Intelligent judicial decision support system based on semantic sovereignty

Abstract

The invention provides a judicial intelligent decision support system based on semantic sovereignty. The judicial intelligent decision support system comprises an ontology law semantic library, a DIKWP judicial reasoning engine and a trial intention management module. According to the system, semantic annotation and knowledge graph construction are carried out on national legal provisions and case knowledge through an ontology legal semantic library, and local semantics of legal concepts are reserved; performing multi-level semantic reasoning on the case by using a judicial reasoning engine comprising five-layer models of data, information, knowledge, wisdom and intention, performing matching analysis on case facts and legal regulations, and generating a decision suggestion with semantic interpretation; and the judgment intention management module allows a judge to input intention parameters such as policy guidance to calibrate the reasoning process of the AI. The system records a complete reasoning semantic chain to ensure that the decision making process is transparent and explainable, the controllability, fairness and credibility of judicial AI aid decision making are remarkably improved, and the system can be widely applied to judicial intelligence fields such as court trial assistance and legislation draft evaluation.

Description

TECHNICAL FIELD

[0001] The present application relates to an intelligent decision support system in the field of justice, in particular to an artificial intelligence system integrating a semantic judicial reasoning framework and a local legal semantic system. Specifically, the present application belongs to the technical field of legal artificial intelligence and knowledge engineering, aiming to provide an intelligent auxiliary decision tool for court trials, legislative evaluation, etc. BACKGROUND

[0002] With the increasing application of artificial intelligence in the legal field, various judicial auxiliary decision systems have emerged. However, most existing judicial AI systems are often based on general deep learning models or large open-source language models from abroad. The internal logic and reasoning process of these models are usually opaque, exhibiting "black box" characteristics, making it difficult for judges and parties to understand the suggestions and decisions given by AI, and may be biased. Compared with blindly pursuing algorithm accuracy, the transparency and explainability of judicial AI are more critical.

[0003] In addition, there are significant differences in legal systems and language expressions among different jurisdictions. If AI models trained from foreign legal corpus are directly used to handle domestic cases, semantic bias or misjudgment may occur. This semantic incompatibility not only affects the correct understanding of AI on domestic legal provisions, but also may introduce tendencies in the decision-making suggestions that are inconsistent with the spirit of domestic law, weakening judicial independence. In the context of increasing cross-border legal AI tools, how to ensure that the understanding of legal provisions by our country's judicial AI is consistent with that of domestic judges and avoid being subject to the bias of foreign models has become a problem to be solved.

[0004] In view of this, the concept of "semantic sovereignty" is proposed in the field, which is to embed the values and semantic system of domestic law in AI systems, so that the unique semantics of domestic legal language are fully retained in the process of legal reasoning. The technical basis of semantic sovereignty is the DIKWP semantic model, in which an "intention" layer is added to the top layer of the traditional data-information-knowledge-wisdom (DIKW) model, forming a five-layer cognitive architecture of data (Data) - information (Information) - knowledge (Knowledge) - wisdom (Wisdom) - purpose (Purpose). By introducing this model into AI, the reasoning process of AI can be integrated with the value orientation of domestic justice, and each step of conclusion can be traced and explained. This lays a foundation for building judicial AI with semantic sovereignty and provides a new idea for improving the credibility and fairness of judicial AI. SUMMARY

[0005] The purpose of the present application is to overcome the semantic deviation and unexplainable defects of the existing judicial AI in the prior art, and provide a judicial intelligent decision support system based on semantic sovereignty. The system comprehensively retains the semantic connotation of the legal language of the country in design, ensures that the understanding of AI to the legal provisions is consistent with the judges of the country, and is not interfered by the foreign legal model bias, so that the sovereignty of judicial AI and the semantic justice are unified.

[0006] In order to achieve the above purpose, the present application provides an intelligent decision support system for the judicial field, which framework integrates the local legal semantic system and semantic judicial reasoning technology. The system includes the following three subsystems:

[0007] Ontological legal semantic library: the data of current legal provisions, judicial cases and the like are semantically annotated and stored in structure, and the ontology model and knowledge graph of legal concepts and their associated relationships are constructed. The semantic library embeds the value consideration specific to Chinese legal culture (for example, the semantic representation reflecting the balance between reason and law), ensuring that the semantic meaning of the legal concept is consistent with the legal tradition of the country. Through the ontological legal semantic library, AI can retain the semantic details of the country when analyzing legal texts, and will not lose key information due to translation deviation or foreign model bias.

[0008] DIKWP judicial reasoning engine: DIKWP refers to a five-layer semantic model of data-information-knowledge-wisdom-intention. The judicial reasoning engine in the present application performs multi-level semantic reasoning on cases based on the five-layer semantic model. The engine constructs a case semantic representation for the input case elements: maps objective legal provisions and evidence materials into a content graph, maps the subjective factors such as the interests of the parties and the background of the case into a stakeholder graph, and then comprehensively reasons and decides in a unified semantic space. The engine has rich semantic reasoning algorithms built in, such as matching the intention layer to realize case judgment (matching the current case with the judgment intention of similar past cases to ensure the same case and the same judgment), and finding the implicit legal principles through the knowledge layer association reasoning (mining the legal principles implied behind the legal provisions to apply to new problems). With the help of the DIKWP engine, multi-source and heterogeneous case information is uniformly represented and processed at the semantic level, so that AI can penetrate the layers of "legal provisions-facts-judgment intention", and give decision suggestions that meet the requirements of legal semantics and judicial justice.

[0009] Trial intent management module: provides a human-computer interaction interface, allowing judges or legislative experts to impose high-level guidance on the AI decision-making process. Through this module, users can input policy guidance, judicial interpretation or specific value inclination, etc. Judgment intention parameters, so that AI fully considers these human intentions when making decision reasoning. For example, for new cases, the judge can set the "priority to maintain national interests" guiding principle, and the AI will reflect the inclination to protect national interests in the decision-making recommendation; for example, for juvenile criminal cases, the policy intention of "education first, punishment second" can be input, and the AI will adjust the reasoning path accordingly to reflect the spirit of combining punishment with leniency. Through the trial intent management module, the invention ensures that the AI decision-making recommendation is not only based on law and evidence, but also dynamically incorporates current judicial policy and value considerations, truly achieving intelligent trial under human-machine collaboration.

[0010] In summary, the system constructed by the invention first utilizes the ontological legal semantic library to understand legal knowledge and case data at the semantic level; secondly, the DIKWP judicial reasoning engine performs multi-layer semantic matching reasoning between case facts and legal provisions; finally, the trial intent management module introduces human judicial intent to calibrate and optimize the AI decision-making results. The key innovation lies in the fact that all reasoning processes of the system record detailed semantic traces: the conclusions obtained at each step of reasoning can be traced back to the corresponding legal semantic basis and value considerations, forming a transparent white-boxed judgment logic chain. This design greatly improves the explainability and credibility of AI decision-making, ensuring that the process and results of AI-assisted adjudication comply with the legal system and the principle of judicial fairness of the country.

[0011] Compared with the prior art, the invention has the following beneficial effects:

[0012] Maintain legal semantic sovereignty: the system is developed based on the legal semantic model of the country, fully retaining the meaning and value inclination of Chinese legal language, avoiding the introduction of biases or misunderstandings that may be introduced by foreign models. Through the realization of semantic sovereignty, the understanding of AI for legal provisions is highly consistent with that of judges in the country, ensuring that the algorithm does not deviate from the legal spirit of China in case judgment and legal analysis, realizing the autonomous control of judicial AI.

[0013] Improve the transparency of judicial intelligence: the system records the semantic traces and judgment basis in the AI reasoning process, so that each step can be explained and reviewed. This white-boxed decision chain enhances the trust of judges and parties in AI recommendations, and also facilitates future auditing and supervision of the AI judgment process, further consolidating judicial credibility.

[0014] Efficiency and decision support: In complex and difficult cases, judges can use the system to obtain multi-angle intelligent analysis and suggestions, including similar case reference, law application analysis and possible decision result prediction. The system's suggestions fully consider legal basis and reasonableness, allowing judges to understand different aspects of the case in a short time, greatly improving the efficiency of the trial.

[0015] Legislation and judicial training: For legislators, the system can simulate the semantic application effect of new legal draft in judicial practice, and find possible ambiguities or conflicts in legal provisions in advance, providing basis for perfecting legislation. For judicial practice and training, the system can be used as a teaching tool to help judges and lawyers understand how AI reasons based on legal semantics, and promote the research and discussion of complex legal semantic problems.

[0016] Maintain judicial independence and security: In the trend of developing legal AI in countries around the world, the invention provides an intelligent judicial platform centered on China's legal semantic system, reflecting the awareness of digital sovereignty in the legal field. As a self-contained technical base, the system avoids excessive dependence on foreign AI technology and prevents the improper influence of foreign models, providing a safe and reliable localization solution for China's smart justice construction. This helps to ensure the independence of the country's legal system in the digital age and ensure that the exercise of judicial power in the AI era is in line with the country's sovereign interests. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 The structure diagram of the judicial intelligent decision support system based on semantic sovereignty of the invention. DETAILED DESCRIPTION

[0018] The technical solutions of the invention will be further described below in conjunction with the drawings and examples. It should be noted that these examples are intended to help understand the principles of the invention and are not intended to limit the scope of protection of the invention.

[0019] Example 1: System architecture and workflow

[0020] As Figure 1As shown, the judicial intelligent decision support system of the present application is composed of three parts: an ontological legal semantic library, a DIKWP judicial reasoning engine and a trial intent management module. Each part works through a shared semantic data interface. When the system is running, first, the information related to the case is input into the ontological legal semantic library for processing, including the applicable legal text, case fact materials (such as evidence list, case statement) and the background of the parties and other information. The semantic library uses the pre-constructed legal knowledge graph to perform semantic annotation and correlation expansion on the input content: for example, it identifies the legal concepts, charges and statutory circumstances involved in the case, indexes the characters, events, time and other elements, and retrieves the semantic correlation of related cases from the existing case library. After processing by the semantic library, the key elements of the case are structured into semantic data, laying the foundation for subsequent reasoning.

[0021] Next, the DIKWP judicial reasoning engine starts working. The engine first constructs the semantic representation of the case under the DIKWP five-layer model according to the structured data output by the semantic library: in the data layer, it obtains the original evidence and legal text data; in the information layer, it extracts the key information of the case (such as the time of the incident, the amount involved, the evidence proof, etc.); in the knowledge layer, it introduces legal knowledge and precedent rules to establish a mapping relationship between the case facts and relevant legal provisions; in the wisdom layer, it combines judicial experience and fairness principles to make a comprehensive evaluation of the case; in the intent layer, it considers the adjudication target and value orientation to make an overall assessment of the case and choose the direction of the ruling. Through the process of semantic analysis of five layers of gradual abstraction and feedback iteration, the engine can simulate a reasoning path close to the thinking of human judges.

[0022] In the process of operation of the above-mentioned five-layer semantic model, the DIKWP engine constructs two types of semantic graphs for reasoning: one is the content graph, mainly containing the objective facts and legal norms involved in the case, such as the correspondence between the evidence chain and the relevant legal provisions; the other is the stakeholder graph, containing the subjective intent-related information of the parties' claims, social influence factors and policy considerations. The engine deduces these two graphs in a unified semantic space: by comparing the matching degree of the case facts and the legal provisions in the content graph, it determines which legal provisions are applicable to the case; by analyzing the stakeholder graph, it assesses the impact of different ruling results on the parties and the public interest. In this process, the system uses a variety of semantic reasoning algorithms. For example, the engine calls the case analogy algorithm in the knowledge layer and the wisdom layer to find similar past cases in terms of semantics and analyze their adjudication intent as a reference for the current case ruling; in the intent layer, the engine uses the value alignment algorithm to ensure that the AI's ruling tendency conforms to the input adjudication intent parameters. If the legal provisions themselves have ambiguities, the system can also extract relevant legal principles or legal interpretations from the knowledge base through the principle extraction algorithm to supplement the basis for ruling, thereby improving the rationality and legality of the conclusion.

[0023] After the above reasoning process, the judicial reasoning engine will generate a preliminary decision proposal, including the responsibility determination of the case, the applicable legal provisions, and the sentencing or handling scheme. At the same time, the system attaches semantic explanations to each proposal: for example, it points out that "according to Article X of the law and relevant cases, it is determined that the defendant's behavior constitutes Y crime", and "considering the reasonable demands of the plaintiff and the requirements of national interest protection, this scheme focuses on … when determining". These explanations are automatically generated by the engine according to its internal reasoning chain, derived from the associated paths in the semantic graph, helping users understand the basis behind the AI recommendations.

[0024] Finally, the trial intent management module is used to guide the human value and adjust the results of the preliminary proposal given by the engine. The judge can view each suggestion and its semantic explanation given by AI through the module interface. If the AI's consideration does not fully cover certain policy requirements or ethical factors, the judge can add or adjust the corresponding intent parameters through the module interface. For example, in an environmental public interest litigation case, the judge wants to emphasize the "environmental priority" principle, and can select the "environmental interest priority" intent option in the module. The system then rebalances the relevant weights in the stakeholder graph, allowing the engine to adjust the decision result, which may manifest as imposing stricter responsibilities on polluting enterprises. During this interaction process, the AI reasoning process is updated synchronously, and new suggestions and explanations are generated again for the judge to review. The whole human-machine collaborative closed-loop process can continue until the judge approves the AI's proposal. Through such a mechanism, artificial judicial experience and value judgment can be directly integrated into AI decision-making, so that the final decision result not only conforms to the legal provisions, but also conforms to reason and policy orientation.

[0025] Example 2: Semantic adaptation evaluation of legislative draft

[0026] In addition to assisting court decisions, the system of the present application can also be used by the legislative department to evaluate the semantic compatibility of new draft laws. Suppose a legislative body intends to introduce a new legal provision, but is concerned about potential conflicts with the existing legal system or feasibility issues in judicial practice. Using the ontological legal semantic library of the system, the draft provision can be compared with existing relevant laws, regulations and court cases in terms of semantics to find differences or overlaps in terminology definitions and legal concepts. The DIKWP judicial reasoning engine simulates the application process of the draft in actual cases, applies the draft provision to a series of hypothetical cases, deduces the judgment results of these cases under the new law framework, and compares them with the results under the existing law. If the AI finds that the wording of the draft may cause ambiguity (for example, an ambiguous definition of a term may lead to different understandings), or there is a conflict with the superior law in a specific situation, the system will mark and explain it in the evaluation results. This simulated evaluation helps lawmakers identify the loopholes and improvement space of the draft in advance, ensuring that the final enacted law is semantically rigorous and easy to apply.

[0027] It should be noted that the above embodiments are only used to illustrate the principles of the present application. For those skilled in the art, various modifications, optimizations or equivalent replacements can be made to the modules and processes of the above system without departing from the spirit and essence of the present application. These modifications or changes should also be considered to fall within the scope of protection of the present application. The present application combines the semantic sovereignty concept with judicial AI technology and proposes an intelligent decision support system that is autonomous, controllable and interpretable, which is of great significance to improve the efficiency and fairness of the judiciary.

Claims

1. A judicial intelligent decision support system based on semantic sovereignty, comprising an ontology-based legal semantic database, a judicial reasoning engine, and a trial intent management module, characterized in that: The ontological legal semantic database is used to semantically annotate and structurally store domestic legal provisions and judicial cases, constructing a legal knowledge ontology that includes legal concepts and their relationships, so as to preserve the local semantic connotation of legal language; the judicial reasoning engine performs multi-level semantic reasoning on case information based on a five-layer semantic model of data-information-knowledge-wisdom-intention, mapping objective case facts into a content semantic graph, mapping subjective demands and value factors into a stakeholder graph, and performing reasoning matching on the content semantic graph and the stakeholder graph in a unified semantic space to generate preliminary ruling suggestions; The trial intent management module provides a human-computer interaction interface, allowing users to input trial intent parameters to dynamically adjust the reasoning path and ruling results of the judicial reasoning engine, thereby integrating human judicial intent into the AI ​​decision-making process.

2. The judicial intelligent decision support system according to claim 1, characterized in that, The ontological legal semantic database constructs a legal knowledge graph, which uses semantic annotation to map elements such as legal provisions, legal concepts in judicial precedents, relationships between people, and event time to ensure that the semantic meaning of legal concepts is consistent with the legal tradition of the country.

3. The judicial intelligent decision support system according to claim 1, characterized in that, The judicial reasoning engine uses the five-layer semantic model to establish a semantic representation of the input case: extracting original evidence and legal text at the data layer, extracting key information of the case at the information layer, establishing a mapping relationship between case facts and legal provisions at the knowledge layer, comprehensively evaluating the case by combining judicial experience and the principle of fairness at the wisdom layer, and conducting an overall assessment and selection of the direction of the judgment by considering the judgment objectives and value orientation at the intention layer.

4. The judicial intelligent decision support system according to claim 1, characterized in that, The judicial reasoning engine constructs a case semantic model that includes a content graph and a stakeholder graph. The content graph includes the correspondence between the case's evidence chain and relevant legal provisions, while the stakeholder graph includes subjective intent information such as the parties' claims, social impact factors, and policy considerations. The judicial reasoning engine compares and reasons with the content graph and the stakeholder graph in a unified semantic space. It determines the applicable legal provisions by matching the case facts and legal provisions in the content graph, and assesses the impact of different rulings on the relevant parties by analyzing the stakeholder graph.

5. The judicial intelligent decision support system according to claim 1, characterized in that, The judicial reasoning engine invokes case analogy and principle extraction algorithms during the reasoning process: the case analogy algorithm retrieves semantically similar historical cases in the knowledge and wisdom layers and extracts their judicial intent as a reference for the current case ruling; when there are ambiguities or gaps in the legal provisions, the principle extraction algorithm extracts relevant legal principles or jurisprudential interpretations from the legal knowledge base to supplement the basis for the ruling.

6. The judicial intelligent decision support system according to claim 1, characterized in that, The trial intent management module allows input parameters for judicial intent, including judicial policy orientation and value orientation, which are used to adjust the weights of various factors in the stakeholder graph in the judicial reasoning engine, thereby changing the tendency of the adjudication recommendation to conform to the preset adjudication objectives.

7. The judicial intelligent decision support system according to claim 1, characterized in that, The system records semantic traces during the AI ​​reasoning process, providing traceable legal basis and value considerations for the conclusions reached at each step of the reasoning, thereby forming a transparent and interpretable chain of adjudication logic.

8. The judicial intelligent decision support system according to claim 1, characterized in that, The system is used to assess the semantic adaptability of new draft legislation. It compares the semantic differences between the draft provisions and the existing legal system through the ontological legal semantic database, uses the judicial reasoning engine to simulate the application effect of the draft provisions in hypothetical cases, and marks potential legal ambiguities or conflicts.

Images