Hierarchical reasoning chain-driven multi-defendant law decision prediction method and system

Abstract

The invention provides a hierarchical reasoning chain-driven multi-defendant law decision prediction method and system, and relates to the technical field of law decision prediction, and the method comprises the steps: obtaining the information of a to-be-predicted case, including a to-be-determined specific defendant, the names of all defendants and the fact description; inputting case information into the trained hierarchical reasoning network, and predicting judgment results of a specific defendant, including criminal relationships, sentencing plots, legal provisions, criminal names and criminal periods; wherein the hierarchical reasoning network is based on a sequence-to-sequence generation technology, adopts a hierarchical reasoning chain, identifies a criminal relationship between a specific bulletin and other bulletin through two-stage reasoning, determines a sentencing plot of the specific bulletin, and performs forward prediction and backward verification on legal provisions, criminal names and criminal periods of the specific bulletin; according to the method, the multi-defendant judgment process is formalized into the hierarchical reasoning chain, and the judgment result of each defendant in the multi-defendant case is effectively predicted according to the reasoning chain.

Description

Technical Field

[0001] The present disclosure relates to the technical field of legal judgment prediction, and in particular to a method and system for predicting legal judgments for multiple defendants driven by a hierarchical reasoning chain. Background Art

[0002] The goal of legal judgment prediction is to predict the judgment outcome based on the facts of the case, which include legal provisions, applicable charges, sentences, and other factors; currently, this field mainly focuses on the study of single-defendant cases, that is, cases involving only one defendant.

[0003] In practical applications, the single-defendant prediction method has some inevitable limitations. For actual cases, the factual descriptions involving multiple defendants account for a considerable proportion; multiple defendants in the criminal factual descriptions usually have complex interactive relationships, which brings challenges to the legal judgment prediction task. The existing single-defendant prediction method mainly targets single-defendant cases, ignoring the complex interactive relationships between multiple defendants and cannot effectively distinguish the judgment results (legal provisions, charges, and sentences) of each defendant in multi-defendant cases. Summary of the invention

[0004] In order to solve the above problems, the present disclosure proposes a multi-defendant legal judgment prediction method and system driven by a hierarchical reasoning chain, which formalizes the multi-defendant judgment process into a hierarchical reasoning chain, and effectively predicts the judgment result of each defendant in a multi-defendant case based on the reasoning chain.

[0005] According to some embodiments, the present disclosure adopts the following technical solutions:

[0006] A hierarchical reasoning chain driven multi-defendant legal judgment prediction method, including:

[0007] Obtain information about the case to be predicted, including the specific defendant to be sentenced, the names of all defendants, and a description of the facts;

[0008] Input case information into the trained hierarchical reasoning network to predict the verdict of a specific defendant, including criminal relationship, sentencing circumstances, legal provisions, crime, and sentence;

[0009] Among them, the hierarchical reasoning network is based on sequence-to-sequence generation technology and adopts a hierarchical reasoning chain. Through two-level reasoning, it identifies the criminal relationship between a specific defendant and other defendants and determines the sentencing circumstances of the specific defendant, and performs forward prediction and backward verification of the legal provisions, charges, and sentences of the specific defendant.

[0010] According to some embodiments, the present disclosure adopts the following technical solutions:

[0011] A multi-defendant legal judgment prediction system driven by a hierarchical reasoning chain, including an information acquisition module and a judgment prediction module:

[0012] The information acquisition module is configured to: obtain information of the case to be predicted, including the specific defendant to be sentenced, the names of all defendants and description of facts;

[0013] The judgment prediction module is configured to: input case information into the trained hierarchical reasoning network to predict the judgment result of a specific defendant, including criminal relationship, sentencing circumstances, legal provisions, crime, and sentence;

[0014] Among them, the hierarchical reasoning network is based on sequence-to-sequence generation technology and adopts a hierarchical reasoning chain. Through two-level reasoning, it identifies the criminal relationship between a specific defendant and other defendants and determines the sentencing circumstances of the specific defendant, and performs forward prediction and backward verification of the legal provisions, charges, and sentences of the specific defendant.

[0015] According to some embodiments, the present disclosure adopts the following technical solutions:

[0016] A computer program product includes a computer program, which, when executed by a processor, implements the multi-defendant legal judgment prediction method driven by a hierarchical reasoning chain.

[0017] According to some embodiments, the present disclosure adopts the following technical solutions:

[0018] A non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the multi-defendant legal judgment prediction method driven by a hierarchical reasoning chain is implemented.

[0019] According to some embodiments, the present disclosure adopts the following technical solutions:

[0020] An electronic device comprises: a processor, a memory and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory so that the electronic device executes the multi-defendant legal judgment prediction method driven by a hierarchical reasoning chain.

[0021] Compared with the prior art, the present invention has the following beneficial effects:

[0022] 1. In terms of innovation, this invention proposes for the first time a legal judgment method that focuses on predicting multi-defendant cases, and formalizes the multi-defendant judgment process into a task of a hierarchical reasoning chain. This formalized reasoning chain provides new ideas and methods for predicting judgments for different defendants in complex cases.

[0023] 2. At the model level, the present invention constructs a hierarchical reasoning network HR-MLJP, which distinguishes the verdict results of different defendants through a hierarchical reasoning chain. By conducting a systematic hierarchical analysis of the verdict process of each defendant, HR-MLJP effectively solves the complexity and diversity of verdict prediction in multi-defendant cases.

[0024] 3. At the application level, the present invention verifies the effectiveness of the HR-MLJP method on a multi-defendant legal judgment prediction dataset; experimental results show that this method performs well in the prediction of multi-defendant cases, can provide legal workers with more accurate judgment prediction references, and improve the efficiency and accuracy of legal judgment work. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The accompanying drawings constituting a part of the present disclosure are used to provide a further understanding of the present disclosure. The illustrative embodiments of the present disclosure and their descriptions are used to explain the present disclosure and do not constitute an improper limitation on the present disclosure.

[0026] Figure 1 This is a hierarchical reasoning network framework diagram of an embodiment of the present disclosure. DETAILED DESCRIPTION

[0027] The present disclosure is further described below in conjunction with the accompanying drawings and embodiments.

[0028] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present disclosure belongs.

[0029] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present disclosure. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0030] Example 1

[0031] In one embodiment of the present disclosure, a multi-defendant legal judgment prediction method driven by a hierarchical reasoning chain is provided, comprising:

[0032] Obtain information about the case to be predicted, including the specific defendant to be sentenced, the names of all defendants, and a description of the facts;

[0033] Input case information into the trained hierarchical reasoning network to predict the verdict of a specific defendant, including criminal relationship, sentencing circumstances, legal provisions, crime, and sentence;

[0034] Among them, the hierarchical reasoning network is based on sequence-to-sequence generation technology and adopts a hierarchical reasoning chain. Through two-level reasoning, it identifies the criminal relationship between a specific defendant and other defendants and determines the sentencing circumstances of the specific defendant, and performs forward prediction and backward verification of the legal provisions, charges, and sentences of the specific defendant.

[0035] As an embodiment, in order to extract information such as the relationship between defendants and sentencing circumstances, and to solve the problem of confusion between criminal facts and legal provisions among multiple defendants, the present disclosure proposes a new multi-defendant sentence prediction method. Based on a hierarchical reasoning network, the multi-defendant sentence process is formalized into a hierarchical reasoning chain, and the sentence result of each defendant in the multi-defendant case is effectively predicted based on the reasoning chain, namely, the criminal relationship, sentencing circumstances, legal provisions, crime, and sentence. The specific implementation process is explained below.

[0036] The hierarchical reasoning chain driven multi-defendant legal judgment prediction method proposed in this embodiment models the reasoning process based on sequence-to-sequence generation technology. The hierarchical reasoning chain contains two levels. The first level of reasoning is used for intermediate tasks, namely, identifying the criminal relationship between a specific defendant and other defendants and determining the sentencing circumstances of the specific defendant. The second level of reasoning is used for judgment prediction tasks, namely, forward prediction and backward verification of the legal provisions, charges, and sentences of the specific defendants. Finally, the decoder fusion training technology is applied to deal with the problem that the average length of the factual descriptions of multiple defendants may exceed the encoder length limit.

[0037] 1. Sequence to Sequence Generation

[0038] From the perspective of sequence-to-sequence (Seq2Seq) generation, each task is modeled as a conditional probability maximization problem of predicting the optimal label sequence y based on the fact description, the specific defendant name and the specific task description. The calculation formula is as follows:

[0039]

[0040] Among them, m represents the length of the label sequence, and the specific task description d provides semantic clues for Seq2Seq to perform the required tasks.

[0041] In order to complete the Seq2Seq generation task, the Seq2Seq language model mT5 is used, which is specifically described as follows:

[0042]

[0043] Among them, ENC represents the encoder of the language model, DEC represents the decoder of the language model, For prediction results consisting of words, when constructing the input of the encoder, this embodiment uses a special [SEP] tag to separate different input information.

[0044] 2. Hierarchical Reasoning Chain

[0045] In order to distinguish different verdicts for different defendants, this embodiment uses a hierarchical reasoning network HR-MLJP to determine the criminal relationship, sentencing circumstances, legal provisions, crime, and sentence of each defendant in a hierarchical reasoning chain manner.

[0046] like Figure 1 As shown, the hierarchical reasoning chain consists of two levels:

[0047] (1) The first level of reasoning is used for intermediate tasks.

[0048] The first intermediate task is to first describe the facts x, the specific defendant e to be sentenced, and the crime relationship prediction task description d r , to determine the relationship between other defendants and the specific defendant e The specific description is as follows:

[0049]

[0050] Among them, the crime relationship prediction task description d r is “what is the criminal relationship between defendant C and others”, defendant C is the specific defendant e to be sentenced, where e∈E, E is the set of all defendants in this case.

[0051] The second intermediate task is to predict the criminal relationship based on the fact description x, the specific defendant e to be sentenced, and the criminal relationship prediction result And the sentencing prediction task description d s , determine the sentencing circumstances of the specific defendant e, namely:

[0052]

[0053] Among them, the sentencing circumstances prediction task description d s The question is "What are the sentencing circumstances for defendant C?"

[0054] (2) The second level of reasoning is used for decision prediction tasks.

[0055] The second-level reasoning chain includes the forward prediction process and the backward verification process.

[0056] Forward prediction process, based on fact description x, specific defendant e to be sentenced, and first-level reasoning results And the forward prediction task description d lct , predict the legal provisions, crime, and sentence in turn, and the specific expressions are as follows:

[0057]

[0058] Among them, the forward prediction task description d lct is "What is the legal provision, crime and sentence of defendant C?". The backward verification process is based on the fact description x, the specific defendant e to be sentenced, and the first-level reasoning result. And the backward verification task description d tcl , verify these decisions in reverse order, namely:

[0059]

[0060] Among them, the backward verification task description d tcl "What is the sentence, crime and law for defendant C?"

[0061] 3. Decoder fusion training

[0062] In order to address the problem that the average length of the factual descriptions of multiple defendants may exceed the encoder length limit, this embodiment adopts a decoder fusion training method (fusion in decoder, FID) to train the hierarchical reasoning network HR-MLJP.

[0063] The FID method is used to encode multiple paragraphs segmented from the fact description. Specifically, the fact description x is first segmented into K paragraphs, each paragraph containing M words; then, each paragraph is input into the encoder for encoding, and the multiple paragraph encoding representations from the encoder are merged. The decoder generates prediction results by focusing on these paragraph representations, as shown below:

[0064]

[0065] Among them, h i represents the encoding representation of the i-th paragraph in the fact description x; since all tasks are modeled as sequence-to-sequence generation tasks, the standard maximum likelihood method is used to train the model and calculate the cross entropy loss for each task. The calculation formula of the overall loss function is as follows:

[0066]

[0067] Among them, the hyperparameter λ determines the trade-off between all subtask losses. and They represent the cross entropy losses of the crime relationship prediction task, sentencing circumstances prediction task, forward prediction process, and backward validation process respectively; during testing, greedy decoding is used to generate prediction results.

[0068] Example 2

[0069] In one embodiment of the present disclosure, a multi-defendant legal judgment prediction system driven by a hierarchical reasoning chain is provided, including an information acquisition module and a judgment prediction module:

[0070] The information acquisition module is configured to: obtain information of the case to be predicted, including the specific defendant to be sentenced, the names of all defendants and description of facts;

[0071] The judgment prediction module is configured to: input case information into the trained hierarchical reasoning network to predict the judgment result of a specific defendant, including criminal relationship, sentencing circumstances, legal provisions, crime, and sentence;

[0072] Among them, the hierarchical reasoning network is based on sequence-to-sequence generation technology and adopts a hierarchical reasoning chain. Through two-level reasoning, it identifies the criminal relationship between a specific defendant and other defendants and determines the sentencing circumstances of the specific defendant, and performs forward prediction and backward verification of the legal provisions, charges, and sentences of the specific defendant.

[0073] Example 3

[0074] In one embodiment of the present disclosure, a computer program product is provided, including a computer program, which, when executed by a processor, implements the multi-defendant legal judgment prediction method driven by a hierarchical reasoning chain.

[0075] Example 4

[0076] In one embodiment of the present disclosure, a non-transitory computer-readable storage medium is provided, wherein the non-transitory computer-readable storage medium is used to store computer instructions, and when the computer instructions are executed by a processor, the multi-defendant legal judgment prediction method driven by a hierarchical reasoning chain is implemented.

[0077] Example 5

[0078] In one embodiment of the present disclosure, an electronic device is provided, including: a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device executes the multi-defendant legal judgment prediction method driven by a hierarchical reasoning chain.

[0079] The present disclosure is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present disclosure. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0080] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0081] Although the above describes the specific implementation methods of the present disclosure in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present disclosure. Technical personnel in the relevant field should understand that on the basis of the technical solution of the present disclosure, various modifications or variations that can be made by those skilled in the art without creative work are still within the scope of protection of the present disclosure.

Claims

1. A hierarchical reasoning chain driven multi-defendant legal judgment prediction method, characterized by: include: Obtain information about the case to be predicted, including the specific defendant to be sentenced, the names of all defendants, and a description of the facts; Input case information into the trained hierarchical reasoning network to predict the verdict of a specific defendant, including criminal relationship, sentencing circumstances, legal provisions, crime, and sentence; Among them, the hierarchical reasoning network is based on sequence-to-sequence generation technology and adopts a hierarchical reasoning chain. Through two-level reasoning, it identifies the criminal relationship between a specific defendant and other defendants and determines the sentencing circumstances of the specific defendant, and performs forward prediction and backward verification of the legal provisions, charges, and sentences of the specific defendant.

2. A hierarchical reasoning chain driven multi-defendant legal judgment prediction method as claimed in claim 1, characterized in that: The two-level reasoning is specifically as follows: The first level of reasoning includes two tasks: identifying the criminal relationship between a specific defendant and other defendants and determining the sentencing circumstances of the specific defendant; The second-level reasoning includes two tasks: forward prediction of the legal provisions, charges, and sentences of specific defendants and backward verification of the legal provisions, charges, and sentences of specific defendants.

3. A hierarchical reasoning chain driven multi-defendant legal judgment prediction method as claimed in claim 2, characterized in that: The task is executed by using sequence-to-sequence generation technology, adopting Seq2Seq language model mT5, and generating prediction results based on the input task description.

4. A hierarchical reasoning chain driven multi-defendant legal judgment prediction method as claimed in claim 3, characterized in that: The identification of the criminal relationship between a specific defendant and other defendants, the specific input is a fact description, the names of all defendants and a description of the criminal relationship prediction task, and the criminal relationship between the specific defendant and other defendants is determined through a language model; Among them, the criminal relationship prediction task is described as "what is the criminal relationship between defendant C and others", and defendant C is a specific defendant.

5. A hierarchical reasoning chain driven multi-defendant legal judgment prediction method as claimed in claim 3, characterized in that: The specific input for determining the sentencing circumstances of a specific defendant is a fact description, a specific defendant, a crime relationship prediction result, and a sentencing circumstances prediction task description, and the sentencing circumstances of a specific defendant are determined through a language model; Among them, the sentencing circumstances prediction task is described as "What are the sentencing circumstances of defendant C?", and defendant C is a specific defendant.

6. A hierarchical reasoning chain driven multi-defendant legal judgment prediction method as claimed in claim 3, characterized in that: The forward prediction of the legal provisions, charges, and sentences of specific defendants is specifically inputted as fact description, specific defendant, first-level reasoning result, and forward prediction task description, and the legal provisions, charges, and sentences of specific defendants are sequentially predicted through the language model; The forward prediction task is described as "What are the legal provisions, charges and sentences for defendant C?", where defendant C is a specific defendant.

7. A hierarchical reasoning chain driven multi-defendant legal judgment prediction method as claimed in claim 3, characterized in that: The backward verification of the legal provisions, charges, and sentences of the specific defendants is specifically inputted as fact description, specific defendants, first-level reasoning results, and backward verification task descriptions, and the judgment result is verified in the order of the specific defendants’ sentences, charges, and legal provisions through the language model; The backward verification task is described as "What is the sentence, crime and legal provision of defendant C?", where defendant C is a specific defendant.

8. A multi-defendant legal judgment prediction system driven by a hierarchical reasoning chain, characterized in that: include: The information acquisition module is configured to: obtain information of the case to be predicted, including the specific defendant to be sentenced, the names of all defendants and description of facts; The judgment prediction module is configured to: input case information into the trained hierarchical reasoning network to predict the judgment result of a specific defendant, including criminal relationship, sentencing circumstances, legal provisions, crime, and sentence; Among them, the hierarchical reasoning network is based on sequence-to-sequence generation technology and adopts a hierarchical reasoning chain. Through two-level reasoning, it identifies the criminal relationship between a specific defendant and other defendants and determines the sentencing circumstances of the specific defendant, and performs forward prediction and backward verification of the legal provisions, charges, and sentences of the specific defendant.

9. A non-transitory computer-readable storage medium, characterized in that: The non-transitory computer-readable storage medium is used to store computer instructions, and when the computer instructions are executed by the processor, a hierarchical reasoning chain-driven multi-defendant legal judgment prediction method as described in any one of claims 1-7 is implemented.

10. An electronic device, characterized in that: include: A processor, a memory and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement a hierarchical reasoning chain driven multi-defendant legal judgment prediction method as described in any one of claims 1-7.

Images