Multidimensional intention embedding interpretable judgment recommendation system based on DIKWP model

Abstract

The invention discloses a multi-dimensional intention embedding interpretable judgment recommendation system based on a DIKWP model. The system is composed of a judgment inference engine, an intention embedding module, an interpretation generation module and an interactive display interface. The engine performs hierarchical reasoning according to data-information-knowledge-intelligence-intention, and generates referee suggestions in combination with fact elements, legal provisions and class case retrieval; the intention module acts value weights of fairness, efficiency, policies and the like on a target layer to regulate and control the cutting amount; and the interpretation module synchronously outputs a semantic reasoning path which comprises a fact basis, a legal basis, a class case reference and an intention description, and provides hierarchical presentation according to user types. The system improves judicial AI transparency and reviewable performance, and is suitable for court case handling assistance, judgment document interpretation, class case research and other scenes.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence and the field of explainable artificial intelligence, and more particularly, to an explainable judgment recommendation system fusing multi-dimensional judgment intention, which is used to provide intelligent judgment suggestions with clear reasoning paths and basis explanations. BACKGROUND

[0002] In judicial adjudication, the adjudicator not only gives the adjudication conclusion, but also must explain the reasons, that is, the so-called "explain the law and reason". Similarly, if the adjudication suggestions given by artificial intelligence (AI) lack sufficient explanation, it will be difficult to gain the trust of judges and parties. With the development of the construction of intelligent courts, more and more AI technologies are introduced into the field of judicial adjudication assistance, but some current legal AI systems (such as sentencing prediction models) often give results as "black boxes" and cannot explain their inference process. This contradicts the high requirements of transparency and accountability in the judicial field. Therefore, explainability has become one of the key bottlenecks for the practical application of judicial AI.

[0003] In addition, judicial adjudication involves multiple purposes and value balance: not only to adjudicate according to law and realize justice, but also to reflect policy guidance and consider social impact. Human judges will consider legal effects and social effects and other factors when making decisions. However, traditional AI models mostly output results based on historical data patterns, which may ignore these value dimensions, resulting in suggestions that do not reflect the latest judicial policies or people's sense of justice. For example, based only on past case data, AI may mechanically continue the pattern of severe punishment in certain cases, but if the current judicial policy emphasizes the principle of combining leniency with severity, such a suggestion is obviously inappropriate. Therefore, in order to make AI adjudication suggestions truly have practical value, the purpose of legal adjudication must be considered and presented in a form understandable to humans.

[0004] It is worth noting that "consciousness relativism" points out that the output of an intelligent agent is only considered to truly embody "consciousness" or intelligence when it is understood and given meaning by the observer. In other words, if the output of AI (such as judgment suggestions) cannot be mapped into the cognitive framework and legal knowledge system of judges, judges are likely to consider the output unreliable and difficult to accept. Therefore, it is crucial for the judgment recommendation process of AI to be "understandable and understandable" to humans. This suggests that we need to build an AI judgment explanation mechanism to convert the reasoning path inside AI into expressions consistent with the thinking way of legal professionals, so that it can be integrated into the cognitive closed loop of humans and realize human-machine collaborative work. Only in this way, can judges regard AI as a "legal wisdom" assistant rather than a black box tool that cannot understand its logic. SUMMARY

[0005] The main purpose of the present application is to provide an interpretable decision recommendation system based on the DIKWP model. The system introduces multi-dimensional intention embedding and semantic path display mechanism, so that AI can provide clear reasoning process explanation while giving decision suggestions, and clearly explain the legal provisions and case support, thereby improving the intelligibility and credibility of the system output. The system not only focuses on "what result is recommended", but also pays attention to "why this result is recommended", so that users can clearly understand the origin and basis of AI decision suggestions.

[0006] The present application innovatively combines multi-dimensional judicial intention and interpretable AI technology. In the decision reasoning process, the core purpose of judicial decision (such as punishment, education, relief, fairness, efficiency, social impact, etc.) is taken as an intention signal, which is embedded in the purpose layer of the DIKWP model to guide the reasoning direction; at the same time, the key steps and supporting basis of each layer in the reasoning process are recorded to form a complete semantic decision path. When the system generates a decision suggestion, it will output an explanation report at the same time, which contains a complete reasoning chain from fact determination to law application to discretionary weighing. For example, the explanation report may list: "the fact A in this case is identified through comparison, which meets the legal requirement X (according to certain relevant provisions); the decision result B in case Y (similar to the case) is analogically referenced; considering the policy guidance Z of the intention layer (such as the current policy of balancing punishment and leniency), the result C is recommended." In this way, users can see that every step in the AI decision process has a basis: there is law to rely on (listing relevant legal provisions), there is case to follow (quoting similar cases), and there is reason to rely on (explaining value intention consideration). This clear reasoning context greatly facilitates human understanding and review of AI conclusions, making AI's decision suggestions a dialogable and verifiable result, rather than a obscure and difficult to understand numerical output.

[0007] The present application also focuses on the hierarchical and customized explanation. Since the DIKWP model has a natural hierarchical structure, the system can provide different levels of explanation for different audiences: for professional users such as judges, the system focuses on showing detailed reasoning details of the knowledge layer and wisdom layer (such as reasons for legal application, comparative analysis of similar cases, etc.); for non-professional users such as parties, a higher level of summary explanation is provided (such as citing a typical case and explaining "why the law judges like this"), so as to avoid using too many professional terms. Through this relativistic perspective, the explanation content can fit the cognitive framework of different observers, truly making it easy for the knowledgeable to understand and easy for the benevolent to believe. When users of different backgrounds form a common understanding of AI output, the information gap between man and machine is eliminated, and AI is more likely to be regarded as a system with "consciousness" and intelligence.

[0008] Compared with the prior art, the application combines the powerful data analysis capability of artificial intelligence with the rigor and transparency of legal reasoning, thereby effectively solving the pain points of "not understanding and not being credible" of judicial AI. On the basis of meeting the legal accuracy, the value orientation is fused, so that the judgment suggestion is not only in line with the legal principles but also reasonable. The explanation report output by the system enhances the transparency and reviewability of the AI decision-making process, and can significantly improve the reliance of judges and parties on AI-assisted judgment.

[0009] Technical scheme

[0010] The DIKWP model summarizes the layer-by-layer cognitive process of artificial intelligence from data to intention. On the basis of the traditional DIKW model (Data-Information-Knowledge-Wisdom), the purpose (intention) layer is added, thereby forming five levels of data (Data), information (Information), knowledge (Knowledge), wisdom (Wisdom) and intention (Purpose). As shown in Figure 1 The layers of the DIKWP model from bottom to top are data layer (D), information layer (I), knowledge layer (K), wisdom layer (W) and intention layer (P), which embodies the process of deepening cognition and integrating target orientation.

[0011] Decision reasoning engine: the core engine module for multi-level semantic analysis and reasoning based on the DIKWP model. The engine is responsible for layer-by-layer reasoning of a case from facts to judgment, including fact extraction, legal requirement matching, case search, legal application and discretionary decision making (similar to the idea of an artificial judge trying a case). During the reasoning process, the engine will mark and store important intermediate outputs. For example, it will identify key factual elements in the case, match the corresponding legal requirements that are met or not met, search for relevant legal provisions and guiding cases, form a preliminary judgment conclusion, and consider the influence of preset intention factors on discretion in the process. The engine ensures that the reasoning process flows through the data, information, knowledge, wisdom layers from bottom to top according to the DIKWP model, and connects with the intention at the highest level. Each step of processing records the traces of "input-processing-output", providing material basis for future explanation.

[0012] Intention embedding module: This module manages the quantified modeling of multi-dimensional judicial intention and its application in the reasoning process. On the one hand, it provides an interface that allows system administrators or users to pre-set various judicial intention parameters and their weights (for example, fairness: efficiency = 7:3, or emphasizing a certain policy orientation); on the other hand, during the runtime of the judgment reasoning engine, the intention embedding module dynamically converts the above intention parameters into guidance signals for the reasoning process. For example, this module can adjust the evaluation function of the case retrieval algorithm to make the retrieval results more biased towards the "fairness" dimension (improve the consistency of the results with similar cases), or biased towards the "efficiency" dimension (solve the dispute more quickly). For another example, embed the current policy rules in the discretionary decision-making stage to impose constraints on the sentencing results - if the current judicial policy is strict on drug crimes, the AI's sentencing recommendation will be correspondingly increased. By applying these guidance signals at the purpose layer of the DIKWP model, it is equivalent to influencing the decisions of the wisdom layer, so that the final recommended judicial result can reflect the pre-set value orientation. The intention embedding module will also record the intention content and its weight applied, as part of the explanation report, to explain "which value targets were considered in this decision and to what extent".

[0013] Explanation generation module: This module aggregates process data recorded from the judgment reasoning engine and the intention embedding module to automatically generate structured explanation content. The explanation generation module extracts and summarizes the reason chain behind the judicial recommendation based on key semantic nodes and references in the judgment reasoning process. Specifically, it includes:

[0014] Factual basis: List the main facts and evidence of the case that the decision recommendation is based on (corresponding to the data layer input of the DIKWP model), and explain how these facts meet or do not meet the relevant legal requirement conditions (related to the information layer logic judgment of DIKWP).

[0015] Legal basis: Enumerate the applicable legal provisions and their content of the case (corresponding to the knowledge layer of DIKWP), and explain how these legal norms support the conclusion of the current case.

[0016] Case reference: Give summaries of cases similar to the current case and their judicial results (corresponding to the wisdom layer application of DIKWP), and explain how the current case refers to these precedents and what insights or comparison results are obtained. For example, it can be pointed out that "the decision recommendation of the current case is generally consistent with the reference cases", or "due to the lighter circumstances of the current case compared to the reference cases, a lighter treatment is recommended".

[0017] Intention description: Clearly describe the value goals considered in the decision-making process (corresponding to the intention layer of DIKWP), such as "considering the defendant's repentance to reflect the principle of combining leniency with severity" and "considering the factor of judicial credibility to balance the feelings of the public", and explain the influence direction and degree of these intention signals on the final judgment result.

[0018] Reasoning process: Use natural language to describe how the system gradually reasons through each layer of DIKWP to finally derive the judgment conclusion. For example: "The system first confirms facts A and B (evidence is sufficient), and based on this, it determines that legal requirement X has been met; then according to relevant law Y and guiding case Z, it judges that the defendant's behavior constitutes the corresponding legal responsibility; finally, combined with the current policy orientation, it decides to take measure C as appropriate." Such a description outlines the complete reasoning link from fact determination, law application to discretionary decision-making.

[0019] The explanation content output by the explanation generation module can be organized in a tree structure (corresponding to the hierarchical expansion of DIKWP model layers) or in the form of question and answer dialogue (for example, the user asks "Why is this result recommended?" and the system answers "because…"). to improve the friendliness and coherence of reading.

[0020] Interactive display interface: The user interaction interface is used to present the above-mentioned judgment suggestion and its explanation in a friendly way to different types of end users. For professional users such as judges, the interface can provide a tree diagram form to display the explanation in a hierarchical expansion: by default, only the conclusion and main basis of AI suggestion are displayed, and users can expand to view more detailed reasoning steps, involved law texts, case details, etc. For non-professional users, the interface can use question and answer explanation or visual chart presentation, such as key factor influence chart, similar case comparison chart, etc. to highlight the key points in an intuitive form, so that ordinary people can also understand the reasons behind the judgment. The interface also supports user feedback on the explanation content, such as providing "Is the explanation easy to understand?" "Do you agree with the recommended result?" etc. feedback options, and the system can optimize the presentation of the explanation accordingly, making it more in line with the user's thinking habits and understanding level. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 is a schematic diagram of the hierarchical structure of the DIKWP model.

[0022] Figure 2 is a schematic diagram of the overall architecture of the explainable judgment recommendation system of the present application.

[0023] Figure 3 is a schematic diagram of the components of the explanation report content described in the present application. DETAILED DESCRIPTION

[0024] Example 1: Court-assisted judgment

[0025] In the practice of court trials, the system can be integrated into the judge's case handling assistant tool and used in conjunction with similar case recommendation and other functions. When the judge receives the AI recommended judgment result and the corresponding explanation report, he or she can easily understand the reasoning basis of the AI. For example, a summary of the explanation report may read:

[0026] Suggested result: dismiss the appeal and uphold the original judgment.

[0027] Reason summary: based on the fact finding, the appellant's breach of contract is clear and the evidence is sufficient; according to Article ** of the Contract Law, the appellant should bear the liability for breach of contract; reference to similar cases (such as Case

number

[0028] Through such an explanation report, the judge can quickly understand the AI's approach - every step has a basis and a reason. From fact finding, legal basis, to reference cases and discretionary considerations, everything is clearly presented. This highly transparent process makes the judge more confident when referring to AI suggestions, and it is easier to integrate the reasonable analysis into his or her own judgment documents, thereby improving the sufficiency of the judgment reasoning. Once the AI's analysis logic is understood and accepted by the judge, AI becomes a "helper" rather than an "opponent" for the judge, which helps to improve the efficiency of the trial and the fairness and consistency of the judgment results.

[0029] Example 2: Online interpretation of judgment documents

[0030] The system can also serve the public-facing judgment document interpretation platform. When the public reads the effective judgment documents online, if they encounter professional terms or complex legal arguments and have difficulty understanding the reasons for the judgment, they can call on the system to re-analyze the semantics of the document and generate an "interpretation view" for their reference. The interpretation view uses simple language and case comparison to explain the reasons behind the judge's decision to the public. For example, the system can explain the application of legal provisions in everyday language: "Why did the court make this decision? Because the law requires both parties to a contract to act in good faith, and in this case one party clearly breached the contract, so they must bear the liability for compensation." At the same time, the interpretation view can cite one or two similar cases to illustrate how this handling reflects fairness and reason in judicial practice. Through the interpretation of the system, the public can understand "why the law judges like this" even without a legal background, thereby improving their understanding and acceptance of the judgment results. This application helps to enhance the transparency and authority of judicial decisions and increase the public's trust in the judiciary.

[0031] Example 3: Similar case analysis and legal research

[0032] Lawyers, prosecutors or legal researchers can also use the system to generate an in-depth case analysis report. The report will not only give the AI's decision recommendation result for the case, but also explain how the outcome may change under different value weight choices. For example, the report can say: "In this case, when the fairness weight is relatively high, the AI tends to give a result closer to the average judgment level; when efficiency is prioritized, the AI is more inclined to recommend a settlement between the two parties to quickly close the case." In addition, the report will compare the subtle differences in the judgment of the case and several similar cases, analyze the reasons for these differences, such as different boundary identifications of legal requirements, different repentance performances of the defendants leading to different sentencing ranges, etc. This multi-angle, multi-dimensional analysis tool has important value for legal practice and research: legal practitioners can use it to understand the rules and policy direction of judicial discretion and provide reference for litigation strategy; legal scholars can use the data produced by the system to study the evolution of judgment standards and evaluate judicial fairness. It can be predicted that this interpretable case analysis report will have broad application prospects in legal consulting, case analysis and decision support scenarios, and has important commercial value.

[0033] In summary, the interpretable judgment recommendation system provided by the present application combines the powerful data processing capability of AI with the explainability of legal reasoning process, solving the pain point problem of "not understanding and not trusting" in the application of judicial AI. By integrating legal purpose consideration into the judgment recommendation and providing clear explanation and description, the system greatly improves the transparency and credibility of AI judgment recommendation. Its application will accelerate the construction process of "intelligent court", making AI a real powerful assistant for judges and a reliable tool for the public. The system has good industrialization prospects and can be applied to court judgment assistance, legal consulting platform, case intelligent evaluation system and other scenarios, providing high-value intelligent services for judicial practitioners and the public.

Claims

1. A multi-dimensional intent embedding interpretable decision recommendation system based on DIKWP model, characterized in that, The system comprises: a judgment reasoning engine for multi-level semantic analysis and reasoning of a case according to a DIKWP model (Data-Information-Knowledge-Wisdom-Purpose) to generate a judgment suggestion; an intention embedding module for embedding multi-dimensional judicial intention parameters and weights into the intention layer of the DIKWP model during reasoning to guide discretion and result generation; an explanation generation module for recording semantic nodes and basis information at each level during reasoning to generate an explanation report containing factual basis, legal basis, case reference, and intention explanation; an interactive display interface for displaying the judgment suggestion and corresponding explanation content to different users in a visual or hierarchical expansion manner.

2. The system of claim 1, wherein, The judgment reasoning engine sequentially performs the steps of fact extraction, legal requirement matching, case retrieval, legal application, and discretion decision-making, and records the process data of input, processing, and output at each step.

3. The system of claim 1 or 2, wherein, The intention embedding module is configured to: receive user or system preset intention weight parameters, including but not limited to fairness, efficiency, policy orientation, and social impact; convert the intention parameters into reasoning guide signals to affect case retrieval, discretion calculation, or result recommendation direction; record the applied intention and its weight in the explanation report to explain the value orientation of the judgment suggestion.

4. The system according to any one of claims 1 to 3, characterized in that, The explanation generation module comprises: a factual basis generation unit for explaining how the facts of the case satisfy the legal requirements; a legal basis generation unit for listing applicable legal provisions; a case reference generation unit for giving similar cases and their handling results; an intention explanation unit for explaining the considered judicial intention and its impact on the result; a reasoning chain construction unit for generating a complete reasoning chain from fact determination to discretion decision-making.

5. The system according to any one of claims 1 to 4, characterized in that, The explanation report is output in a tree structure or question and answer form, supporting user interaction of layer-by-layer expansion or questioning.

6. The system according to any one of claims 1 to 5, characterized in that, The interactive display interface can adaptively display the explanation content according to the user type: for professional users, display detailed reasoning steps, legal provision content, and case comparison; for ordinary users, display simplified summary explanation or visual relationship diagram.

7. The system according to any one of claims 1 to 6, characterized in that, The system further comprises a user feedback module for collecting user understanding and recognition of the explanation content and adjusting the explanation presentation strategy accordingly.

8. The system according to any one of claims 1 to 7, characterized in that, The system is suitable for application scenarios such as court case handling assistance, judgment document interpretation, case analysis, and judicial research.

Images