Multivariate evidence correlation analysis method in judicial field

Abstract

The invention relates to a multivariate evidence correlation analysis method in the judicial field, and belongs to the technical field of intelligent processing of documents. The method comprises the following steps: loading judicial case file data and extracting an evidence content text; constructing an evidence rule template according to judicial domain knowledge and expert opinions; extracting evidence elements based on the rule template and constructing an evidence library; building a multivariate evidence information network, calculating the condition probability of association information among evidence elements based on a Bayesian network, and forming a multivariate evidence association analysis model; optimizing the multivariate evidence information network through a random strategy reinforcement learning algorithm; and calculating an information probability value of the multivariate evidence subnet, and selecting the most valuable multivariate evidence information network. According to the method, a multivariate evidence correlation analysis event of a judicial case is converted into an updating and optimizing task of a multivariate evidence information network, the complex correlation problem among multivariate evidences is simplified, a correlation model among the multivariate evidences is given, and certain auxiliary support is provided in a judicial decision support system.

Description

Technical Field

[0001] This invention belongs to the field of document intelligent processing technology, specifically relating to a method for multi-evidence correlation analysis in the judicial field. Background Technology

[0002] With the increasing application of new-generation information technologies such as big data and artificial intelligence in various industries, investigating, trying, and adjudicating various cases from massive amounts of data in the judicial field has become an important task in the construction of smart justice in my country, and also a future development trend in this field.

[0003] In the course of judicial proceedings, faced with a large amount of diverse evidence, judicial staff often struggle to extract key evidentiary information. Furthermore, diverse evidence often suffers from low value density and poor relevance. Quickly and effectively obtaining relevant evidentiary information from redundant and complex evidence documents is a major challenge for judicial staff. Therefore, the analysis of the relevance of diverse evidence is particularly important for judicial staff, especially in cases involving multiple types of evidence, where the evidence is diverse, scattered, and concealed. For example, in cases involving multiple types of evidence, the relevance and complexity of the evidence place higher technical demands on judicial staff.

[0004] The analysis and organization of diverse evidence is inherently labor-intensive and inefficient. Therefore, leveraging big data and artificial intelligence—new-generation information technologies—to perform correlation analysis of diverse evidence and obtain key information between them becomes essential. This reduces the workload of judicial staff, improves case-handling efficiency, and empowers the construction of smart justice.

[0005] Currently, domestic research on the correlation analysis of multiple types of evidence is still in its early stages, focusing primarily on information extraction from legal documents and intelligent writing of judgments. This invention represents a new exploration in the field of multiple types of evidence. The judicial field was chosen for this research because: firstly, judicial evidence is highly digitized, facilitating intelligent processing; and secondly, with the advancement of the rule of law in my country, evidence in cases is becoming increasingly diverse and complex. Therefore, to address the issue of evidence correlation in the judicial field, this invention employs a Bayesian network model and a stochastic policy reinforcement learning algorithm to achieve correlation analysis of multiple types of evidence. Summary of the Invention

[0006] (a) Technical problems to be solved

[0007] The technical problem to be solved by this invention is how to provide a method for multi-source evidence correlation analysis in the judicial field, so as to solve the problem of effectively obtaining and analyzing evidence information through technical means in cases involving multiple sources of evidence and complex correlations in the judicial field.

[0008] (II) Technical Solution

[0009] To address the aforementioned technical problems, this invention proposes a method for analyzing the correlation of multiple pieces of evidence in the judicial field, which includes the following steps:

[0010] Step 1: Load the judicial case file data and extract the text content of various types of evidence documents; these include: documentary evidence, physical evidence, witness testimony, expert opinions, and statements from the parties involved.

[0011] Step 2: Based on legal knowledge and expert opinions, construct a table of evidence trigger words and evidence elements, and form rule templates for various types of evidence;

[0012] Step 3: Based on various evidence rule templates, extract evidence elements from the text content of evidence documents, construct an evidence database, and optimize the evidence rule templates and element tables according to the iterative updates of evidence content and elements.

[0013] Step 4: Construct a multi-evidence information network, where each node represents an evidence element, and the edges between nodes represent the conditional probabilities of the association information between two evidence elements; based on the prior knowledge of the Bayesian network, calculate the conditional probabilities of the association information between various types of evidence elements to form a multi-evidence association analysis model.

[0014] Step 5: Automatically search for key evidence points in the multivariate evidence information network using the Monte Carlo policy gradient algorithm, obtain reward values, and update the policy to achieve the updating and optimization of the multivariate evidence information network.

[0015] Step Six: Obtain multi-evidence subnets from the multi-evidence information network by enumeration, calculate the information probability value of each subnet, and select the multi-evidence subnet with the highest probability value as the most valuable multi-evidence information network.

[0016] (III) Beneficial Effects

[0017] This invention proposes a method for multi-faceted evidence correlation analysis in the judicial field. Its innovation lies in introducing artificial intelligence technology into multi-faceted evidence analysis in the judicial field. It initializes a multi-faceted evidence information network model using a Bayesian network, updates and optimizes the model through stochastic policy reinforcement learning, and incorporates the subjective initiative of judicial personnel into the stochastic policy reinforcement learning process, enabling the multi-faceted evidence information network model to learn autonomously from historical data. Simultaneously, the model itself has experimental properties; based on the model results, it can be generalized to the entire judicial field and further extended to incorporate deep learning algorithms to better serve the judicial system.

[0018] This invention transforms the analysis of multiple evidence correlations in judicial cases into the task of updating and optimizing a multiple evidence information network, simplifies the complex correlation problem between multiple pieces of evidence, and provides a correlation model between multiple pieces of evidence. The model has generalization ability and flexibility for general judicial cases, and provides certain auxiliary support in judicial decision support systems. Attached Figure Description

[0019] Figure 1 This is a flowchart of the method for analyzing the correlation of multiple pieces of evidence in the judicial field according to the present invention;

[0020] Figure 2 The flowchart shows the multivariate evidence association analysis method based on Bayesian network model and Monte Carlo policy gradient reinforcement learning algorithm of the present invention.

[0021] Figure 3 This is a schematic diagram of the structure of the multi-source evidence information network of the present invention;

[0022] Figure 4 This is a graphical model of a Markov decision process, where dashed lines represent the decision process. Detailed Implementation

[0023] To make the objectives, contents, and advantages of the present invention clearer, the specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples.

[0024] This invention relates to the field of intelligent technology applications in judicial documents, and is mainly a multivariate evidence association analysis method based on Bayesian network models and stochastic policy reinforcement learning algorithms. It belongs to the application technology that combines natural language processing with smart judicial business.

[0025] This invention aims to address the problem of effectively acquiring and analyzing evidentiary information through technical means in cases involving multiple sources of evidence and complex relationships in the judicial field. It proposes a multivariate evidence correlation analysis method based on Bayesian network models and stochastic policy reinforcement learning algorithms. Bayesian networks are directed acyclic graph (DAG) models, a product of the combination of probability theory and graph theory. On the one hand, they can intuitively reveal the structure of a problem using the language of graph theory; on the other hand, they can analyze and utilize the problem structure according to the principles of probability theory. It is a language that systematically describes the relationships between random variables, simulating the uncertainty of causal relationships in human reasoning. Reinforcement learning is another field of machine learning that focuses on how an agent takes actions in an environment to maximize a certain cumulative reward. That is, given data, it learns how to choose a series of actions to maximize long-term gains. The Monte Carlo policy gradient (REINFORCE) algorithm is a stochastic policy parameter learning method where the policy parameters are updated using gradient ascent. Its goal is to maximize performance metrics, thereby achieving the purpose of reinforcement learning.

[0026] The technical solution described in this invention requires the following steps:

[0027] Step 1: Load the judicial case file data and extract the text content of various evidentiary documents such as documentary evidence, physical evidence, witness testimonies, expert opinions, and party statements;

[0028] Step 2: Based on legal knowledge and expert opinions, construct a table of evidence trigger words and evidence elements, and form rule templates for various types of evidence such as documentary evidence, physical evidence, witness testimony, expert opinions, and party statements;

[0029] Step 3: Based on various evidence rule templates, extract evidence elements from the evidence content, construct an evidence database, and optimize the evidence rule templates and element tables according to the iterative updates of the evidence content and elements.

[0030] Step 4: Construct a multi-evidence information network, where each node represents an evidence element, and the edges between nodes represent the conditional probabilities of the association information between two evidence elements; based on the prior knowledge of the Bayesian network, calculate the conditional probabilities of the association information between various types of evidence elements to form a multi-evidence association analysis model.

[0031] Step 5: Automatically search for key evidence points in the multivariate evidence information network using a stochastic policy reinforcement learning algorithm (such as the Monte Carlo policy gradient algorithm), obtain reward values, and update the policy to achieve the updating and optimization of the multivariate evidence information network model.

[0032] Step 6: Obtain multi-evidence subnets from the multi-evidence information network by enumeration, calculate the information probability value (information value) of each subnet, and select the multi-evidence subnet with the highest probability value as the most valuable multi-evidence information network.

[0033] In one embodiment of the method for multi-evidence correlation analysis in the judicial field, this invention loads judicial case files and extracts the text content of various evidentiary documents such as documentary evidence, physical evidence, witness testimonies, expert opinions, and party statements.

[0034] In one embodiment of the present invention, a method for analyzing the correlation of multiple types of evidence in the judicial field, evidence trigger words and evidence element tags are collected and organized based on judicial knowledge and expert opinions. Based on the evidence trigger words and evidence elements, rule templates for various types of evidence, such as documentary evidence, physical evidence, witness testimony, expert opinions, and party statements, are constructed.

[0035] In one embodiment of the present invention, a method for analyzing the correlation of multiple pieces of evidence in the judicial field, evidence elements are extracted from the corresponding evidence content based on various evidence rule templates and an evidence database is constructed. The evidence rule templates are iteratively updated according to the evidence content and evidence elements to form a complete rule template and evidence element table, which is convenient to be extended to general judicial cases.

[0036] In one embodiment of the multi-evidence correlation analysis method in the judicial field, this invention constructs a multi-evidence information network, where each node represents an evidentiary element, and the edges between nodes represent the conditional probabilities of correlation information between two evidentiary elements. Based on the prior knowledge of Bayesian networks, the conditional probabilities of correlation information between various types of evidentiary elements are calculated to form a multi-evidence correlation analysis model.

[0037] In one embodiment of the present invention, a method for analyzing the correlation of multiple evidences in the judicial field, the present invention uses a randomized policy reinforcement learning algorithm (such as the Monte Carlo policy gradient algorithm) to automatically search for key evidence points in a multiple evidence information network, obtain reward values ​​and update the policy, thereby realizing the updating and optimization of the multiple evidence information network model.

[0038] In one embodiment of the multi-evidence association analysis method in the judicial field, the present invention uses an enumeration method to obtain multi-evidence subnets on the multi-evidence information network, calculates the information probability value (information value) of each subnet, and selects the multi-evidence subnet with the highest probability value as the most valuable multi-evidence information network.

[0039] This invention provides a method for multi-faceted evidence correlation analysis in the judicial field, belonging to the application technology of combining natural language processing and smart judicial business. The method includes: loading judicial case file data and extracting the text of evidence content; constructing evidence rule templates based on judicial knowledge and expert opinions; extracting evidence elements based on the rule templates and constructing an evidence database; building a multi-faceted evidence information network, calculating the conditional probabilities of correlation information between evidence elements based on Bayesian networks, forming a multi-faceted evidence correlation analysis model; optimizing the multi-faceted evidence information network through a stochastic strategy reinforcement learning algorithm; calculating the information probability values ​​of the multi-faceted evidence subnets, and selecting the most valuable multi-faceted evidence information network.

[0040] Example 1:

[0041] To more clearly illustrate the technical solution and advantages of the present invention, the technical solution of the present invention will be further described below in conjunction with the accompanying drawings and embodiments:

[0042] S1: Retrieve case files of judicial cases from the unified business system;

[0043] S2: Based on the case file catalog subsystem, extract various types of evidentiary documents such as documentary evidence, physical evidence, witness testimonies, expert opinions, and statements of the parties involved;

[0044] S3: Extract the text content from the corresponding evidence document, and represent it as follows: ,in Indicate the type of evidence, specifically:

[0045] Witness Testimony: {

[0046] "Witness Zhang's Testimony Record": "",

[0047] "Witness Zhang Mou2's Testimony Record": ""},

[0048] "Statement from the parties involved": {

[0049] "Wang": "",

[0050] "Li Mou": ""},

[0051] …}

[0052] S4: Collect and organize evidence trigger words and evidence element tags based on legal knowledge and expert opinions;

[0053] S5: Based on the results collected and organized in step S4, construct an evidence rule template, represented as follows: Specifically:

[0054] Witness Testimony: {

[0055] "Loan details": {

[0056] Borrower: Wang Mou

[0057] Loan Amount: xx yuan

[0058] "Loan Date": "Year xx Month xx Day"},

[0059] "Company Registration": {

[0060] Registrant: "Wang Mou",

[0061] "Registered Amount": "xx yuan",

[0062] Registration Date: [Year / Month / Day]

[0063] "Statement from the parties involved": {

[0064] "Amount defrauded": {

[0065] Victim: Zhao Mou

[0066] "Amount defrauded": "xx yuan",

[0067] "Fraudulent claims": "",

[0068] "Date of the scam": "Year xx Month xx Day"

[0069] "Company Financial Records": {

[0070] "Transferor": "Wang Mou",

[0071] "Transferee": "Li Mou",

[0072] "Destination of Transfer": "",

[0073] "Transfer Amount": "xx yuan",

[0074] "Transfer-out date": "xx year xx month xx day"},

[0075] …}

[0076] S6: Through the text content of evidence documents Evidence Rules Template Obtain evidence element values superscript The subscript indicates the type of evidence, and the subscript indicates the evidentiary element under the corresponding evidentiary type. An evidence database is then constructed. ;

[0077] S7: Update the evidence trigger words and evidence element tags in step S4 based on the iteration results of steps S3, S5, and S6, thereby improving the evidence rule template. ;

[0078] S8: Based on the evidence library of step S6 Establish a multi-source evidence information network , It is a directed acyclic graph, and its expression is:

[0079]

[0080] in, yes The set of nodes, i.e., the evidence repository. Each node represents an element of evidence. , yes The set of edges represents the directions and probabilities between pieces of evidence;

[0081] edge set The association constraints are established based on the prior knowledge of the Bayesian network. It is a causal relationship, represented by arrows, and its value is calculated using the conditional probability of the Bayesian network. If two nodes are connected by a single arrow, it means one node is the "cause" and the other is the "effect," and the two nodes will generate a conditional probability value. (Assume evidence...) Directly affects the evidence ,Right now Then use evidence Pointing evidence Arrows establish evidence Evidence Directed arc The weights, i.e., the connection strength, are expressed using conditional probabilities. To indicate, that is:

[0082]

[0083] This allows us to calculate the conditional probabilities among all multiple pieces of evidence. ,in, Indicate the elements of evidence, Indicate the type of evidence, Representing the types of evidence respectively Different evidentiary elements, as a multi-faceted evidentiary information network The initial connection weights are used to form the initial model for multivariate evidence association analysis;

[0084] S9: The Monte Carlo (MC) method is a way to solve for the optimal policy in reinforcement learning. MC methods do not require all information about the environment; they can learn based solely on past experience. They can achieve good results by learning from experience with very little prior knowledge of the environment. The Monte Carlo policy gradient algorithm belongs to the MC method, where the policy parameter θ represents the policy function π used to model the policy. θ In the context of the multi-evidence information network of this invention, the state-action features of (a|s) can be composed of the conditional probability, causal strength, and node features of the evidence elements, and θ is the weight parameter applied to these features. The update method for the policy parameter θ is gradient ascent, with the goal of maximizing the performance index. .

[0085] The role of Bayesian networks is to provide the initial causal structure and conditional probability distribution among nodes, which is used to define the state space, state transitions, and reward functions. Monte Carlo methods, on the other hand, use trajectory sampling and policy updates to search for and optimize evidence subnets with higher joint probabilities.

[0086] S10: Based on the multi-source evidence information network obtained in step S9 For any set of evidence element nodes (i.e., a subset of random variables) .

[0087] in Their joint probability can be obtained by multiplying their respective local conditional probability distributions:

[0088]

[0089] The optimization objective J(θ) of the Monte Carlo policy gradient algorithm is equivalent to finding the joint probability P(G) in this scenario. i The largest subnet is J(θ), so maximizing J(θ) can be viewed as maximizing the joint probability, and then optimizing the joint probability using the Monte Carlo policy gradient algorithm.

[0090] S11: According to the Monte Carlo theorem, the gradient of performance indicators... The calculation yields:

[0091]

[0092] in, This is a parameterized representation of the policy function. For parameter vectors, This represents the cumulative reward value obtained by taking action a in state s. This represents the mathematical expectation of the trajectory sampling.

[0093] Based on the above formula, the Monte Carlo policy gradient update can be written as:

[0094]

[0095] in, For learning rate, As a reward discount factor, This refers to the cumulative reward. In this invention, reward discounts are not considered; therefore, a [specific reward structure is used]. ;

[0096] In reinforcement learning, a trajectory is a sequence of states, actions, and rewards:

[0097]

[0098] Wherein, the initial state s0 is initially distributed, and action a is executed. t Reward r t+1 And transition to the next state s t+1 The cumulative reward of the trajectory is used to update the policy parameter θ. By continuously sampling multiple trajectories and iteratively optimizing θ, the optimal policy eventually converges. .

[0099] The following is the Monte Carlo policy gradient algorithm flow:

[0100]

[0101] S13: Based on steps S9 to S12, the Monte Carlo policy gradient algorithm implements the initial state multi-evidence information network. Interconnection weights The updates, through continuous updates Get state Until convergence This leads to the updating and optimization of the multi-source evidence information network model, forming a new multi-source evidence information network. ;

[0102] S14: Through a multi-source evidence information network All possible multi-source evidence subnets are obtained by enumeration. ,in And calculate the multi-evidence subnet. joint probability The multivariate evidence subnet with the highest output probability value. As the most valuable multi-source evidence information network.

[0103] Beneficial effects:

[0104] The innovation of this invention lies in introducing artificial intelligence technology into the analysis of multiple evidence in the judicial field. It initializes a multiple evidence information network model using a Bayesian network, updates and optimizes the model through stochastic policy reinforcement learning, and incorporates the subjective initiative of judicial personnel into the stochastic policy reinforcement learning process, enabling the multiple evidence information network model to learn autonomously from historical data. Simultaneously, the model itself has experimental properties; based on the model results, it can be generalized to the entire judicial field and further extended with the addition of deep learning algorithms to better serve the judicial system.

[0105] This invention transforms the analysis of multiple evidence correlations in judicial cases into the task of updating and optimizing a multiple evidence information network, simplifies the complex correlation problem between multiple pieces of evidence, and provides a correlation model between multiple pieces of evidence. The model has generalization ability and flexibility for general judicial cases, and provides certain auxiliary support in judicial decision support systems.

[0106] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for analyzing the correlation of multiple pieces of evidence in the judicial field, characterized in that, The method includes the following steps: Step 1: Load the judicial case file data and extract the text content of various types of evidence documents; these include: documentary evidence, physical evidence, witness testimony, expert opinions, and statements from the parties involved. Step 2: Based on legal knowledge and expert opinions, construct a table of evidence trigger words and evidence elements, and form rule templates for various types of evidence; Step 3: Based on various evidence rule templates, extract evidence elements from the text content of evidence documents, construct an evidence database, and optimize the evidence rule templates and element tables according to the iterative updates of evidence content and elements. Step 4: Construct a multi-evidence information network, where each node represents an evidence element, and the edges between nodes represent the conditional probabilities of the association information between two evidence elements; based on the prior knowledge of the Bayesian network, calculate the conditional probabilities of the association information between various types of evidence elements to form a multi-evidence association analysis model. Step 5: Automatically search for key evidence points in the multivariate evidence information network using the Monte Carlo policy gradient algorithm, obtain reward values, and update the policy to achieve the updating and optimization of the multivariate evidence information network. Step Six: Obtain multi-evidence subnets from the multi-evidence information network by enumeration, calculate the information probability value of each subnet, and select the multi-evidence subnet with the highest probability value as the most valuable multi-evidence information network.

2. The method for analyzing the correlation of multiple pieces of evidence in the judicial field as described in claim 1, characterized in that, Step one includes: extracting the text content from the corresponding evidence document, represented as: ,in Indicates the type of evidence.

3. The method for analyzing the correlation of multiple pieces of evidence in the judicial field as described in claim 2, characterized in that, Step two includes: collecting and organizing evidence trigger words and evidence element tags based on judicial knowledge and expert opinions; and constructing an evidence rule template based on the collected and organized results, as follows: .

4. The method for analyzing the correlation of multiple pieces of evidence in the judicial field as described in claim 3, characterized in that, Step three includes: Through the content of the evidence documents Evidence Rules Template Obtain evidence element values superscript The subscript indicates the type of evidence, and the subscript indicates the evidentiary element under the corresponding evidentiary type. An evidence database is then constructed. ; By iteratively updating evidence trigger words and evidence element tags, the evidence rule template can be improved. .

5. The method for analyzing the correlation of multiple pieces of evidence in the judicial field as described in claim 4, characterized in that, Step four includes: According to the evidence library Building a multi-source evidence information network , It is a directed acyclic graph, and its expression is: in, yes The set of nodes, i.e., the evidence repository. Each node represents an element of evidence. , yes The set of edges represents the directions and probabilities between pieces of evidence; edge set The association constraints are established based on the prior knowledge of the Bayesian network. It is a causal relationship, represented by arrows, and its value is calculated using the conditional probability of the Bayesian network. If two nodes are connected by a single arrow, it means one node is the "cause" and the other is the "effect," and the two nodes will generate a conditional probability value. Assuming evidence... Directly affects the evidence ,Right now Then use evidence Pointing evidence Arrows establish evidence Evidence Directed arc The weights, i.e., the connection strength, are expressed using conditional probabilities. To indicate, that is: This allows us to calculate the conditional probabilities among all multiple pieces of evidence. ,in, Indicate the elements of evidence, Indicate the type of evidence, Representing the types of evidence respectively Different evidentiary elements, as a multi-faceted evidentiary information network The initial connection weights are used to form the initial model for multivariate evidence association analysis.

6. The method for analyzing the correlation of multiple pieces of evidence in the judicial field as described in claim 5, characterized in that, In step five, the policy parameter θ of the Monte Carlo policy gradient algorithm represents the policy function π used for modeling. θ In a multi-evidence information network scenario, the state-action features of (a|s) are composed of the conditional probabilities, causal strengths, and node features of the evidence elements. θ represents the weight parameters applied to these features. The policy parameter θ is updated using gradient ascent, with the goal of maximizing performance metrics. ; The role of Bayesian networks is to provide the initial causal structure and conditional probability distribution among nodes, which is used to define the state space, state transitions, and reward functions. Monte Carlo methods, on the other hand, use trajectory sampling and policy updates to search for and optimize evidence subnets with higher joint probabilities.

7. The method for analyzing the correlation of multiple pieces of evidence in the judicial field as described in claim 6, characterized in that, In step five, based on the multi-source evidence information network For any set of evidence element nodes ; in, Their joint probability can be obtained by multiplying their respective local conditional probability distributions: The optimization objective J(θ) of the Monte Carlo policy gradient algorithm is equivalent to finding the joint probability P(G) in this scenario. i The largest subnet is J(θ), so maximizing J(θ) is considered as maximizing the joint probability, and then the joint probability is optimized by the Monte Carlo policy gradient algorithm.

8. The method for analyzing the correlation of multiple pieces of evidence in the judicial field as described in claim 7, characterized in that, In step five, the gradient of performance indicators is determined according to the Monte Carlo theorem. The calculation yields: in, This is a parameterized representation of the policy function. For parameter vectors, This represents the cumulative reward value obtained by taking action a in state s. This represents the mathematical expectation of the trajectory sampling. Based on the above formula, the Monte Carlo policy gradient update can be written as: in, For learning rate, As a reward discount factor, The cumulative reward; In reinforcement learning, a trajectory is a sequence of states, actions, and rewards: Wherein, the initial state s0 is initially distributed, and action a is executed. t Reward r t+1 And transition to the next state s t+1 The cumulative reward of the trajectory is used to update the policy parameter θ. By continuously sampling multiple trajectories and iteratively optimizing θ, the optimal policy eventually converges. .

9. The method for analyzing the correlation of multiple pieces of evidence in the judicial field as described in claim 8, characterized in that, In step five, the Monte Carlo policy gradient algorithm is used to process the initial state of the multivariate evidence information network. Interconnection weights The updates, through continuous updates Get state Until convergence This leads to the updating and optimization of the multi-source evidence information network model, forming a new multi-source evidence information network. .

10. The method for analyzing the correlation of multiple pieces of evidence in the judicial field as described in claim 9, characterized in that, Step six includes: using a multi-source evidence information network All possible multi-source evidence subnets are obtained by enumeration. ,in And calculate the multi-evidence subnet. joint probability The multivariate evidence subnet with the highest output probability value. As the most valuable multi-source evidence information network.

Images