Auxiliary generation method and system for judicial decision, electronic equipment and storage medium

Abstract

The embodiment of the invention provides a judicial decision auxiliary generation method and system, electronic equipment and a storage medium, and the method comprises the steps: extracting a law article suitable for a given case fact from a preset law article list based on case description information, and obtaining a law article related to a case; the given case facts are case facts corresponding to the case description information; analyzing case elements contained in the case description information to obtain corresponding case interpretations; wherein the case interpretation comprises information of case elements which are related to the given case facts and can influence a case judgment result; and based on the case description information, the law articles related to the case and the case interpretation, generating a judgment result. By applying the embodiment of the invention, interpretable and traceable reasoning of intelligent judicial decision is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, in particular to a method and system for assisting in generating a judicial decision, an electronic device and a storage medium. BACKGROUND

[0002] The traditional case review and judicial decision rely on the professional answers and argumentation processes of legal professionals such as judges, lawyers, prosecutors, etc. In the face of complex criminal cases, etc., a large amount of manpower and material resources need to be spent for review. In response to the huge needs of legal practitioners and ordinary citizens, legal decision assistants are rapidly developing. With the rapid development of big data technology, the development process of social digitization and intelligentization is also accelerating. Based on the excellent performance of large language models (LLMs) in various tasks, it is considered to apply large language models to the legal field to assist in legal decisions to achieve intelligent judicial decisions, which helps to improve the work efficiency in the legal field.

[0003] How to realize the explainable and traceable reasoning of intelligent judicial decisions has become a problem to be solved at present. SUMMARY

[0004] The purpose of the embodiments of the present application is to provide a method and system for assisting in generating a judicial decision, an electronic device and a storage medium to realize the explainable and traceable reasoning of intelligent judicial decisions. The specific technical solutions are as follows:

[0005] In a first aspect, the embodiments of the present application provide a method for assisting in generating a judicial decision, which comprises:

[0006] Based on the case description information, a law applicable to a given case fact is extracted from a preset law list to obtain a law related to the case; the given case fact is a case fact corresponding to the case description information;

[0007] The case elements contained in the case description information are analyzed to obtain a corresponding case interpretation; wherein the case interpretation contains information of case elements related to the given case fact that can affect the case decision result;

[0008] Based on the case description information, the law related to the case, and the case interpretation, a decision result is generated.

[0009] Optionally, the method for assisting in generating a judicial decision comprises:

[0010] determining, based on the case description information and a law in the preset law list, a law applicable to a given case fact by using a pre-trained first large language model, to obtain a law related to the case;

[0011] Alternatively, a similarity matching calculation is performed on the case description information and the laws in the preset law list, and a law related to the case is determined according to a calculation result.

[0012] Optionally, the case elements contained in the case description information are parsed to obtain a corresponding case interpretation, including:

[0013] The case description information is input into a pre-trained second large language model for case element parsing to obtain a corresponding case interpretation; wherein the second large language model is trained according to sample case description information and label information corresponding to the sample case description information; the case interpretation includes events reflecting case facts, elements and attribute information of the elements contained in the events, and relationships between the events.

[0014] Optionally, the case description information, the law related to the case, and the case interpretation are used to generate a judgment result, including:

[0015] The law related to the case is used to prune a with-or-graph AOG case knowledge base that is pre-constructed, to obtain a target AOG case knowledge base; the AOG case knowledge base is pre-constructed based on the laws in the preset law list;

[0016] The case description information, the case interpretation, and the target AOG case knowledge base are used to generate a judgment result.

[0017] Optionally, the case description information, the case interpretation, and the target AOG case knowledge base are used to generate a judgment result, including:

[0018] The matching result between the case description information and the case interpretation and the content corresponding to a bottom leaf node in the target AOG case knowledge base is determined to obtain the matching result corresponding to the bottom leaf node;

[0019] In a bottom-up manner, the matching result corresponding to the bottom leaf node is passed to an upper node of the bottom leaf node in the target AOG case knowledge base, and a matching result corresponding to the upper node is determined, until a matching result of a root node in the target AOG case knowledge base is obtained;

[0020] The matching result of the root node is used to generate a judgment result.

[0021] Optionally, the construction of the AOG case knowledge base includes:

[0022] Identifying and analyzing the legal provisions in the preset legal provision list to determine the intermediate nodes of the AOG case knowledge base;

[0023] For each intermediate node, based on the analysis of the internal logic of each legal provision in the preset legal provision list, determine the leaf nodes contained in the intermediate node;

[0024] The judgment result is taken as the root node;

[0025] Based on the root node, the intermediate node and the leaf node, an AOG case knowledge base is constructed.

[0026] In a second aspect, the embodiments of the present application provide a method for assisting the generation of judicial decisions, which comprises:

[0027] Analyzing the case elements contained in the case description information to obtain the corresponding case interpretation; wherein the case interpretation contains information about case elements related to a given case fact that can affect the judgment result of the case; the given case fact is the case fact corresponding to the case description information;

[0028] Based on the case description information, the case interpretation and the pre-constructed AND / OR graph AOG case knowledge base, a judgment result is generated; the AOG case knowledge base is pre-constructed based on the legal provisions in the preset legal provision list.

[0029] Optionally, the analysis of the case elements contained in the case description information to obtain the corresponding case interpretation comprises:

[0030] The case description information is input into a pre-trained second large language model for case element analysis to obtain the corresponding case interpretation; wherein the second large language model is trained according to sample case description information and sample case description information corresponding to the labeled information; the case interpretation includes events reflecting case facts, elements and attribute information of elements contained in the events, and relationships between events.

[0031] Optionally, the generation of a judgment result based on the case description information, the case interpretation and the pre-constructed AND / OR graph AOG case knowledge base comprises:

[0032] Determine the matching result between the case description information and the case interpretation and the content corresponding to the bottom leaf node in the AOG case knowledge base to obtain the matching result corresponding to the bottom leaf node;

[0033] In a bottom-up manner, the matching result corresponding to the bottom layer leaf node is transmitted to the upper layer node of the bottom layer leaf node in the AOG case knowledge base, and the matching result of the corresponding upper layer node is determined, until the matching result of the root node in the AOG case knowledge base is obtained;

[0034] The matching result of the root node is used to generate a judgment result.

[0035] Optionally, the construction of the AOG case knowledge base comprises:

[0036] The legal provisions in the preset legal provision list are identified and analyzed to determine the intermediate nodes of the AOG case knowledge base;

[0037] For each intermediate node, the leaf nodes contained in the intermediate node are determined based on the analysis of the internal logic of each legal provision in the preset legal provision list;

[0038] The judgment result is taken as the root node;

[0039] Based on the root node, the intermediate node and the leaf node, the AOG case knowledge base is constructed.

[0040] In a third aspect, an embodiment of the present application provides a system for assisting in generating a judicial judgment, and the system comprises:

[0041] A legal provision extraction module is configured to extract legal provisions applicable to a given case fact from a preset legal provision list based on case description information, to obtain legal provisions related to the case; the given case fact is a case fact corresponding to the case description information;

[0042] A first case interpretation module is configured to analyze case elements contained in the case description information to obtain a corresponding case interpretation; wherein the case interpretation contains information about case elements related to the given case fact and capable of affecting the judgment result of the case;

[0043] A first judgment generation module is configured to generate a judgment result based on the case description information, the legal provisions related to the case, and the case interpretation.

[0044] In a fourth aspect, an embodiment of the present application provides a system for assisting in generating a judicial judgment, and the system comprises:

[0045] A second case interpretation module is configured to analyze case elements contained in the case description information to obtain a corresponding case interpretation; wherein the case interpretation contains information about case elements related to the given case fact and capable of affecting the judgment result of the case; the given case fact is a case fact corresponding to the case description information;

[0046] The second decision generation module is configured to generate a decision result based on the case description information, the case interpretation, and a pre-constructed AOG case knowledge base.

[0047] In a fifth aspect, an electronic device is provided, which includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus.

[0048] The memory is configured to store a computer program.

[0049] The processor is configured to execute the program stored in the memory to implement the method described above.

[0050] In a sixth aspect, a computer readable storage medium is provided, which stores a computer program, and the computer program is executed by a processor to implement the method described above.

[0051] In a seventh aspect, a computer program product containing instructions is provided, which, when executed on a computer, causes the computer to perform the method described above.

[0052] The embodiments of the present application have the following beneficial effects:

[0053] The embodiments of the present application provide a method and system for assisting the generation of judicial decisions, an electronic device and a storage medium, which extract a legal provision applicable to a given case fact, so that the generated decision result has a legal basis, analyze the case elements contained in the case description information to obtain a case interpretation containing case elements related to the given case fact and capable of affecting the decision result of the case, so that the case facts can be expressed in a structured manner, and then generate a decision result based on the case description information, the legal provision related to the case and the case interpretation, the process of generating the decision result includes the legal basis and the structured expression of the case facts, so that the judicial decision has explainability and traceable reasoning, therefore, the explainability and traceable reasoning of intelligent judicial decision are realized, so as to facilitate the analysis and use of relevant personnel and improve the work efficiency in the legal field.

[0054] Of course, implementing any product or method of the present application does not necessarily require achieving all the advantages described above at the same time. BRIEF DESCRIPTION OF DRAWINGS

[0055] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings described below are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained based on these drawings.

[0056] Figure 1 A schematic diagram of the auxiliary generation method of judicial decisions provided by the embodiments of the present application;

[0057] Figure 2 A flowchart of the auxiliary generation method of judicial decisions provided by the embodiments of the present application;

[0058] Figure 3 A case interpretation schematic diagram provided by the embodiments of the present application;

[0059] Figure 4 Another flowchart of the auxiliary generation method of judicial decisions provided by the embodiments of the present application;

[0060] Figure 5 An AOG case knowledge base schematic diagram provided by the embodiments of the present application;

[0061] Figure 6 Another flowchart of the auxiliary generation method of judicial decisions provided by the embodiments of the present application;

[0062] Figure 7 A structural schematic diagram of the auxiliary generation system of judicial decisions provided by the embodiments of the present application;

[0063] Figure 8 Another structural schematic diagram of the auxiliary generation system of judicial decisions provided by the embodiments of the present application;

[0064] Figure 9 A structural schematic diagram of the electronic device provided by the embodiments of the present application. DETAILED DESCRIPTION

[0065] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art based on the present application are within the scope of protection of the present application.

[0066] Firstly, the related terms in the embodiments of the present application are introduced:

[0067] Law extraction: based on the given case facts, according to the event elements contained in the facts and the expert experience, based on the semantic understanding and reasoning ability of the model, one or more law articles are extracted as the basis for the legal judgment. The model here can be a large language model.

[0068] Case interpretation: based on the given case facts, the criminal event and the event type related to the fact, the content of the criminal event and the event related to the fact, and the attributes of the people and things related to the criminal event and the event related to the fact, and the time, place, and other criminal events of the criminal event and the event related to the fact are extracted.

[0069] AOG (And-Or Graph): based on the content of the law article, the logical relationship between the law articles in the law article is extracted, and a complete law article and or graph (as a knowledge base for judging cases, providing basis and reference for judicial judgment) is constructed. AOG is a tool that uses logical relationships such as and, or, and not to organize various facts and descriptions into a graphical representation.

[0070] With the rapid development of big data technology, the development process of social digitization and intelligentization is also accelerating. Based on the excellent performance of large language models in various tasks, large language models can be considered for application in the legal field. Therefore, in the case of applying large language models in the legal field, how to realize the explainable and traceable reasoning of intelligent judicial decision has become a problem to be solved.

[0071] To solve the above problems, the embodiments of the present application provide a method and system for assisting the generation of judicial decisions, electronic devices and storage media. The method and system for assisting the generation of judicial decisions provided by the embodiments of the present application can be applied to electronic devices, which can be terminal devices or server devices. For example, the method for assisting the generation of judicial decisions can be applied to an intelligent judicial simulation judgment system, which can be deployed in a terminal device or a server device.

[0072] The method for assisting the generation of judicial decisions provided by the embodiments of the present application, as shown in Figure 1 According to the law extraction based on the case facts, the law articles related to the case facts (i.e. the judgment basis law articles) are obtained, the AOG case knowledge base constructed based on the law articles in the pre-set law article list is pruned, the dynamic AOG extraction is realized, and the event interpretation graph is obtained based on the case interpretation. The target AOG case knowledge base obtained by dynamic AOG extraction and the event interpretation graph are combined to perform AOG reasoning to generate a judgment result.

[0073] The method for assisting the generation of judicial decisions provided by the embodiments of the present application will be described in detail.

[0074] Referring to Figure 2 The method for assisting in generating a judicial decision provided by the embodiment of the present application can include the following steps:

[0075] In S201, the law applicable to the given case facts is extracted from the preset law list based on the case description information, and the law related to the case is obtained.

[0076] The given case facts are the case facts corresponding to the case description information.

[0077] In S202, the case elements contained in the case description information are parsed to obtain the case interpretation.

[0078] The case interpretation contains information about the case elements related to the given case facts that can affect the case decision.

[0079] In S203, the decision result is generated based on the case description information, the law related to the case, and the case interpretation.

[0080] By applying the embodiment of the present application, the law applicable to the given case facts is extracted, so that the generated decision result has a legal basis. The case elements contained in the case description information are parsed to obtain the case interpretation containing the case elements related to the given case facts that can affect the case decision. The case facts can be expressed in a structured manner. Furthermore, the decision result is generated based on the case description information, the law related to the case, and the case interpretation. The process of generating the decision result includes the legal basis and the structured expression of the case facts, so that the judicial decision has explainability and traceable reasoning. Therefore, the explainability and traceable reasoning of intelligent judicial decision are realized, which facilitates the analysis and use of relevant personnel and improves the work efficiency in the legal field.

[0081] In S201, the case description information is information describing the basic situation (such as the process) of the event. The given case facts are the case facts corresponding to the case description information. The preset law list can be a set of existing known laws. The laws contained in the law list can be obtained after being manually annotated and removed by law experts. For example, the laws contained in the law list can be laws and judicial interpretations in the criminal law and / or the general provisions / particular provisions of civil law, or they can be laws related to one or more representative criminal or civil causes, or they can be laws in the criminal law and / or the general provisions of civil law and the corresponding provisions of the particular provisions of the relevant causes to be decided, etc. For example, the criminal or civil causes can be the four-level causes of smuggling, trafficking, transporting, and manufacturing drugs, and the two-level causes of smuggling and trafficking of prohibited articles. Specifically, the laws contained in the preset law list and the number of laws are not limited in the embodiment of the present application.

[0082] The case description information is matched with the legal provisions in the legal provision list one by one to extract legal provisions suitable for the given case facts and capable of assisting in the judgment of the case from the legal provision list, and then the legal provisions related to the case are obtained.

[0083] In S202, the case description information is parsed to extract case elements included in the given case facts, which may include, for example, events embodying the case facts, elements (people, objects, time, place, etc.) contained in the events and attribute information (specific information corresponding to people, objects, time, place, etc.) of the elements, and relationships between related events.

[0084] For example, as shown in Figure 3 , taking the drug trafficking case corresponding to the case description information as an example, the content formed in the case interpretation includes: events embodying the case facts (i.e. content such as drug trafficking, confession); wherein the elements contained in the first drug trafficking event (the leftmost event in Figure 3 ) are: drug trafficking objects, amounts, buyers and sellers, and the attribute information corresponding to each element is: drugs (quantity ag (grams)), AA yuan, Yuanmianmian, Xiangmianmian; the elements contained in the second drug trafficking event (the event in the middle of Figure 3 ) are: sellers, buyers, amounts, drug trafficking objects, and the attribute information corresponding to each element is: Xiangmianmian, Zhoumianmian, BB yuan, drugs (quantity bg); the elements contained in the confession event are: confessor, and the attribute information corresponding to the event is Xiangmianmian, and the relationships between the events (the confession event is respectively associated with the first drug trafficking event and the second drug trafficking event).

[0085] In S203, the case description information, the legal provisions related to the case extracted, and the case interpretation are combined to generate the corresponding judgment result by using legal reasoning.

[0086] Referring to Figure 4 , another method for assisting in the generation of judicial judgments provided by an embodiment of the application can include the following steps:

[0087] In S401, based on the case description information and the legal provisions in the preset legal provision list, a pre-trained first large language model is used to determine legal provisions suitable for the given case facts, and legal provisions related to the case are obtained.

[0088] In the embodiments of the present application, the first large language model can be trained in advance according to sample case description information and sample legal provisions, and the applicable results corresponding to the sample case description information and the sample legal provisions. The first large language model can also be fine-tuned and trained based on an open source large language model, such as a GPT4 (Generative Pre-trained Transformer 4) language model, a Llama large language model, and a Chinese open source text embedding model text2vec, using the collected sample case description information and the legal provisions applicable to the sample case description information, to obtain a pre-trained first large language model. The data or task instructions used to train the first large language model can be determined according to the case involved, the scene in which the model is used (such as judicial consultation), and the effect required by the model, and can be adaptively adjusted. The applicable results corresponding to the sample case description information and the sample legal provisions, and the legal provisions applicable to the sample case description information can be manually annotated data by legal field experts. In the process of training the first large language model, based on the semantic understanding of the sample case description information and the sample legal provisions, it is determined whether the sample case description information and the sample legal provisions are applicable, so as to accurately train the first large language model.

[0089] In one example, each legal provision in the legal provision list and the case description information can be input into the pre-trained first large language model one by one to make the first large language model determine whether the current legal provision is applicable to the given case facts and output the determination result of whether it is applicable. Each legal provision in the legal provision list and the case description information can also be input into the pre-trained first large language model to make the first large language model determine whether each legal provision in the legal provision list is applicable to the given case facts and output the determination result of whether each legal provision is applicable to the given case facts. Specifically, the use method of the first large language model corresponds to the training method, that is, when the input is one legal provision and case description information during training, the input is also one legal provision and case description information during use; when the input is the entire legal provision list and case description information during training, the input is also the entire legal provision list and case description information during use.

[0090] When all the legal provisions in the legal provision list are traversed, the legal provisions in the determination result indicating that the legal provisions are applicable to the given case facts are determined as the legal provisions related to the case. The legal provisions related to the case obtained can specifically include the serial number or identifier and content description corresponding to the legal provisions, and the like.

[0091] Exemplary, the case description information is specifically: on aa year bb month cc day dd, the defendant Long drives a motorcycle to A market after selling drugs by telephone, sells drugs p to Liu 2 for w yuan, and is arrested by public security personnel. The public security personnel seized the above-mentioned drug p (net weight P kg) at Liu 2, and seized drug funds w yuan and crime tools mobile phone and motorcycle at the defendant Long. The above facts are also not disputed by the defendant Long in the court trial, and are confirmed by the evidence of witnesses Liu 1 and Su, and are sufficient to be identified. The above case description information and the criminal law and / or general provisions / special provisions of civil law and judicial interpretation of the law contained in the law list are input into the pre-trained first large language model, and the judgment result of whether each law is applicable to the given case fact is output, and the law related to the case is obtained: “Article 64”, “Article 347”, “Article 52”, “Article 67”, “Article 53”.

[0092] Because the training of the first large language model is based on the semantic understanding of the sample case description information and the sample law, whether the sample case description information and the sample law are applicable, so that the law related to the case obtained by using the pre-trained first large language model is more accurate and has more legal basis, so as to better assist judicial decision.

[0093] In one possible implementation, the first large language model can also be pre-trained, and the law applicable to the given case fact is extracted from the preset law list to obtain the law related to the case in the following way:

[0094] Similar matching calculation is performed on the case description information and the laws in the preset law list, and the laws related to the case are determined according to the calculation result.

[0095] Specifically, the case description information is matched with the laws in the law list one by one, and the matching (i.e. applicable) law is determined as the law related to the case. Wherein, the similar matching calculation can be the similarity calculation between the semantics of the case description information and the semantics of the law, such as by calculating the cosine similarity, Euclidean distance and the like, of course, other similarity calculation methods can also be used, and the embodiments of the present application are not limited thereto.

[0096] Through the semantic similarity of the case description information and the law, the law related to the case determined is more accurate and has more legal basis, so as to better assist judicial decision.

[0097] S402, input the case description information into the pre-trained second large language model to analyze the case elements, and obtain the corresponding case interpretation.

[0098] The second large language model is trained according to sample case description information and label information corresponding to the sample case description information. Here, the label information corresponding to the sample case description information is the label case interpretation corresponding to the sample case description information. For example, the second large language model can also use the Llama 7B large language model as a base, and use a decoder-based Transformer generation model to fine-tune the training on the labeled data.

[0099] By using the pre-trained second large language model to analyze the case elements in the case description information, the corresponding case interpretation is obtained, which realizes the intelligent case interpretation, so as to further realize the intelligent judicial decision.

[0100] In one example, the format of the label information corresponding to the sample case description information can be set according to legal professional knowledge and case judgment experience, and the format includes a case paradigm and an outline. The outline can include events reflecting case facts in the sample case description information, elements contained in the events, and relationships between the events, etc. For example, the events include drug trafficking, drug possession, truthful confession, etc., and the relationship between the events is that the truthful confession event is for the drug trafficking event. When the event is drug trafficking, the drug trafficking event contains elements such as subject, object, time, and place. Accordingly, the case paradigm can be represented in the form of “if the labeled element is …, the attribute of the element contains …”. For example, if the labeled element is a drug, the attribute of the element can include components and content, and if the labeled element is a person, the attribute of the element can include age and gender. The corresponding information is extracted from the sample case description information according to the set case paradigm and outline to obtain the label information corresponding to the sample case description information.

[0101] Further, the case description information can be input into the pre-trained second large language model, the second large language model encodes the case description information, and extracts the case elements contained in the encoded case description information according to the format of the label information corresponding to the sample case description information, to obtain the case interpretation including events reflecting case facts, elements contained in the events and attribute information of the elements, and relationships between the events.

[0102] For example, the case description information is as follows: On ccc day of bbb month of aaa, Lei and the defendant Li contacted each other by phone to trade drugs. Lei gave Li W yuan of drug money outside a net bar. Li returned to the entrance of the net bar after t minutes and sold q small bags of suspected drugs (net weight Q grams) to Lei. After the transaction, Li was arrested by the police. The police seized the suspected drugs from Lei on the spot. The above facts are not disputed by the defendant Li during the trial, and are confirmed by the defendant's confession and defense, and the witness Lei's testimony.

[0103] The above case description information is input into the pre-trained second large language model to obtain the following output:

[0104]

[0105]

[0106] Using the pre-trained second large language model to analyze the case description information, the semantic understanding and case fact extraction of the case description information are realized, and the extracted case elements can be structured and displayed, which is helpful for assisting judicial decisions and facilitating analysis and use by relevant personnel.

[0107] S403, pruning the pre-constructed AOG case knowledge base using the law related to the case to obtain a target AOG case knowledge base.

[0108] The pre-constructed AOG case knowledge base is constructed based on the law in the pre-set law list. In one possible implementation, the construction process of the AOG case knowledge base can include:

[0109] Identify and analyze the law in the pre-set law list to determine the intermediate nodes of the AOG case knowledge base; for each intermediate node, determine the leaf nodes contained in the intermediate node based on the analysis of the internal logic of each law in the pre-set law list; the judgment result is taken as the root node; based on the root node, the intermediate node and the leaf node, the AOG case knowledge base is constructed.

[0110] In one example, the legal provisions in the preset legal provision list can be identified by semantic matching, and the legal provisions in the preset legal provision list can be parsed by named entity recognition, semantic disambiguation, etc. to determine the intermediate nodes of the AOG cause knowledge base, which represent the upper nodes (or parent nodes) containing the leaf nodes. The legal provisions in the preset legal provision list can also be identified and parsed by expert annotation or natural language capabilities of large language models to determine the intermediate nodes of the AOG cause knowledge base. For example, the preset legal provision list includes legal provisions in the Criminal Law, such as the main types of criminal punishment, including: (1) control; (2) detention; (3) fixed-term imprisonment; (4) life imprisonment; (5) death penalty. Accordingly, the intermediate node is “main punishment”, and the corresponding leaf nodes are “control”, “detention”, “fixed-term imprisonment”, “life imprisonment” and “death penalty”.

[0111] For example, the internal logic of each legal provision in the preset legal provision list can be parsed by a large language model. For example, the large language model can be fine-tuned and trained in advance according to the content of sample legal provisions, task instructions, and annotated legal provision logical relationships, and then when the large language model is used to parse the internal logic of each legal provision in the preset legal provision list, the content of the legal provision and the task instruction are directly input into the trained large language model, and the logical relationship of the corresponding legal provision is output.

[0112] Further, based on the parsing of the internal logic of each legal provision in the preset legal provision list, the leaf nodes contained in each intermediate node are determined. For example, the content of the legal provision is: “Smuggling, trafficking, transporting, manufacturing drugs, regardless of the amount, shall be held criminally responsible and punished. Smuggling, trafficking, transporting, manufacturing drugs, if one of the following conditions is met, shall be punished with 15 years of fixed-term imprisonment, life imprisonment or death penalty, and shall be subject to confiscation of property: (1) Smuggling, trafficking, transporting, manufacturing drugs XX grams or more or large amounts of drugs; (2) The chief of the smuggling, trafficking, transporting, manufacturing drug group … … ” The task instruction is: if XXX condition is met, it is the child node of YY node. The parsed legal provision logical relationship is: “Smuggling, trafficking, transporting, manufacturing drugs XX grams or more … … ” as the child node of “15 years of fixed-term imprisonment”, “life imprisonment” and “death penalty”. In other words, one of the leaf nodes of the intermediate nodes “15 years of fixed-term imprisonment”, “life imprisonment” and “death penalty” is determined to be “smuggling, trafficking, transporting, manufacturing drugs XX grams or more”. By analogy, the leaf nodes contained in each intermediate node are determined by expanding from the upper node to the lower node.

[0113] And, the decision result is taken as the root node of all nodes, at this time the decision result has no substantial content. Specifically, in the process of assisting judicial decision by using the constructed AOG case knowledge base, the specific content of the decision result is determined based on the logical relationship of each intermediate node, and the final result is output.

[0114] Then, a complete AND / OR graph is constructed by using the determined root node, intermediate node and leaf node, forming an AOG case knowledge base as a knowledge base for judging cases.

[0115] For example, part of the constructed AOG case knowledge base is shown in Figure 5 The constructed AOG case knowledge base can be represented as a tree-shaped AND / OR graph, which contains root nodes (decisions), intermediate nodes (penalties, others, principal penalties, additional penalties, sentencing ranges, etc.), and leaf nodes (principal penalties, additional penalties, sentencing ranges, controls, detention, …, mitigation, leniency, exemption from punishment, and satisfaction conditions, etc.), each node corresponds to different article content or logical judgment results for article content, and the relationship between each node and other nodes is any one of AND, OR and NOT.

[0116] The construction of the case knowledge base in the form of AOG makes the case knowledge base reusable and can be expanded to more cases to facilitate intelligent judicial reasoning.

[0117] Further, for the pre-constructed AOG case knowledge base, based on the extracted articles from the pre-set article list that are applicable to the given case facts and related to the case, the nodes in the AOG case knowledge base corresponding to the extracted articles related to the case are retained, the pruning of the AOG case knowledge base is realized, and the target AOG case knowledge base is obtained.

[0118] For example, the nodes corresponding to different articles can be removed by using the identification of each article, and the nodes corresponding to the same article are retained, and the target AOG case knowledge base is obtained.

[0119] S404, based on the case description information, the case interpretation and the target AOG case knowledge base, a decision result is generated.

[0120] In one possible implementation, the implementation process of generating a decision result based on case description information, case interpretation and target AOG case knowledge base includes:

[0121] The matching result between the case description information and the case interpretation and the content corresponding to the bottom layer leaf node in the target AOG case knowledge base is determined to obtain the matching result corresponding to the bottom layer leaf node. In a bottom-up manner, the matching result corresponding to the bottom layer leaf node is transmitted to the upper layer node of the bottom layer leaf node in the target AOG case knowledge base, and the matching result corresponding to the upper layer node is determined, until the matching result of the root node in the target AOG case knowledge base is obtained. The judgment result is generated by using the matching result of the root node.

[0122] In one example, the case description information and the case interpretation can be taken as a whole, and the third large language model is used to determine the matching result between the case description information and the case interpretation and the content corresponding to the bottom layer leaf node in the target AOG case knowledge base. For example, the third large language model can be trained according to sample case description information and case interpretation, sample statute content, and sample statute content whether it conforms to the sample case description information and case interpretation. Then, the case description information and the case interpretation can be taken as a whole, and the content corresponding to the bottom layer leaf node in the target AOG case knowledge base is input into the pre-trained third large language model to output the judgment result of whether the content corresponding to the bottom layer leaf node exists in the case description information and the case interpretation. The judgment result can be represented by TRUE / FALSE, wherein TRUE indicates that the content corresponding to the bottom layer leaf node (the condition described by the node) conforms to the given case fact, and FALSE indicates that the content corresponding to the bottom layer leaf node does not conform to the given case fact. The case description information and the case interpretation can also be taken as a whole, and the similarity judgment is performed on the content corresponding to the bottom layer leaf node in the target AOG case knowledge base to determine whether the content corresponding to the bottom layer leaf node appears in the case description information and the case interpretation.

[0123] In the embodiment of the application, the first large language model, the second large language model and the third large language model can be three independent models, or can be one large language model. In the case of one large language model, the first large language model, the second large language model and the third large language model correspond to three different tasks of the large language model.

[0124] Further, in a bottom-up manner, the matching result corresponding to the bottom layer leaf node is transmitted to the upper layer node (i.e. the parent node corresponding to the leaf node) of the bottom layer leaf node in the target AOG case knowledge base, and the matching result corresponding to the upper layer node is determined, until the matching result of the root node in the target AOG case knowledge base is obtained, that is, the matching result of each leaf node is transmitted to the root node layer by layer using the “AND / OR” logical relationship of the intermediate node, the reasoning result is transmitted from bottom to top, and the final judgment result is generated.

[0125] Exemplary, case description information is specifically: aaa year bbb month ccc day ddd hours, Lei and the defendant Li telephone contact drug transactions, first outside a net bar drug funds W yuan to the defendant Li. Li left t minutes after returning to the net bar door, and will q small bag of suspected drugs (net weight Q grams) sold to Lei, after the transaction was arrested by the police. Police on the spot from Lei body seized the suspected drug package. The above facts, the defendant Li in the process of trial also no objection, and there are defendant's confession and defense, witness Lei testimony and other evidence to prove, enough to identify.

[0126] The relevant law for the current case is: [‘Article 347’, ‘Article 64’, ‘Article 25’, ‘Article 27’, ‘Article 67’].

[0127] The case interpretation is:

[0128]

[0129]

[0130] Correspondingly, the generated judgment result is represented as:

[0131]

[0132] By applying the embodiment of the present application, the law applicable to the given case facts is extracted, so that the generated judgment result has a legal basis, the case elements contained in the case description information are analyzed, the case interpretation containing the case elements related to the given case facts and capable of affecting the case judgment result is obtained, so that the case facts can be structured expressed, and then based on the case description information, the law related to the case and the case interpretation, the judgment result is generated by AOG reasoning, the process of generating the judgment result includes the legal basis, the structured expression of the case facts and the AOG reasoning process, so that the judicial judgment has explainability and traceable reasoning, therefore, the explainability and traceable reasoning of intelligent judicial decision are realized, so as to facilitate the analysis and use of relevant personnel, and improve the work efficiency in the legal field.

[0133] Referring to Figure 6 Another judicial decision auxiliary generation method provided by the embodiment of the present application can include the following steps:

[0134] S601, the case elements contained in the case description information are analyzed, and the corresponding case interpretation is obtained.

[0135] Case description information is information describing the basic situation (such as the process) of an event. A given case fact is a case fact corresponding to the case description information. The case description information is parsed to extract case elements contained in the given case fact, which may include, for example: events embodying the case fact, elements (people, objects, time, place, etc.) contained in the event, and attribute information of the elements (specific information corresponding to people, objects, time, place, etc.), and the relationship between related events, and the like. The case interpretation is formed by these case elements related to the given case fact that can affect the case judgment result, and accordingly, the case interpretation contains information of the case elements related to the given case fact that can affect the case judgment result.

[0136] As an example, as shown in Figure 3 , taking a drug trafficking case corresponding to the case description information as an example, the content of the formed case interpretation includes: events embodying the case fact (i.e. content, such as drug trafficking, confession); among them, the elements contained in the first drug trafficking event (the leftmost event in Figure 3 ) are: drug trafficking objects, amount, buyers, and sellers, and the attribute information corresponding to each element is: drugs (quantity ag), AA yuan, Yuanmianmian, Xiangmianmian; the elements contained in the second drug trafficking event (the event in the middle of Figure 3 ) are: seller, buyer, amount, drug trafficking object, and the attribute information corresponding to each element is: Xiangmianmian, Zhoumianmian, BB yuan, drugs (quantity bg); the elements contained in the confession event are: confessor, and the corresponding attribute information is Xiangmianmian, and the relationship between events (the confession event is respectively associated with the first drug trafficking event and the second drug trafficking event).

[0137] S602, based on the case description information, the case interpretation, and the pre-constructed AOG case knowledge base, generate a judgment result.

[0138] The case description information and the case interpretation are combined, legal reasoning is performed on the pre-constructed AOG case knowledge base, and a corresponding judgment result is generated. The pre-constructed AOG case knowledge base is pre-constructed based on the law articles in the preset law article list. The preset law article list can be a set of existing known law articles, and the law articles included in the law article list can be obtained after artificial labeling and removal by legal experts. For example, the law articles included in the law article list can be law articles in criminal law and / or general provisions / particular provisions of civil law and judicial interpretations, or can be law articles related to one or more representative criminal cases or civil cases, or can be law articles in criminal law and / or general provisions of civil law and relevant case-specific provisions, and the like. The criminal cases or civil cases can be, for example, four-level cases of smuggling, trafficking, transporting, and manufacturing drugs, and two-level cases of smuggling and trafficking of prohibited articles. Specifically, the number of law articles included in the preset law article list is not limited in the embodiments of the present application.

[0139] By applying the embodiments of the present application, the case elements included in the case description information are analyzed to obtain case interpretations including case elements related to given case facts and capable of affecting the judgment result of the case, so that the case facts can be expressed in a structured manner. Furthermore, based on the case description information, the case interpretation, and the pre-constructed AOG case knowledge base, the judgment result is generated. Since the AOG case knowledge base is pre-constructed based on the law articles in the preset law article list, the process of generating the judgment result includes legal basis and structured expression of case facts, so that the judicial judgment has explainability and traceable reasoning. Therefore, the explainability and traceable reasoning of intelligent judicial judgment are realized, which facilitates analysis and use by relevant personnel and improves the work efficiency in the legal field.

[0140] In a possible implementation, the implementation of the step S601 of analyzing the case elements included in the case description information to obtain the corresponding case interpretation includes: inputting the case description information into a pre-trained second large language model to analyze the case elements and obtain the corresponding case interpretation.

[0141] The second large language model is trained according to sample case description information and corresponding labeled information of the sample case description information. Here, the labeled information corresponding to the sample case description information is the labeled case interpretation corresponding to the sample case description information. For example, the second large language model can also be based on the Llama 7B large language model, and a decoder-based Transformer generation model is used to fine-tune the training on the labeled data.

[0142] The second large language model is pre-trained, and case elements in the case description information are parsed by the second large language model to obtain corresponding case interpretation, so that the intelligent case interpretation is realized, and intelligent judicial decision is further realized.

[0143] In one example, according to legal professional knowledge and case judgment experience, the format of the label information corresponding to the sample case description information can be set to include a case paradigm and an outline. The outline can include events reflecting case facts in the sample case description information, arguments (i.e., elements) contained in the events, and relationships between the events, etc. For example, the events include drug trafficking, drug possession, truthful confession, etc. The relationship between the events is: the “truthful confession event” confesses the “drug trafficking event”. When the event is drug trafficking, the drug trafficking event contains arguments such as subject, object, time, and location. Accordingly, the case paradigm can be represented in the form of “if the labeled argument is …, the attribute of the argument contains …”. For example, if the labeled argument is a drug, the attribute of the argument can include components and content. If the labeled argument is a person, the attribute of the argument can include age and gender. The corresponding information is extracted from the sample case description information according to the set case paradigm and outline to obtain the label information corresponding to the sample case description information.

[0144] Further, the case description information can be input into the pre-trained second large language model. The second large language model encodes the case description information, and extracts the case elements contained in the encoded case description information according to the format of the label information corresponding to the sample case description information, to obtain the case interpretation including events reflecting case facts, elements contained in the events, attribute information of the elements, and relationships between the events.

[0145] The pre-trained second large language model is used to parse the case elements in the case description information, which realizes semantic understanding and case fact extraction of the case description information, and can structure the extracted case elements for display, which is helpful for assisting judicial decision and facilitating analysis and use by relevant personnel.

[0146] In one possible implementation, an AOG case knowledge base can be pre-constructed based on the legal provisions in the preset legal provision list. The construction process of the AOG case knowledge base can include:

[0147] The legal provisions in the preset legal provision list are identified and parsed to determine the intermediate nodes of the AOG case knowledge base. For each intermediate node, the leaf nodes contained in the intermediate node are determined based on the analysis of the internal logic of each legal provision in the preset legal provision list. The judgment result is taken as the root node. The AOG case knowledge base is constructed based on the root node, the intermediate nodes, and the leaf nodes.

[0148] In one example, the legal provisions in the preset legal provision list can be identified by semantic matching, and the legal provisions in the preset legal provision list can be parsed by named entity recognition, semantic disambiguation, etc. to determine the intermediate nodes of the AOG cause knowledge base, which represent the upper nodes (or parent nodes) containing the leaf nodes. The legal provisions in the preset legal provision list can also be identified and parsed by expert annotation or natural language capabilities of large language models to determine the intermediate nodes of the AOG cause knowledge base. For example, the preset legal provision list includes legal provisions in the Criminal Law, such as the main types of criminal punishment in the Criminal Law, including: (1) control; (2) detention; (3) fixed-term imprisonment; (4) life imprisonment; (5) death penalty. Correspondingly, the intermediate node is “main punishment”, and the corresponding leaf nodes are “control”, “detention”, “fixed-term imprisonment”, “life imprisonment” and “death penalty”.

[0149] For example, the internal logic of each legal provision in the preset legal provision list can be parsed by a large language model. For example, the large language model is fine-tuned and trained in advance according to the content of sample legal provisions, task instructions and annotated legal provision logical relationships, and then when the large language model is used to parse the internal logic of each legal provision in the preset legal provision list, the content of the legal provision and the task instruction are directly input into the trained large language model, and the logical relationship of the corresponding legal provision is output.

[0150] Further, based on the parsing of the internal logic of each legal provision in the preset legal provision list, the leaf nodes contained in each intermediate node are determined. For example, the content of the legal provision is: “Smuggling, trafficking, transporting and manufacturing drugs, regardless of the amount, shall be held criminally responsible and punished. Smuggling, trafficking, transporting and manufacturing drugs shall be punished with 15 years of fixed-term imprisonment, life imprisonment or death penalty, and confiscation of property: (1) Smuggling, trafficking, transporting and manufacturing drugs XX grams or more or large amounts of drugs; (2) The chief of the smuggling, trafficking, transporting and manufacturing drug group … … ”, the task instruction is: if XXX condition is met, it is the child node of YY node, and the parsed legal provision logical relationship is: “Smuggling, trafficking, transporting and manufacturing drugs XX grams or more … … ” as the child node of “15 years of fixed-term imprisonment”, “life imprisonment” and “death penalty”. In other words, it is determined that one of the leaf nodes of the intermediate nodes “15 years of fixed-term imprisonment”, “life imprisonment” and “death penalty” is “smuggling, trafficking, transporting and manufacturing drugs XX grams or more … … ”. By analogy, the leaf nodes contained in each intermediate node are determined by expanding from the upper node to the lower node.

[0151] And, the decision result is taken as the root node of all nodes, at this time the decision result has no substantial content. Specifically, in the case of using the constructed AOG case knowledge base to assist judicial decision, the specific content of the decision result is determined based on the logical relationship of each intermediate node, and is output as the final result.

[0152] Then, using the determined root node, intermediate node and leaf node, a complete AND-OR graph is constructed to form an AOG case knowledge base as a knowledge base for case decision.

[0153] For example, part of the constructed AOG case knowledge base is shown in Figure 5 The constructed AOG case knowledge base can be represented as a tree-shaped AND-OR graph, which includes root nodes (decisions), intermediate nodes (penalties, others, principal punishment, additional punishment, sentencing range, etc.), and leaf nodes (principal punishment, additional punishment, sentencing range, control, detention, …, mitigation, leniency, exemption from punishment, and the like). Each node corresponds to different article content or logical judgment results for article content, and the relationship between each node and other nodes is any one of AND, OR and NOT.

[0154] The construction of the case knowledge base in the form of AOG makes the case knowledge base reusable and can be expanded to more cases to facilitate intelligent judicial reasoning.

[0155] In one possible implementation, the implementation of the above step S602 based on the case description information, case interpretation and the pre-constructed AOG case knowledge base to generate the decision result includes:

[0156] Determine the matching result between the case description information and the case interpretation and the content corresponding to the bottom leaf node in the AOG case knowledge base, to obtain the matching result corresponding to the bottom leaf node; in a bottom-up manner, pass the matching result corresponding to the bottom leaf node to the upper layer node of the bottom leaf node in the AOG case knowledge base, and determine the matching result corresponding to the upper layer node, until the matching result of the root node in the AOG case knowledge base is obtained; and generate the decision result using the matching result of the root node.

[0157] In one example, the case description information and the case interpretation can be taken as a whole, and a third large language model is used to determine the matching result between the case description information and the case interpretation and the content corresponding to the bottom leaf node in the AOG case knowledge base. For example, the third large language model can be trained according to sample case description information and case interpretation, sample legal provision content, and sample legal provision content whether it conforms to the sample case description information and case interpretation. Then, the case description information and the case interpretation can be taken as a whole, and the content corresponding to the bottom leaf node in the AOG case knowledge base is input into the pre-trained third large language model to output a judgment result of whether the content corresponding to the bottom leaf node exists in the case description information and the case interpretation. The judgment result can be represented by TRUE / FALSE, where TRUE indicates that the content (condition described by the node) corresponding to the bottom leaf node conforms to the given case fact, and FALSE indicates that the content corresponding to the bottom leaf node does not conform to the given case fact. The case description information and the case interpretation can also be taken as a whole, and similarity judgment is performed on the content corresponding to the bottom leaf node in the AOG case knowledge base to determine whether the content corresponding to the bottom leaf node appears in the case description information and the case interpretation.

[0158] Further, in a bottom-up manner, the matching result of the bottom leaf node is transmitted to the upper node (i.e., the parent node of the leaf node) of the bottom leaf node in the AOG case knowledge base, and the matching result corresponding to the upper node is determined, so as to obtain the matching result of the root node in the AOG case knowledge base, that is, the matching result of each leaf node is transmitted to the root node layer by layer using the “AND / OR” logical relationship of the intermediate node, the reasoning result is transmitted from bottom to top, and the final judgment result is generated.

[0159] By applying the embodiment of the present application, the case elements contained in the case description information are parsed to obtain the case interpretation containing the case elements related to the given case fact and capable of affecting the case judgment result, so that the case fact can be expressed in a structured manner. Then, based on the case description information and the case interpretation, the AOG reasoning is used to generate the judgment result. Since the AOG case knowledge base is pre-constructed based on the legal provisions in the preset legal provision list, the process of generating the judgment result includes the structured expression of the legal basis and the case fact and the AOG reasoning process, so that the judicial judgment has explainability and traceable reasoning. Therefore, the explainability and traceable reasoning of the intelligent judicial judgment are realized, so as to facilitate the analysis and use of relevant personnel and improve the work efficiency in the legal field.

[0160] Referring to Figure 7 The embodiment of the present application also provides an auxiliary generation system of judicial judgment, which comprises:

[0161] The law clause extraction module 701 is configured to extract a law clause applicable to a given case fact from a preset law clause list based on case description information, to obtain a law clause related to the case;

[0162] The first case interpretation module 702 is configured to parse a case element included in the case description information, to obtain a corresponding case interpretation; the case interpretation includes information of a case element related to the given case fact and capable of affecting a case judgment result;

[0163] The first judgment generation module 703 is configured to generate a judgment result based on the case description information, the law clause related to the case, and the case interpretation.

[0164] Optionally, the law clause extraction module 701 is specifically configured to:

[0165] determine, based on the case description information and the law clauses in the preset law clause list, a law clause applicable to the given case fact by using a pre-trained first large language model, to obtain the law clause related to the case;

[0166] Alternatively, the case description information and the law clauses in the preset law clause list are subjected to similarity matching calculation, and the law clause related to the case is determined according to a calculation result.

[0167] Optionally, the first case interpretation module 702 is specifically configured to input the case description information into a pre-trained second large language model to parse a case element, to obtain a corresponding case interpretation; the second large language model is trained according to sample case description information and labeled information corresponding to the sample case description information; the case interpretation includes an event reflecting a case fact, an element included in the event and attribute information of the element, and a relationship between the events.

[0168] Optionally, the first judgment generation module 703 includes:

[0169] a pruning unit configured to prune a pre-constructed AOG case knowledge base by using the law clause related to the case, to obtain a target AOG case knowledge base; the AOG case knowledge base is pre-constructed based on the law clauses in the preset law clause list;

[0170] a judgment generation unit configured to generate a judgment result based on the case description information, the case interpretation, and the target AOG case knowledge base.

[0171] Optionally, the judgment generation unit is specifically configured to:

[0172] determine a matching result between the case description information and the case interpretation and content corresponding to a bottom leaf node in the target AOG case knowledge base, to obtain a matching result corresponding to the bottom leaf node;

[0173] In a bottom-up manner, the matching result corresponding to the bottom layer leaf node is transmitted to the upper layer node of the bottom layer leaf node in the target AOG case knowledge base, and the matching result corresponding to the upper layer node is determined, until the matching result of the root node in the target AOG case knowledge base is obtained;

[0174] The matching result of the root node is used to generate a judgment result.

[0175] Optionally, the system further comprises a first construction module.

[0176] The first construction module is configured to identify and analyze the legal provisions in the preset legal provision list, determine the intermediate nodes of the AOG case knowledge base, determine the leaf nodes contained in each intermediate node based on the analysis of the internal logic of each legal provision in the preset legal provision list, take the judgment result as the root node, and construct the AOG case knowledge base based on the root node, the intermediate nodes and the leaf nodes.

[0177] Referring to Figure 8 The embodiment of the present application also provides another auxiliary generation system of judicial judgments, which comprises:

[0178] The second case interpretation module 801 is configured to analyze the case elements contained in the case description information to obtain a corresponding case interpretation; wherein the case interpretation contains information of case elements related to a given case fact and capable of affecting the judgment result of the case; the given case fact is a case fact corresponding to the case description information.

[0179] The second judgment generation module 802 is configured to generate a judgment result based on the case description information, the case interpretation and a pre-constructed AND-OR graph AOG case knowledge base; the AOG case knowledge base is pre-constructed based on the legal provisions in the preset legal provision list.

[0180] Optionally, the second case interpretation module 801 is specifically configured to input the case description information into a pre-trained second large language model to analyze the case elements and obtain a corresponding case interpretation; wherein the second large language model is trained according to sample case description information and labeled information corresponding to the sample case description information; the case interpretation includes events reflecting case facts, element and attribute information of the elements contained in the events, and relationships between the events.

[0181] Optionally, the second judgment generation module 802 is specifically configured to:

[0182] determine the matching result between the case description information and the case interpretation and the content corresponding to the bottom layer leaf node in the AOG case knowledge base, and obtain the matching result corresponding to the bottom layer leaf node;

[0183] In a bottom-up manner, the matching result corresponding to the bottom leaf node is transmitted to the upper node of the bottom leaf node in the AOG case knowledge base, and the matching result of the corresponding upper node is determined, until the matching result of the root node in the AOG case knowledge base is obtained.

[0184] The decision result is generated by using the matching result of the root node.

[0185] Optionally, the system further comprises a second construction module.

[0186] The second construction module is configured to identify and analyze the legal provisions in the preset legal provision list, determine the intermediate nodes of the AOG case knowledge base, determine the leaf nodes contained in each intermediate node based on the analysis of the internal logic of each legal provision in the preset legal provision list, take the decision result as the root node, and construct the AOG case knowledge base based on the root node, the intermediate nodes and the leaf nodes.

[0187] The embodiment of the present application further provides an electronic device, as shown in the figure. Figure 9 The processor 901, the communication interface 902 and the memory 903 can communicate with each other through the communication bus 904.

[0188] The memory 903 is configured to store a computer program.

[0189] The processor 901 is configured to execute the program stored in the memory 903, and implement the steps of the auxiliary generation method of the judicial decision, so as to achieve the same technical effects.

[0190] The communication bus mentioned in the above electronic device can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The communication bus can be divided into an address bus, a data bus and a control bus. For the convenience of representation, only one thick line is shown in the figure, but it does not mean that there is only one bus or one type of bus.

[0191] The communication interface is configured to communicate between the above electronic device and other devices.

[0192] The memory can include a random access memory (RAM) and a non-volatile memory (NVM), such as at least one disk memory. Optionally, the memory can also be at least one storage device located away from the above-mentioned processor.

[0193] The processor described above can be a general processor, including a central processing unit (CPU), a network processor (NP), etc.; can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component.

[0194] In another embodiment provided by the present application, a computer readable storage medium is also provided, and the computer readable storage medium stores a computer program. The computer program is executed by a processor to implement the steps of the method for assisting the generation of a judicial decision in any of the above embodiments, so as to achieve the same technical effects.

[0195] In another embodiment provided by the present application, a computer program product containing instructions is also provided, and when the computer program product is run on a computer, the computer is caused to execute the steps of the method for assisting the generation of a judicial decision in any of the above embodiments, so as to achieve the same technical effects.

[0196] In the above embodiments, all or part of the embodiments can be implemented by software, hardware, firmware or any combination thereof. When implemented by software, all or part of the embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network or other programmable apparatus. The computer instructions can be stored in a computer readable storage medium or transmitted from one computer readable storage medium to another computer readable storage medium, for example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center through wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.) manner. The computer readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. integrated with one or more available media. The available medium can be a magnetic medium (for example, floppy disk, hard disk, magnetic tape), an optical medium (for example, DVD), or a semiconductor medium (for example, solid state disk (SSD)) and the like.

[0197] It is to be noted that, in the present document, relational terms such as first and second and the like can be used solely to distinguish one entity or action from another entity or action without necessarily implying any actual relationship or order between such entities or actions. Also, the terms "comprises", "comprising", or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises... a" does not, without more constraints, exclude the presence of additional identical elements in the process, method, article, or apparatus that comprises the element.

[0198] Each of the embodiments in the present document is described in a related manner, and the same or similar parts among the embodiments can be referred to each other. Each of the embodiments focuses on the difference from other embodiments. In particular, for the system / electronic device / storage medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method embodiments.

[0199] The preferred embodiments of the present application have been described above with the aid of drawing figures, and are not intended to limit the scope of the present application. Any modification, equivalent replacement, or improvement made without departing from the spirit and principle of the present application shall fall within the scope of the present application.

Claims

1. A method of assisting in the production of a judicial decision, characterized in that, The method comprises: extracting a law applicable to a given case fact from a preset law list based on case description information to obtain a law related to the case; the given case fact is a case fact corresponding to the case description information; parsing a case element contained in the case description information to obtain a corresponding case interpretation; wherein the case interpretation contains information of a case element related to the given case fact and capable of affecting the case judgment result; generating a judgment result based on the case description information, the law related to the case, and the case interpretation.

2. The method of claim 1, wherein, The method comprises: determining a law applicable to a given case fact by using a pre-trained first large language model based on the case description information and the laws in the preset law list to obtain a law related to the case; or, performing similarity matching calculation on the case description information and the laws in the preset law list, and determining a law related to the case according to the calculation result.

3. The method of claim 1, wherein, The method comprises: inputting the case description information into a pre-trained second large language model for case element parsing to obtain a corresponding case interpretation; wherein the second large language model is trained according to sample case description information and labeled information corresponding to the sample case description information; the case interpretation includes events reflecting case facts, element and attribute information of the elements contained in the events, and relationships between the events.

4. The method of claim 1, wherein, The method comprises: pruning a pre-constructed AOG case knowledge base by using the law related to the case to obtain a target AOG case knowledge base; the AOG case knowledge base is pre-constructed based on the laws in the preset law list; generating a judgment result based on the case description information, the case interpretation, and the target AOG case knowledge base.

5. The method of claim 4, wherein, The method comprises: determining a matching result between the case description information and the case interpretation and content corresponding to a bottom leaf node in the target AOG case knowledge base to obtain a matching result corresponding to the bottom leaf node; in a bottom-up manner, passing the matching result corresponding to the bottom leaf node to an upper node of the bottom leaf node in the target AOG case knowledge base and determining a matching result corresponding to the upper node, until a matching result of a root node in the target AOG case knowledge base is obtained; generating a judgment result by using the matching result of the root node.

6. The method according to claim 4 or 5, characterized in that, The construction of the AOG case knowledge base comprises: identifying and parsing the laws in the preset law list to determine an intermediate node of the AOG case knowledge base; For each intermediate node, based on the analysis of the internal logic of each law in the preset law list, determine the leaf nodes contained by the intermediate node; The judgment result is taken as the root node; Based on the root node, the intermediate node and the leaf node, an AOG case knowledge base is constructed.

7. A method of assisting in the production of a judicial decision, characterized in that, The method comprises: The case element contained in the case description information is analyzed to obtain the corresponding case interpretation; wherein, the case interpretation contains information of the case element related to the given case fact which can affect the case judgment result; the given case fact is the case fact corresponding to the case description information; Based on the case description information, the case interpretation and the pre-constructed AND / OR graph AOG case knowledge base, a judgment result is generated; the AOG case knowledge base is pre-constructed based on the laws in the preset law list.

8. The method of claim 7, wherein, The case element contained in the case description information is analyzed to obtain the corresponding case interpretation, comprising: The case description information is input into the pre-trained second large language model for case element analysis to obtain the corresponding case interpretation; wherein, the second large language model is trained according to sample case description information and sample case description information corresponding to the labeled information; the case interpretation includes events reflecting case facts, elements and attribute information of elements contained in the events, and relationships between events.

9. The method of claim 7, wherein, Based on the case description information, the case interpretation and the pre-constructed AND / OR graph AOG case knowledge base, a judgment result is generated, comprising: Determine the matching result between the case description information and the case interpretation and the content corresponding to the bottom leaf node in the AOG case knowledge base to obtain the matching result corresponding to the bottom leaf node; In a bottom-up manner, the matching result corresponding to the bottom leaf node is transmitted to the upper node of the bottom leaf node in the AOG case knowledge base, and the matching result of the corresponding upper node is determined, until the matching result of the root node in the AOG case knowledge base is obtained; The matching result of the root node is used to generate a judgment result.

10. The method according to any of claims 7-9, characterized by, The construction of the AOG case knowledge base comprises: The laws in the preset law list are identified and analyzed to determine the intermediate nodes of the AOG case knowledge base; For each intermediate node, based on the analysis of the internal logic of each law in the preset law list, determine the leaf nodes contained by the intermediate node; The judgment result is taken as the root node; Based on the root node, the intermediate node and the leaf node, an AOG case knowledge base is constructed.

11. An assisted generation system of judicial decisions, characterized by, The system comprises: A law extraction module for extracting laws applicable to a given case fact from a preset law list based on case description information to obtain case-related laws; the given case fact is the case fact corresponding to the case description information; A first case interpretation module for analyzing the case elements contained in the case description information to obtain the corresponding case interpretation; wherein, the case interpretation contains information of the case element related to the given case fact which can affect the case judgment result; A second case interpretation module for inputting the case description information into a pre-trained second large language model for case element analysis to obtain the corresponding case interpretation; wherein, the second large language model is trained according to sample case description information and sample case description information corresponding to the labeled information; the case interpretation includes events reflecting case facts, elements and attribute information of elements contained in the events, and relationships between events. The first decision generation module is configured to generate a decision result based on the case description information, the law related to the case, and the case interpretation.

12. An assisted generation system of judicial decisions, characterized by, The system comprises: The second case interpretation module is configured to analyze case elements contained in the case description information to obtain corresponding case interpretation; wherein the case interpretation contains information of case elements related to a given case fact and capable of influencing a case decision result; the given case fact is a case fact corresponding to the case description information. The second decision generation module is configured to generate a decision result based on the case description information, the case interpretation, and a pre-constructed AOG case knowledge base; the AOG case knowledge base is pre-constructed based on law in a pre-set law list.

13. An electronic device, comprising: The system comprises a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory complete mutual communication through the communication bus. The memory is configured to store a computer program. The processor is configured to execute the program stored on the memory to implement the method in any one of claims 1-10.

14. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program is executed by the processor to implement the method in any one of claims 1-10.

Images