loan contract dispute.

[0023] 2. Information Layer Extraction: The system further extracts structured information: The plaintiff claims "a loan of 1 million yuan was made, with an agreed annual interest rate of 10%, a loan term of one year, and the loan has matured but not been repaid"; the defendant claims "500,000 yuan of principal has been repaid, but the interest is too high and requests an adjustment." The system uses a legal knowledge base to map this information to standard legal elements: contract principal, interest rate, term, repayment status, etc. Simultaneously, based on the built-in contract law knowledge graph, it identifies relevant legal concepts such as "the agreed interest rate is protected by law at its maximum limit" and "overdue interest," and associates them with case information. Probability confirmation calculations show that the plaintiff and defendant's statements are completely consistent on the principal amount (100% match), differ on the interest rate (the defendant believes the interest rate is too high, implying it may exceed the legal limit, approximately 60% match), and are significantly inconsistent on the repayment facts (the plaintiff claims no repayment vs. the defendant claims 500,000 yuan has been repaid, 0% match). Based on this, the system marks "whether part of the loan has been repaid" and "whether the interest is legal" as potential points of contention.

[0024] 3. Knowledge-Level Reasoning: Entering the intelligent analysis module, the system conducts in-depth reasoning based on all case information. Regarding the repayment dispute, the system retrieves the bank statements provided by the defendant, finding a transfer record of 500,000 yuan, and matching the time to the day after the loan's due date, thus inferring that the defendant did indeed make partial repayments. This newly inferred fact is temporarily stored as a "candidate fact." The system also retrieves legal provisions, confirming that courts will not support annual interest rates for private loans exceeding 24% (the upper limit stipulated by judicial interpretation). Comparing this to the agreed annual interest rate of 10%, it does not exceed the limit and requires no adjustment. The system applies this knowledge to the analysis, inferring that the interest rate objection may lack legal basis. Through these reasonings, the system updates the list of disputed issues: retaining "whether the 500,000 yuan loan has been repaid" as the focus, while weakening the priority of "whether the interest rate is too high," because according to knowledge-based reasoning, this dispute has little impact on the judgment outcome.

[0025] 4. Intent-Level Calibration: The system invokes the adjudication intent association module to calibrate and prioritize the current points of contention based on the civil trial's goal of debt repayment and dispute resolution. Since "whether the loan has been repaid" directly relates to the defendant's repayment responsibility and is crucial to the case's ruling, it is given the highest priority. The interest rate dispute, determined by the knowledge layer, is not illegal, and even adjustments would have little impact on the principal repayment obligation; therefore, its priority is lowered. The judge reviews the system's suggested priorities, confirms they align with their case-handling experience, and accepts this prioritization. If the judge has a different judgment (e.g., believing the interest rate issue involves the clarification of judicial interest calculations), they can manually adjust the priority of the priorities. The system will record this human feedback for further optimization of the intent-level decision-making model.

[0026] 5. Output and Explanation of Points of Contention: Ultimately, the system outputs the main point of contention: "Whether the defendant has repaid part of the loan (500,000 yuan)." A white-box explanation report is also generated: listing the differences between the plaintiff's and defendant's statements on this issue, citing evidence (bank statements) to support the defendant's claim of repayment, and explaining that this matter becomes a point of contention due to the inconsistency in the statements and the need for judicial verification of the evidence. Regarding the issue of "excessive interest rate," which was downgraded by the system, the report states: after searching legal provisions, an annual interest rate of 10% is legal; therefore, this issue has a relatively small impact on the judgment and is not currently listed as a major point of contention. The entire report is presented in chart form, clearly indicating the process nodes of fact extraction, legal connections, and point of contention identification, allowing both the judge and both parties to intuitively understand how the system arrived at the above conclusions.

[0027] This embodiment demonstrates the process of accurately extracting facts and focusing on disputes in complex civil cases. The system effectively alleviates the burden on judges manually reviewing case details, improving efficiency. Actual testing showed that parties involved approved of the system-generated list of disputed issues and its explanations, believing that their concerns were fully covered and the reasons were clear. This proves the practicality and reliability of this invention in real-world applications.

[0028] Example 2: Semantic Analysis of Facts in a Criminal Case. In a case of burglary and robbery, the suspect claimed that his initial intention was not robbery but theft, and that he only resorted to violence after being discovered. Applying the system of this invention to this case can quickly clarify the focus of the factual dispute. For example, the system extracts from the indictment and confession: "The suspect entered the victim's residence at night with the intention of stealing, and after being discovered by the victim, he violently robbed the victim of his property." Simultaneously, the system learns from the criminal law knowledge base that this situation falls under the category of burglary converted into robbery. The suspect's confession may emphasize "it was just a casual theft," attempting to deny the intent to rob. Through semantic comparison, the system finds a conflict between the legal characterization and the suspect's self-defense, thus identifying "whether the criminal intent was robbery" as the focus of the dispute. In the output interpretation, the system cites relevant articles of the criminal law and similar cases to clarify that regardless of the initial purpose, as long as violence is used to seize property, it constitutes robbery (judicial interpretation regarding converted robbery). Such semantic analysis and interpretation assist the judge in determining that the suspect's defense is untenable. This embodiment demonstrates that the present invention is also applicable to criminal cases, helping judges to grasp the essential disputes from complex testimonies and improve the accuracy of conviction and sentencing.

Claims

1. A full-process adjudication-assisted legal semantic system supporting factual semantic extraction and dispute focus, comprising a data acquisition and preprocessing module, an information extraction and semantic association module, an intelligent analysis and dispute focus identification module, a adjudication intent association and priority ranking module, a white-box reasoning chain recording and visualization module, and a bug detection and consistency verification module, characterized in that: The system constructs a semantic parsing path based on a five-layer semantic model of "Data-Information-Knowledge-Wisdom-Intent" (DIKWP). It cleans and structures the original case materials at the data layer, extracts factual elements and associates them with legal knowledge at the information layer, identifies the points of contention between the parties at the wisdom layer, and prioritizes the points of contention in conjunction with the adjudication objectives at the intent layer. The semantic results at each layer are iteratively optimized in a network feedback manner, and finally, an interpretable list of points of contention is output.

2. The system according to claim 1, characterized in that: The dispute focus identification module uses the principle of consciousness relativity to semantically align the statements of different judicial roles, identify and measure semantic differences and misunderstandings among the parties, and eliminate semantic biases through semantic association and terminology unification, thereby accurately extracting the dispute focus between the parties.

3. The system according to claim 1, characterized in that: The white-box reasoning chain recording and visualization module is used to record the reasoning basis of the system in the process of fact extraction and dispute identification, and generate a visualized semantic chain; each point of contention is accompanied by a traceable explanation of the evidence, legal knowledge and reasoning steps on which it was generated, so as to improve the transparency and understandability of the system's decision-making.

4. The system according to claim 1, characterized in that: The Bug Detection and Consistency Verification module monitors the analysis results at each layer of semantic parsing. When evidence conflicts, logical inconsistencies, or abnormal deviations occur, they are marked as "Bugs" and an error correction mechanism is triggered. This error correction mechanism includes sending feedback to higher-level semantic models to adjust the parsing results or reminding manual intervention to ensure that the final output list of points of contention is logically consistent and without contradictions.

5. The system according to claim 1, characterized in that: The system is deployed in the court's internal network environment or a securely certified third-party legal service platform, and is implemented through a combination of hardware and software. The software component includes a natural language processing engine, a legal knowledge graph, and reasoning algorithms, while the hardware component preferably uses an artificial intelligence processing unit that supports semantic computing acceleration to ensure case data security and system operating efficiency.

Images