CN-121981102-A - Decision making book generation method and system

CN121981102ACN 121981102 ACN121981102 ACN 121981102ACN-121981102-A

Abstract

One or more embodiments of the present disclosure provide a method and system for generating a decision, which implements alignment of an artificial intelligence reasoning path and legal proof logic by decomposing a judge thinking of a judge into a chain type verifiable judge flow of "fact structuring-dispute focus recognition-evidence analysis-legal applicability-judge item generation". The input and the output of each stage are patterned and structured, so that the whole decision generation process is visible, the decision is traceable according to clear and reasoning steps, and a controllable, reliable, strict, universal and transparent intelligent auxiliary judgment mechanism is constructed.

Inventors

LIU KAI
CHEN BANGHONG
ZHONG WEI
ZHANG WENQIAN
XIE QIAOMING
ZHANG JINGJING
WANG KELI
Tao Huifei
WANG CHAO

Assignees

共道网络科技有限公司

Dates

Publication Date: 20260505
Application Date: 20260407

Claims (10)

1. A method for generating a decision book, the method comprising: Inputting the prosecution, the answer dialect and the court trial strokes of the legal case into a fact generation model, carrying out structural processing on the facts of the legal case by entity identification and relation extraction by utilizing the fact generation model, generating disputed facts and non-disputed facts related to the legal case, and determining a dispute focus; carrying out three-dimensional identification on the structured evidence of the legal case by utilizing an evidence identification model to obtain three-dimensional identification results, and identifying contradictory evidence and determining a letter picking result; Inferentially finding the dispute focus according to the structural evidence, the tri-statement identification result and the credit acquisition result by using a fact identification model so as to generate a focus identification conclusion of the dispute focus; performing legal screening according to the dispute focus and the focus identification conclusion by using a legal screening model to obtain applicable legal cases; generating an approval finding content according to the non-disputed facts, the disputed focus, the focus identification conclusion, the tri-property identification result and the letter collection result by using a content generation model; Utilizing the content generation model to find out content and the applicable legal rules according to the trial, and generating the content considered by the court; Utilizing the content generation model to generate judgment item content according to the trial and approval content, the court thought content, the applicable legal laws and litigation request; and generating a judgment book of the legal case according to the trial finding content, the court considered content and the judgment item content.
2. The method of claim 1, wherein the fact generation model determines a process of dispute focus comprising: Calculating semantic similarity between different disputed facts; Clustering the disputed facts according to a preset semantic similarity threshold and the semantic similarity to merge the semantically related disputed facts into the same dispute cluster to obtain one or more dispute clusters; a dispute focus is determined from the one or more dispute clusters.
3. The method according to claim 1, wherein the method further comprises: Screening the legal rules in the legal rules library based on the rules of the legal cases to obtain first-level candidate legal rules; And performing legal screening according to the dispute focus and the focus identification conclusion by using a legal screening model to obtain applicable legal cases, wherein the method comprises the following steps: Inputting the first-level candidate legal strips, the dispute focus and the focus identification conclusion into the legal strip screening model, determining legal relation of the legal case according to the fact identification conclusion by the legal strip screening model, screening second-level candidate legal strips matched with the legal relation in the first-level candidate legal strips, and determining the second-level candidate legal strips required to be quoted for judging the dispute focus as the applicable legal strips.
4. The method of claim 1, wherein the content generation model performs legal compliance verification, factual closure verification, and program compliance verification on the criterion content after generating the criterion content, and outputs the criterion content after the criterion content passes the legal compliance verification, the factual closure verification, and the program compliance verification.
5. The method of claim 1, wherein the original evidence of the legal case is multi-modal evidence, the method further comprising: inputting original multi-modal evidence into a multi-modal model, carrying out semantic extraction on the multi-modal evidence by the multi-modal model to obtain semantic information corresponding to each multi-modal evidence, carrying out semantic alignment on the semantic information of each multi-modal evidence based on a preset format, and outputting the structured evidence.
6. The method according to claim 1, wherein the method further comprises: displaying one or more of the letter picking result, the focus identification conclusion and the applicable legal through a visual interface; Under the condition that the displayed letter collecting result is updated, the fact identification model is reused to carry out reasoning and finding out on the dispute focus according to the structural evidence, the trisection identification result and the updated letter collecting result so as to generate a new focus identification conclusion of the dispute focus; under the condition that the focus identification conclusion is updated, reusing the law screening model to determine a new applicable law of the legal case according to the updated focus identification conclusion; and when any input of the content generation model is updated, re-inputting updated data into the content generation model to re-generate the decision book.
7. The method of claim 1, wherein the fact-recognition model inferentially ascertains the dispute focus from the structured evidence, the triangulated recognition result, and the credit result to generate a focus-recognition conclusion for the dispute focus, comprising: Judging whether the focus identification conclusion can be deduced according to the structural evidence, the three-dimensional identification result and the letter acquisition result; In the case that the focus identification conclusion cannot be inferred, making a plurality of finding plans for the dispute focus; each finding plan is respectively inferred according to the structured evidence, the three-dimensional identification result and the credit acquisition result, and a corresponding plan finding result is obtained; and determining the focus identification conclusion according to plan finding results corresponding to the plurality of finding plans.
8. The method of claim 7, wherein the method further comprises: The finding plan, the plan finding result and the focus identification conclusion are displayed in the form of a mind map through a visual interface, wherein the dispute focus is a main node of the mind map, the finding plan is a child node of the main node, the plan finding result is a child node corresponding to the finding plan, and the focus identification conclusion is a summarized node of the plan finding results.
9. The method of claim 1 wherein the fact generation model is further configured to determine a dependency relationship between a first dispute focus and a second dispute focus if the number of disputed focuses is multiple, and wherein the fact identification model is further configured to first infer the first dispute focus if the dependency relationship is a premise that the first dispute focus is the second dispute focus, to obtain a first focus identification conclusion for the first dispute focus, and to determine a second focus identification conclusion for the second dispute focus based on the first focus identification conclusion.
10. A decision book generation system, the system comprising: The fact generation module is used for inputting the complaint, the answer dialect and the court trial strokes of the legal cases into the fact generation model, carrying out structural processing on the facts of the legal cases through entity identification and relation extraction by utilizing the fact generation model, generating disputed facts and non-disputed facts related to the legal cases, and determining a dispute focus; The evidence identification module is used for identifying the authenticity, the legality and the relativity of the structured evidence of the legal case by utilizing the evidence identification model to obtain a tri-identification result, identifying contradictory evidence and determining a letter picking result; the fact identification module is used for carrying out reasoning and finding out on the dispute focus according to the structural evidence, the trisection identification result and the credit acquisition result by using a fact identification model so as to generate a focus identification conclusion of the dispute focus; the legal case screening module is used for screening legal cases according to the dispute focus and the focus identification conclusion by utilizing a legal case screening model to obtain applicable legal cases; The content generation module is used for generating an approval finding content according to the non-disputed facts, the disputed focuses, the focus identification conclusions, the three-dimensional identification results and the credit acquisition results by utilizing a content generation model, generating a court identification content according to the approval finding content and the applicable legal rules by utilizing the content generation model, generating a judgment item content according to the approval finding content, the court identification content, the applicable legal rules and the litigation request by utilizing the content generation model, and generating a judgment book of the legal case according to the approval finding content, the court identification content and the judgment item content.

Description

Decision making book generation method and system Technical Field One or more embodiments of the present disclosure relate to the field of artificial intelligence, and in particular, to a method and system for generating a decision book. Background In the trial practice, the judge needs to manually read, analyze evidence, apply law and write a judge document on the file, and the process is highly dependent on personal experience and intellectual labor, has efficiency bottleneck and is easily influenced by subjective factors. In order to improve the quality and efficiency of judgment, an intelligent technology is developed, and the development of the intelligent technology mainly goes through four stages of document templating, element type judgment, knowledge graph application and direct generation of a large model. However, these approaches still have certain limitations that are difficult to meet in practice the high-level requirements for decision quality, logic stringency, and process interpretability. Disclosure of Invention In view of this, one or more embodiments of the present disclosure provide the following technical solutions: According to a first aspect of one or more embodiments of the present specification, there is provided a method of generating a decision book, the method comprising: Inputting the prosecution, the answer dialect and the court trial strokes of the legal case into a fact generation model, carrying out structural processing on the facts of the legal case by entity identification and relation extraction by utilizing the fact generation model, generating disputed facts and non-disputed facts related to the legal case, and determining a dispute focus; carrying out three-dimensional identification on the structured evidence of the legal case by utilizing an evidence identification model to obtain three-dimensional identification results, and identifying contradictory evidence and determining a letter picking result; Inferentially finding the dispute focus according to the structural evidence, the tri-statement identification result and the credit acquisition result by using a fact identification model so as to generate a focus identification conclusion of the dispute focus; performing legal screening according to the dispute focus and the focus identification conclusion by using a legal screening model to obtain applicable legal cases; generating an approval finding content according to the non-disputed facts, the disputed focus, the focus identification conclusion, the tri-property identification result and the letter collection result by using a content generation model; Utilizing the content generation model to find out content and the applicable legal rules according to the trial, and generating the content considered by the court; utilizing the content generation model to generate judgment content according to the trial and approval content, the court thought content, the applicable legal rules and the litigation request; and generating a judgment book of the legal case according to the trial finding content, the court considered content and the judgment item content. Optionally, the process of determining the dispute focus by the fact generation model includes: Calculating semantic similarity between different disputed facts; Clustering the disputed facts according to a preset semantic similarity threshold and the semantic similarity to merge the semantically related disputed facts into the same dispute cluster to obtain one or more dispute clusters; a dispute focus is determined from the one or more dispute clusters. Optionally, the method further comprises: Screening the legal rules in the legal rules library based on the rules of the legal cases to obtain first-level candidate legal rules; And performing legal screening according to the dispute focus and the focus identification conclusion by using a legal screening model to obtain applicable legal cases, wherein the method comprises the following steps: inputting the first-level candidate legal strips, the dispute focus and the focus identification conclusion into the legal strip screening model, determining legal relation of the legal case according to the fact identification conclusion by the legal strip screening model, screening second-level candidate legal strips matched with the legal relation from the first-level candidate legal strips, and determining the second-level candidate legal strips required to be quoted for judging the dispute focus as the applicable legal strips. Optionally, the content generation model performs legal consistency verification, fact closure verification and program compliance verification on the judgment content after the judgment content is generated, and outputs the judgment content after the judgment content passes the legal consistency verification, the fact closure verification and the program compliance verification. Optionally, the original evidence of the legal case is multi-moda