Method, device, equipment and medium for constructing evidence and fact association graph

Through deep language models and graph neural networks, the evidence and facts are correlated and processed, and the correlation map is generated, which solves the problem of time-consuming, labor-intensive and susceptible to bias in manual operations, realizes efficient analysis of complex data, and improves the accuracy of legal judgments Sex and fairness.

CN118627609BActive Publication Date: 2024-11-12SHENZHEN DIBO ENTERPRISE RISK MANAGEMENT TECH CO LTD +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410963159.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-07-18
Publication Date
2024-11-12
Estimated Expiration
2044-07-18

AI Technical Summary

Technical Problem

In the prior art, the correlation between evidence and facts is dependent on manual operations, which is time-consuming and labor-intensive and susceptible to subjective judgments and cognitive biases, and it is difficult to deal with complex data types, such as electronic evidence and big data, which limits the development of legal practice.

Method used

The deep language model is used to vectorize legal files, and the text features are extracted using the pre-trained recognition model. The evidence and claim vectors are updated in combination with the graph neural network to generate association maps to reduce human errors and biases.

Benefits of technology

Accurately extract the complex relationship between evidence and facts, reduce the interference of human error and prejudice, and improve the accuracy and fairness of legal judgments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118627609B_ABST
    Figure CN118627609B_ABST
Patent Text Reader

Abstract

The present application relates to the field of artificial intelligence technology, and discloses a method, device, equipment and medium for constructing an evidence and fact association map. The method includes: vectorizing legal documents and evidence to be associated to obtain text vectors and evidence vectors; using a pre-trained recognition model to extract information and perform feature recognition processing on the text vector to output a corresponding claim vector; using a pre-trained analysis model to update the attention weight of the litigation relationship graph to obtain the attention weight, and performing correlation scoring processing on the evidence vector and the claim vector based on the attention weight to obtain the correlation scoring result; based on the correlation scoring result, generating a corresponding association map. The embodiment of the present application can accurately extract the complex relationship between the evidence to be associated and the facts of the case, and reduce the interference of human errors and prejudices.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a method, device, equipment and medium for constructing a graph of associations between evidence and facts. Background Art

[0002] Correlating evidence with facts is a core part of judicial practice, requiring judges to conduct in-depth analysis of case-related materials, determine the authenticity of facts, and make fair decisions based on them. This process involves the collection, organization, and evaluation of a large amount of evidence such as documents, confessions, and physical evidence to ensure the accuracy and fairness of legal decisions.

[0003] At present, the method of associating evidence with facts relies on manual operation. The manual analysis process is time-consuming and labor-intensive. Secondly, manual analysis is easily affected by subjective judgment and cognitive bias, which may lead to unfair or inaccurate judgments. In addition, traditional methods have difficulty in processing complex data types, such as electronic evidence and big data, which limits the development of legal practice. Summary of the invention

[0004] The purpose of this application is to provide a method, device, equipment and medium for constructing an evidence and fact association map, which can accurately extract the complex relationship between the evidence to be associated and the case facts, and reduce the interference of human errors and prejudices.

[0005] The present application embodiment provides a method for constructing a graph of evidence and fact association, including:

[0006] Vectorize the legal documents and the evidence to be associated to obtain text vectors and evidence vectors;

[0007] Using a pre-trained recognition model, extract information and perform feature recognition processing on the text vector to output a corresponding claim vector; the claim vector includes legal claim information or factual claim information;

[0008] Using a pre-trained analysis model, the litigation relationship graph is updated with attention weights to obtain attention weights, and the evidence vector and the claim vector are scored for relevance based on the attention weights to obtain relevance scoring results; the litigation relationship graph is composed of the evidence vector and the claim vector, and the pre-trained recognition model and the pre-trained analysis model are obtained by joint training using a gradient descent method and relevance training data, and the relevance training data is training data with label information that characterizes the relevance between the evidence vector and the claim vector;

[0009] According to the correlation scoring result, a corresponding association map is generated.

[0010] In some embodiments, the vectorization of the legal document and the evidence to be associated to obtain the text vector and the evidence vector includes:

[0011] Performing text extraction processing on the legal documents in turn to obtain document texts, and performing vectorization processing on the document texts to obtain text vectors;

[0012] The evidence to be associated is subjected to text conversion processing to obtain evidence text, an identifier is configured for the evidence text and vectorization processing is performed to obtain an evidence vector.

[0013] In some embodiments, the pre-trained recognition model includes a BERT model and a linear mapping layer;

[0014] The method of using the pre-trained recognition model to extract information and perform feature recognition processing on the text vector to output a corresponding proposition vector includes:

[0015] Using the BERT model, based on a multi-head attention mechanism and a feedforward neural network, the text vector is subjected to text feature extraction processing, and a text feature vector is output;

[0016] The linear mapping layer is used to perform linear mapping classification processing on the text feature vector to obtain a classification result of the text feature vector, and a vector conversion representation processing is performed on the text feature vector according to the classification result to output the proposition vector.

[0017] In some embodiments, the use of the pre-trained analysis model to update the attention weight of the litigation relationship graph to obtain the attention weight, and the evidence vector and the claim vector are subjected to relevance scoring according to the attention weight to obtain the relevance scoring result, including:

[0018] Using the evidence vector as a first node, using the claim vector as a second node, and combining the first node and the second node to construct an edge, thereby generating the litigation relationship graph;

[0019] The litigation relationship graph is input into the pre-trained analysis model, and the weight of the edge is updated through the self-attention mechanism to obtain an updated edge; the weight of the updated edge is updated to the attention weight, and the pre-trained analysis model is obtained by training the graph attention network;

[0020] Normalizing the weights of the updated edges to obtain normalized weights;

[0021] An average value of each of the normalized weights is calculated to obtain a correlation scoring result; the correlation scoring result represents the correlation between the first node and the second node.

[0022] In some embodiments, the functional expression of the analysis model is:

[0023] ,

[0024] Among them, E updated is the updated edge, G is the litigation relationship graph, E init is the edge before update, f GAT () is the self-attention weight update method;

[0025] The calculation expression of the correlation score result is:

[0026] ,

[0027] Among them, R is the correlation score result, w e is the weight of the updated edge, and softmax() is the softmax function used for normalization.

[0028] In some embodiments, generating a corresponding association map according to the correlation scoring result includes:

[0029] According to the relevance scoring result, the evidence to be associated is screened and sorted for significant relevance to obtain strongly relevant evidence text;

[0030] Performing text extraction processing on the legal document to obtain a case text; the case text describes the case background and relevant legal issues of the corresponding litigation case;

[0031] Generate a review text based on the relevance score result corresponding to the strongly relevant evidence text; the review text describes the impact of the strongly relevant evidence text on the facts of the litigation case;

[0032] The strongly correlated evidence text, the case text and the commentary text are respectively imported into corresponding positions in a preset graph template to generate the association graph.

[0033] In some embodiments, before vectorizing the legal document and the evidence to be associated to obtain the text vector and the evidence vector, the following step is further included:

[0034] Acquire a recognition model to be trained, an analysis model to be trained, and the correlation training data;

[0035] Vectorize the sample documents and sample evidence to obtain sample text vectors and sample evidence vectors;

[0036] Using the recognition model to be trained, performing information extraction and feature recognition processing on the sample text vector to output a corresponding sample proposition vector;

[0037] Using the analysis model to be trained, performing attention weight update processing on the relationship between the sample evidence vector and the sample claim vector to obtain a sample attention weight, and performing correlation scoring processing on the sample evidence vector and the sample claim vector according to the sample attention weight to obtain a sample correlation scoring result;

[0038] According to the model loss information, the gradient descent method is used to adjust the weight parameters of the recognition model to be trained and the weight parameters of the analysis model to be trained. When the model loss information meets the termination conditions, the pre-trained recognition model and the pre-trained analysis model are obtained; the model loss information characterizes the degree of matching between the correlation training data and the sample correlation scoring results for the correlation description of the sample evidence vector and the sample claim vector.

[0039] The present application also provides a device for constructing a graph of evidence and fact association, including:

[0040] The first module is used to vectorize the legal documents and the evidence to be associated to obtain the text vector and the evidence vector;

[0041] The second module is used to perform information extraction and feature recognition processing on the text vector using a pre-trained recognition model to output a corresponding claim vector; the claim vector includes legal claim information or factual claim information;

[0042] The third module is used to use the pre-trained analysis model to perform attention weight update processing on the litigation relationship graph to obtain the attention weight, and perform correlation scoring processing on the evidence vector and the claim vector according to the attention weight to obtain the correlation scoring result; the litigation relationship graph is composed of the evidence vector and the claim vector, and the pre-trained recognition model and the pre-trained analysis model are obtained by joint training through the gradient descent method and correlation training data, and the correlation training data is training data with label information representing the correlation between the evidence vector and the claim vector;

[0043] The fourth module is used to generate a corresponding association map based on the correlation scoring result.

[0044] An embodiment of the present application also provides an electronic device, which includes a memory and a processor, wherein the memory stores a computer program, and the processor implements the above-mentioned method for constructing a graph of associations between evidence and facts when executing the computer program.

[0045] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the above-mentioned method for constructing a graph of associations between evidence and facts.

[0046] The beneficial effects of the present application are as follows: by using a deep language model, information extraction and feature recognition are performed on the text vector obtained by vectorizing the legal document to obtain a claim vector containing legal claim information or factual claim information, and by using graph neural networks and other technologies, attention weights are updated on the litigation relationship graph composed of the claim vector and the evidence vector obtained by vectorizing the evidence to be associated to obtain the attention weights, and then the evidence vectors and the claim vectors are correlated and scored based on the attention weights to obtain the correlation scoring results, so as to determine whether there is a significant relationship between each piece of evidence to be associated and each factual claim, or to determine whether there is a significant relationship between each piece of evidence to be associated and each legal claim, and then select the evidence to be associated that has a significant relationship with the relevant factual claims and / or legal claims, and generate a corresponding correlation map, which can accurately extract the complex relationship between the evidence to be associated and the facts of the case and reduce the interference of human errors and prejudices. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is an optional flowchart of the method for constructing a map of associations between evidence and facts provided in an embodiment of the present application.

[0048] Figure 2 It is a flowchart of the specific method of step S103 provided in an embodiment of the present application.

[0049] Figure 3 It is a flowchart of the specific method of step S104 provided in an embodiment of the present application.

[0050] Figure 4 It is a flowchart of the method before step S101 provided in an embodiment of the present application.

[0051] Figure 5 It is an optional structural diagram of the device for constructing the evidence and fact association map provided in the embodiment of the present application.

[0052] Figure 6 It is a schematic diagram of the hardware structure of the electronic device provided in the embodiment of the present application. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0054] It should be noted that, although the functional modules are divided in the device schematic diagram and the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first", "second", etc. in the specification, claims and the above drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0055] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0056] First, some nouns involved in this application are analyzed:

[0057] Artificial intelligence (AI) is a new technical science that studies and develops theories, methods, technologies and application systems for simulating, extending and expanding human intelligence. AI is a branch of computer science that attempts to understand the essence of intelligence and produce a new type of intelligent machine that can respond in a similar way to human intelligence. Research in this field includes robotics, speech recognition, image recognition, natural language processing and expert systems. AI can simulate the information process of human consciousness and thinking. AI is also a theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0058] Natural language processing (NLP): NLP uses computers to process, understand and apply human languages ​​(such as Chinese, English, etc.). NLP is a branch of artificial intelligence and an interdisciplinary subject between computer science and linguistics. It is often referred to as computational linguistics. Natural language processing includes grammatical analysis, semantic analysis, and text comprehension. Natural language processing is often used in technical fields such as machine translation, handwritten and printed character recognition, speech recognition and text-to-speech conversion, information intent recognition, information extraction and filtering, text classification and clustering, and opinion mining. It involves data mining related to language processing, machine learning, knowledge acquisition, knowledge engineering, artificial intelligence research, and linguistic research related to language computing.

[0059] Information Extraction: A text processing technology that extracts specified types of entity, relationship, event and other factual information from natural language text and forms structured data output. Information extraction is a technology that extracts specific information from text data. Text data is composed of some specific units, such as sentences, paragraphs, and chapters. Text information is composed of some small specific units, such as characters, words, phrases, sentences, paragraphs, or a combination of these specific units. Extracting noun phrases, names, place names, etc. from text data are all text information extraction. Of course, the information extracted by text information extraction technology can be various types of information.

[0060] Cross Entropy: It is an important concept in Shannon information theory, mainly used to measure the difference information between two probability distributions. The performance of language models is usually measured by cross entropy and complexity (perplexity). The meaning of cross entropy is the difficulty of text recognition using the model, or from the perspective of compression, how many bits are used to encode each word on average. The meaning of complexity is the average number of branches that the model uses to represent the text, and its reciprocal can be regarded as the average probability of each word. Smoothing refers to assigning a probability value to the unobserved N-gram combination to ensure that the word sequence can always get a probability value through the language model. Commonly used smoothing techniques include Turing estimation, deletion interpolation smoothing, Katz smoothing and Kneser-Ney smoothing.

[0061] Self-Attention Mechanism: It can "dynamically" generate weights for different connections in order to establish long-distance dependencies between input information. The self-attention mechanism is actually a type of attention mechanism. It is also a network configuration. The problem it wants to solve is that the input received by the network is a lot of vectors, and the size of the vectors is also uncertain, such as machine translation (sequence to sequence problem, the machine decides how many tags to use), part-of-speech tagging (Pos tagging one vector corresponds to one tag), semantic analysis (multiple vectors correspond to one tag), and other text processing and image processing.

[0062] At present, the method of associating evidence with facts relies on manual operation. The manual analysis process is time-consuming and labor-intensive. Secondly, manual analysis is easily affected by subjective judgment and cognitive bias, which may lead to unfair or inaccurate judgments. In addition, traditional methods have difficulty in processing complex data types, such as electronic evidence and big data, which limits the development of legal practice.

[0063] Based on this, the embodiments of the present application provide a method, device, equipment and medium for constructing an evidence and fact association map, which can accurately extract the complex relationship between the evidence to be associated and the case facts, and reduce the interference of human errors and prejudices.

[0064] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0065] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0066] The method for constructing a graph of evidence and facts provided in the embodiments of the present application relates to the field of artificial intelligence technology. The method for constructing a graph of evidence and facts provided in the embodiments of the present application can be applied to a terminal, can be applied to a server side, or can be software running in a terminal or a server side. In some embodiments, the terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, etc.; the server side can be configured as an independent physical server, or can be configured as a server cluster or a distributed system composed of multiple physical servers, or can be configured as a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (Content Delivery Network, CDN) and big data and artificial intelligence platforms; the software can be an application that implements a text classification method, etc., but is not limited to the above forms.

[0067] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronic devices, network personal computers (Personal Computer, PC), minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0068] See also Figure 1 , Figure 1 is an optional flow chart of the method for constructing a graph of evidence and facts provided in the embodiments of the present application. In some embodiments of the present application, Figure 1 The method may specifically include but is not limited to steps S101 to S104. Figure 1 These four steps are introduced in detail.

[0069] Step S101, vectorize the legal document and the evidence to be associated to obtain a text vector and an evidence vector.

[0070] It is understandable that legal documents are based on the principle of correctness and clarity. All documents in which the parties to the litigation put forward their own claims, reasons for the litigation and the basis for the litigation, and which give rise to the occurrence of litigation procedures, are divided into five types: civil complaint, civil appeal, criminal private prosecution, criminal appeal, and defense, according to different circumstances. They may also include corresponding court record documents.

[0071] It can be understood that the evidence to be related is the objective and true material used in the litigation to prove the true situation of the case (or capable of proving the true situation of the case). The types or specific forms of the evidence to be related may include: 1. physical evidence and documentary evidence; 2. witness testimony; 3. plaintiff's statement; 4. defendant's confession and defense; 5. appraisal conclusion; 6. investigation and inspection records; 7. audio-visual materials.

[0072] In some embodiments, step S101 specifically includes:

[0073] The legal documents are sequentially subjected to text extraction to obtain the document text, and the document text is vectorized to obtain the text vector; the associated evidence is subjected to text conversion to obtain the evidence text, and an identifier is assigned to the evidence text and vectorized to obtain the evidence vector.

[0074] The legal documents are sequentially subjected to text extraction processing, which may be performed by using optical character recognition (OCR) technology to perform text extraction processing from scanned legal documents to extract corresponding text content to obtain the file text, and then the file text is cleaned and normalized, using a custom cleaning function to remove non-text elements such as headers, footers, special symbols, and perform case unification processing to obtain the preprocessed file text, and the preprocessed file text is vectorized, using the word embedding technology Word2Vec to convert the text into a feature vector of fixed dimension to obtain a text vector.

[0075] For example, let the original set of legal documents be D 1 ,D 2 ,...,D n , the document text set obtained after OCR processing is T 1 ,T 2 ,...,T n , then, for each file text T i (i=1,2,...,n) for cleaning and normalization. The expression of the cleaning function used is T cleaned =f clean (T i ), T cleaned The preprocessed file text is then converted into a fixed-dimensional evidence vector using the word embedding technology Word2Vec. The evidence vector is represented by V text , the expression of the evidence vector is V text =Word2Vec(T cleaned ), Word2Vec() is a word embedding technology method.

[0076] The text conversion processing of the evidence to be associated can be to uniformly collect different forms of evidence to be associated, such as written evidence, confessions, physical evidence, etc., and use corresponding optical character recognition (OCR) technology, speech recognition and / or image processing technology to convert the evidence to be associated into evidence text in text form, and then assign a unique identifier to each piece of evidence to be associated, store its detailed information in text format, and perform feature vectorization processing on the evidence text assigned with the identifier to generate an evidence vector.

[0077] For example, let the set of evidence to be associated be D, and the set of evidence text converted into text form be T evidence , T evidence=Text(D), Text() is a function that converts the evidence to be associated into evidence text. Then, the evidence text T evidence Perform feature vectorization to generate evidence vector V evidence , which can be achieved through a specific feature extraction method, which can be to use entity recognition technology to extract key information and convert it into numerical features, such as V evidence =Word2Vec(T evidence ), Word2Vec() is a word embedding technology method.

[0078] Step S102, using the pre-trained recognition model, performs information extraction and feature recognition processing on the text vector to output a corresponding claim vector.

[0079] The claim vector contains legal claim information or factual claim information.

[0080] Specifically, information extraction and feature recognition processing of text vectors can be performed by using a Transformer-based recognition model to extract the plaintiff's and defendant's legal claim information and factual claim information from the text vector. The recognition model takes the above text vector as input, extracts relevant legal claim information and factual claim information, and outputs a vector representation of the legal claim information and factual claim information, namely, a claim vector.

[0081] In some embodiments, the pre-trained recognition model includes a BERT model and a linear mapping layer, and step S102 specifically includes: using the BERT model, based on a multi-head attention mechanism and a feedforward neural network, performing text feature extraction processing on the text vector, and outputting a text feature vector; using the linear mapping layer, performing linear mapping classification processing on the text feature vector to obtain a classification result of the text feature vector, performing vector conversion representation processing on the text feature vector based on the classification result, and outputting a proposition vector.

[0082] Using the BERT model as the core architecture can effectively handle long-distance dependencies and capture complex semantic information in text vectors. Specifically, add [CLS] characters to the beginning of the text vector and [SEP] characters to the end, pass it into the pre-trained and fine-tuned BERT model, extract features from the text vector through the self-attention mechanism, that is, downsample the key features of the text vector and perform data enhancement, convolution pooling, and linear transformation to obtain the sparse feature representation of the text vector, obtain the downsampled features, perform attention processing on the downsampled features through the multi-head attention mechanism, obtain the initial attention weights, use these initial attention weights to transform the dimension of the text vector to meet the requirements of the classification output result, obtain the text feature vector, pass the text feature vector into the linear mapping layer, normalize the text feature vector, judge and output the classification result of the text feature vector, transform the dimension of the text feature vector according to the classification result, and obtain the corresponding proposition vector.

[0083] Step S103, using the pre-trained analysis model, the litigation relationship graph is updated with attention weights to obtain attention weights, and the evidence vector and the claim vector are scored for relevance based on the attention weights to obtain relevance scoring results.

[0084] Among them, the litigation relationship diagram is composed of the evidence vector and the claim vector, and the pre-trained recognition model and the pre-trained analysis model are obtained by jointly training the gradient descent method and the correlation training data. The correlation training data is training data with label information that characterizes the correlation between the evidence vector and the claim vector.

[0085] It can be understood that the recognition model and the analysis model are jointly trained by the gradient descent method and the correlation training data. The correlation training data can be divided into a training set and a verification set in proportion, and the correlation training data of the training set is input into the recognition model for information extraction and feature recognition processing. The result output by the recognition model is then input into the analysis model to determine the correlation score result of the relationship between the evidence vector and the claim vector. During the training, the principle of the gradient descent method is used to gradually reduce the deviation between the correlation score result and the label information. After meeting the preset deviation threshold or number of training times, the correlation training data of the verification set is used to train the trained recognition model and analysis model to confirm that both the recognition model and the analysis model meet the training end conditions.

[0086] Specifically, evidence vectors and claim vectors are used to form a litigation relationship graph, and the evidence vectors and claim vectors are connected in pairs to initialize the relationship between the evidence vectors and the claim vectors. Then, the litigation relationship graph is input into a pre-trained analysis model, and the pre-trained analysis model is used to update the attention weight of the relationship between the evidence vector and the claim vector in the litigation relationship graph. The self-attention mechanism is used to assign weights to the relationship between the evidence vector and the claim vector, and finally the attention weight is obtained. Then, the correlation score of the relationship between the evidence vector and the claim vector is calculated in the pre-trained analysis model according to the preset correlation score calculation rules to obtain the correlation score result. The correlation score result is used to characterize the significance of the relationship between the evidence vector and the claim vector.

[0087] Step S104, generating a corresponding association map according to the correlation scoring result.

[0088] Specifically, based on the correlation scoring results, determine whether there is a significant relationship between each evidence vector and each principal vector, thereby determining whether there is a significant relationship between each piece of evidence to be associated and each factual claim, or determining whether there is a significant relationship between each piece of evidence to be associated and each legal claim, so as to select the evidence to be associated that has a significant relationship with the relevant factual claims and / or legal claims, import the selected evidence to be associated in the form of text into a preset graph template, and generate a corresponding association graph in combination with other text content related to the litigation case.

[0089] In steps S101 to S104 shown in the embodiment of the present application, a deep language model is used to perform information extraction and feature recognition processing on the text vector obtained by vectorizing the legal document to obtain a claim vector containing legal claim information or factual claim information, and a graph neural network and other technologies are used to perform attention weight update processing on the litigation relationship graph composed of the claim vector and the evidence vector obtained by vectorizing the evidence to be associated to obtain the attention weight, and then the evidence vector and the claim vector are subjected to correlation scoring processing based on the attention weight to obtain the correlation scoring result, so as to determine whether there is a significant relationship between each piece of evidence to be associated and each factual claim, or to determine whether there is a significant relationship between each piece of evidence to be associated and each legal claim, and select the evidence to be associated that has a significant relationship with the relevant factual claims and / or legal claims, and generate a corresponding correlation map, so as to accurately extract the complex relationship between the evidence to be associated and the facts of the case, and reduce the interference of human errors and prejudices.

[0090] See also Figure 2 , Figure 2is a flowchart of a specific method of step S103 provided in an embodiment of the present application. In some embodiments of the present application, step S103 may specifically include but is not limited to steps S201 to S204. Figure 2 These four steps are introduced in detail.

[0091] Step S201, using the evidence vector as the first node, the claim vector as the second node, and combining the first node and the second node to construct an edge to generate a litigation relationship graph.

[0092] Specifically, the litigation relationship graph is defined as G=(V,E), where V is the node set, E is the edge set, and the evidence vector is the first node V. evidence , the proposed vector is the second node V claims , each first node and each second node are added to the node set V, that is, V={V evidence ,V claims}, connecting any first node and any second node is taken as an edge and added to the edge set E, thereby constructing the node set V and the edge set E.

[0093] Step S202: input the litigation relationship graph into the pre-trained analysis model, update the edge weights through the self-attention mechanism, and obtain updated edges.

[0094] Among them, the updated edge weights are updated to attention weights, and the pre-trained analysis model is trained by the graph attention network.

[0095] Before inputting the litigation relationship graph into the pre-trained analysis model, the edge weights may be initialized using a random initialization method to randomly generate an initial weight for each edge. The initial weight for each edge may be expressed as w e_int =RandomInitalize(V evidence ,V claims ), where w e_int is the initial weight of the edge, and RandomInitalize() is the random initialization method.

[0096] Then, the litigation relationship graph is input into the pre-trained analysis model, and the edge weights are updated through the self-attention mechanism. The correlation degree (or weight) between each first node and each second node is iteratively calculated based on the self-attention mechanism, so as to update the edge weights, wherein the edge weights reflect the mutual relationship between each first node and each second node.

[0097] In some embodiments, the functional expression of the analysis model is:

[0098] ,

[0099] Among them, E updated is the updated edge, which contains the correlation information between the first node and the second node after the current update. G is the litigation relationship graph, and E init is the edge before update, f GAT () is the self-attention weight update method.

[0100] Step S203, normalizing the updated edge weights to obtain normalized weights.

[0101] Specifically, the weights of the updated edges are normalized by first using a preset weight matrix to perform weighted fitting processing on each first node and each second node, so as to obtain the feature vectors of each first node and each second node through linear transformation. The feature vector includes a query vector, a key vector, and a value vector. Then, the feature vector is input into the softmax function for normalization processing, and each element of the feature vector is converted into a probability representation to obtain a normalized weight as a probability representation of a factual claim or a legal claim. The sum of each normalized weight is equal to 1. The self-attention mechanism of the analysis model is expressed as:

[0102] ,

[0103] Among them, O is the normalized weight, W Q is the query weight matrix, W K is the key weight matrix, d k is the dimension of the key, W V is the weight matrix of the value, w e is the updated edge weight.

[0104] Step S204, calculating the average value of each normalized weight to obtain a correlation score result.

[0105] The correlation score result represents the correlation between the first node and the second node.

[0106] In some embodiments, the calculation expression of the relevance score result is:

[0107] ,

[0108] Among them, R is the correlation score result, w e is the updated edge weight.

[0109] Therefore, by considering the weights of all edges in the litigation relationship graph, a correlation score result is provided regarding the overall correlation between the evidence to be associated and the factual claims or legal claims. A higher correlation score result indicates that the evidence to be associated has a higher correlation with the factual claims or legal claims. Conversely, a lower correlation score result indicates that the evidence to be associated has a lower correlation with the factual claims or legal claims.

[0110] See also Figure 3 , Figure 3 is a flowchart of a specific method of step S104 provided in an embodiment of the present application. In some embodiments of the present application, step S104 may specifically include but is not limited to steps S301 to S304. Figure 3 These four steps are introduced in detail.

[0111] Step S301, based on the correlation scoring result, the evidence to be associated is screened and sorted for significant correlation to obtain strongly correlated evidence text.

[0112] Specifically, a correlation threshold is preset to determine whether the correlation between the evidence to be associated and the factual claim or the legal claim is significant. The correlation threshold can be determined based on data analysis of previous cases, or set according to the opinions of legal experts. Then, for each correlation scoring result, each correlation scoring result is compared with the preset correlation threshold. If the value of the correlation scoring result is greater than the correlation threshold, it means that the evidence to be associated has a higher correlation with the factual claim or the legal claim, and the evidence to be associated corresponding to the correlation scoring result is strongly correlated evidence. The evidence to be associated is selected, and then the selected evidence to be associated is sorted in descending order according to the value of the correlation scoring result, and the evidence to be associated in text form after descending sorting is output, thereby obtaining a strongly correlated evidence text.

[0113] Step S302, extracting text from the legal document to obtain the case text.

[0114] Among them, the case text describes the case background and relevant legal issues of the corresponding litigation case.

[0115] Step S303, generating a review text according to the correlation score result corresponding to the strongly relevant evidence text.

[0116] Among them, the commentary text describes the impact of strongly related evidence text on the facts of the litigation case.

[0117] Step S304, import the strongly related evidence text, case text and commentary text into the corresponding positions in the preset graph template respectively to generate an association graph.

[0118] Specifically, the graph template may include a case overview part, a key evidence analysis part and a comprehensive evaluation part. The case text is inserted into the text filling area corresponding to the case overview part, the strongly related evidence text is inserted into the text filling area corresponding to the key evidence analysis part, and the commentary text is inserted into the text filling area corresponding to the comprehensive evaluation part, thereby generating an associated graph.

[0119] See also Figure 4 , Figure 4 is a flowchart of the method provided by the embodiment of the present application before step S101. In some embodiments of the present application, before step S101, steps including but not limited to steps S401 to S405 may also be performed. Figure 4 These five steps are introduced in detail.

[0120] Step S401, obtaining a recognition model to be trained, an analysis model to be trained, and correlation training data.

[0121] Step S402, vectorizing the sample document and the sample evidence to obtain a sample text vector and a sample evidence vector.

[0122] Step S403: using the recognition model to be trained, perform information extraction and feature recognition processing on the sample text vector to output a corresponding sample proposition vector.

[0123] Step S404, using the analysis model to be trained, perform attention weight update processing on the relationship between the sample evidence vector and the sample claim vector to obtain the sample attention weight, and perform correlation scoring processing on the sample evidence vector and the sample claim vector based on the sample attention weight to obtain the sample correlation scoring result.

[0124] Step S405, based on the model loss information, the gradient descent method is used to adjust the weight parameters of the recognition model to be trained and the weight parameters of the analysis model to be trained. When the model loss information meets the end condition, the pre-trained recognition model and the pre-trained analysis model are obtained.

[0125] Among them, the model loss information represents the matching degree between the correlation training data and the sample correlation scoring results for the correlation description of the sample evidence vector and the sample claim vector.

[0126] It can be understood that the sample file is essentially the legal document of the above-mentioned embodiment, and the sample evidence is essentially the evidence to be associated in the above-mentioned embodiment. The sample file, sample evidence and correlation training data serve as training materials for the recognition model and the analysis model. The specific processes of step S402, step S403 and step S404 are essentially the same as the specific processes of the above-mentioned step S101, step S102 and step S103, respectively, and will not be repeated in this embodiment.

[0127] In step S405 of some embodiments, in order to train and optimize the recognition model and the analysis model, a model loss information is first defined, and the model loss information is used to measure the difference between the sample relevance score results predicted by the recognition model and the analysis model and the relevance training data. The model loss information can be expressed as:

[0128] ,

[0129] Among them, L is the model loss information, N is the number of training samples, and y i is the i-th relevant training data, Score the sample relevance results.

[0130] The gradient descent method is used to optimize the weights of the recognition model and the analysis model. In each iteration, the gradient of the model loss information relative to the weights of the recognition model and the analysis model is calculated, and the weights of the recognition model and the analysis model are updated. The update formula can be expressed as:

[0131] ,

[0132] Among them, Weights new To identify new weights for a model or to analyze a model, Weights old To identify the model or analyze the old weights of the model, α is the learning rate, ∇L(Weights old ) is the gradient of the model loss information with respect to the old weights of the recognition model or the analysis model.

[0133] See also Figure 5 The embodiment of the present application also provides a device for constructing a graph of association between evidence and facts, which can implement the above-mentioned method for constructing a graph of association between evidence and facts, and the device includes:

[0134] The first module 501 is used to vectorize the legal document and the evidence to be associated to obtain a text vector and an evidence vector;

[0135] The second module 502 is used to perform information extraction and feature recognition processing on the text vector using a pre-trained recognition model to output a corresponding claim vector; the claim vector includes legal claim information or factual claim information;

[0136] The third module 503 is used to use the pre-trained analysis model to perform attention weight update processing on the litigation relationship graph to obtain the attention weight, and perform correlation scoring processing on the evidence vector and the claim vector according to the attention weight to obtain the correlation scoring result; the litigation relationship graph is composed of the evidence vector and the claim vector, and the pre-trained recognition model and the pre-trained analysis model are obtained by joint training with the gradient descent method and the correlation training data, and the correlation training data is training data with label information representing the correlation between the evidence vector and the claim vector;

[0137] The fourth module 504 is used to generate a corresponding association map according to the correlation scoring result.

[0138] The specific implementation of the apparatus for constructing a graph of associations between evidence and facts is substantially the same as the specific implementation of the method for constructing a graph of associations between evidence and facts described above, and will not be described in detail herein.

[0139] Figure 6 It is a block diagram of an electronic device according to an exemplary embodiment.

[0140] Refer to the following Figure 6 The electronic device 600 according to this embodiment of the present disclosure is described. Figure 6 The electronic device 600 shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0141] like Figure 6 As shown, the electronic device 600 is in the form of a general computing device. The components of the electronic device 600 may include, but are not limited to: at least one processing unit 610, at least one storage unit 620, a bus 630 connecting different system components (including the storage unit 620 and the processing unit 610), a display unit 640, etc.

[0142] The storage unit stores program codes, which can be executed by the processing unit 610, so that the processing unit 610 executes the steps according to various exemplary embodiments of the present disclosure described in the above-mentioned evidence and fact association graph construction method section of this specification. For example, the processing unit 610 can execute the following steps: Figure 1 , Figure 2 , Figure 3 and Figure 4 Follow the steps shown in .

[0143] The storage unit 620 may include a readable medium in the form of a volatile storage unit, such as a random access memory unit (RAM) 6201 and / or a cache memory unit 6202 , and may further include a read-only memory unit (ROM) 6203 .

[0144] The storage unit 620 may also include a program / utility 6204 having a set (at least one) of program modules 6205, such program modules 6205 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0145] Bus 630 may represent one or more of several types of bus structures, including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

[0146] The electronic device 600 may also communicate with one or more external devices 600' (e.g., keyboards, pointing devices, Bluetooth devices, etc.), may also communicate with one or more devices that enable a user to interact with the electronic device 600, and / or communicate with any device that enables the electronic device 600 to communicate with one or more other computing devices (e.g., routers, modems, etc.). Such communication may be performed via an input / output (I / O) interface 650. Furthermore, the electronic device 600 may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) via a network adapter 660. The network adapter 660 may communicate with other modules of the electronic device 600 via a bus 630. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 600, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0147] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the above-mentioned method for constructing a graph of associations between evidence and facts.

[0148] The method, device, equipment and storage medium for constructing a graph of evidence and fact associations provided in the embodiments of the present application use a deep language model to perform information extraction and feature recognition processing on the text vector obtained by vectorizing the legal document to obtain a claim vector containing legal claim information or factual claim information, and use graph neural networks and other technologies to perform attention weight update processing on the litigation relationship graph composed of the claim vector and the evidence vector obtained by vectorizing the evidence to be associated to obtain the attention weight, and then perform correlation scoring processing on the evidence vector and the claim vector based on the attention weight to obtain the correlation scoring result, so as to determine whether there is a significant relationship between each piece of evidence to be associated and each factual claim, or to determine whether there is a significant relationship between each piece of evidence to be associated and each legal claim, and select the evidence to be associated that has a significant relationship with the relevant factual claims and / or legal claims, and generate a corresponding association graph, which can accurately extract the complex relationship between the evidence to be associated and the facts of the case and reduce the interference of human errors and prejudices.

[0149] Through the description of the above implementation, it is easy for those skilled in the art to understand that the example implementation described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the implementation of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, or a network device, etc.) to execute the above method according to the implementation of the present disclosure.

[0150] The program product may use any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0151] Computer readable storage media may include data signals propagated in baseband or as part of a carrier wave, wherein readable program codes are carried. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The readable storage medium may also be any readable medium other than a readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, device, or device. The program codes contained on the readable storage medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination thereof.

[0152] Program code for performing the operations of the present disclosure may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, etc., and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a separate software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving a remote computing device, the remote computing device may be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., through the Internet using an Internet service provider).

[0153] Those skilled in the art will appreciate that the above modules can be distributed in the device according to the description of the embodiment, or can be changed accordingly and only used in one or more devices different from the embodiment. The modules of the above embodiments can be combined into one module, or further divided into multiple sub-modules.

[0154] Through the description of the above embodiments, it is easy for those skilled in the art to understand that the example embodiments described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the embodiment of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, and includes several instructions to enable a computing device (which can be a personal computer, a server, a mobile terminal, or a network device, etc.) to execute the method according to the embodiment of the present disclosure.

[0155] The exemplary embodiments of the present disclosure are specifically shown and described above. It should be understood that the present disclosure is not limited to the detailed structures, configurations or implementations described herein; on the contrary, the present disclosure is intended to cover various modifications and equivalent configurations included in the spirit and scope of the appended claims.

Claims

1. A method for constructing a graph of evidence and fact association, characterized in that: include: Vectorize the legal documents and the evidence to be associated to obtain text vectors and evidence vectors; Using a pre-trained recognition model, extract information and perform feature recognition processing on the text vector to output a corresponding claim vector; the claim vector includes legal claim information or factual claim information; Using a pre-trained analysis model, the litigation relationship graph is updated with attention weights to obtain attention weights, and the evidence vector and the claim vector are scored for relevance based on the attention weights to obtain relevance scoring results; the litigation relationship graph is composed of the evidence vector and the claim vector, and the pre-trained recognition model and the pre-trained analysis model are obtained by joint training using a gradient descent method and relevance training data, and the relevance training data is training data with label information that characterizes the relevance between the evidence vector and the claim vector; According to the correlation scoring result, a corresponding association map is generated.

2. The method for constructing a graph of evidence and fact association according to claim 1, characterized in that: The vectorization process of the legal document and the evidence to be associated to obtain the text vector and the evidence vector includes: Performing text extraction processing on the legal documents in turn to obtain document texts, and performing vectorization processing on the document texts to obtain text vectors; The evidence to be associated is subjected to text conversion processing to obtain evidence text, an identifier is configured for the evidence text and vectorization processing is performed to obtain an evidence vector.

3. The method for constructing a graph of evidence and facts according to claim 1, characterized in that: The pre-trained recognition model includes a BERT model and a linear mapping layer; The method of using the pre-trained recognition model to extract information and perform feature recognition processing on the text vector to output a corresponding proposition vector includes: Using the BERT model, based on a multi-head attention mechanism and a feedforward neural network, the text vector is subjected to text feature extraction processing, and a text feature vector is output; The linear mapping layer is used to perform linear mapping classification processing on the text feature vector to obtain a classification result of the text feature vector, and a vector conversion representation processing is performed on the text feature vector according to the classification result to output the proposition vector.

4. The method for constructing a graph of evidence and fact association according to claim 1, characterized in that: The pre-trained analysis model is used to update the attention weight of the litigation relationship graph to obtain the attention weight, and the evidence vector and the claim vector are subjected to relevance scoring according to the attention weight to obtain the relevance scoring result, including: Using the evidence vector as a first node, using the claim vector as a second node, and combining the first node and the second node to construct an edge, thereby generating the litigation relationship graph; The litigation relationship graph is input into the pre-trained analysis model, and the weight of the edge is updated through the self-attention mechanism to obtain an updated edge; the weight of the updated edge is updated to the attention weight, and the pre-trained analysis model is obtained by training the graph attention network; Normalizing the weights of the updated edges to obtain normalized weights; An average value of each of the normalized weights is calculated to obtain a correlation scoring result; the correlation scoring result represents the correlation between the first node and the second node.

5. The method for constructing a graph of evidence and fact association according to claim 4, characterized in that: The functional expression of the analysis model is: , Among them, E updated is the updated edge, G is the litigation relationship graph, E init is the edge before update, f GAT () is the self-attention weight update function; The calculation expression of the correlation score result is: , Among them, R is the correlation score result, w e is the weight of the updated edge, and softmax() is the softmax function used for normalization.

6. The method for constructing a graph of evidence and fact association according to claim 1, characterized in that: The generating a corresponding association map according to the correlation scoring result includes: According to the relevance scoring result, the evidence to be associated is screened and sorted for significant relevance to obtain strongly relevant evidence text; Performing text extraction processing on the legal document to obtain a case text; the case text describes the case background and relevant legal issues of the corresponding litigation case; Generate a review text based on the relevance score result corresponding to the strongly relevant evidence text; the review text describes the impact of the strongly relevant evidence text on the facts of the litigation case; The strongly correlated evidence text, the case text and the commentary text are respectively imported into corresponding positions in a preset graph template to generate the association graph.

7. The method for constructing a graph of evidence and fact association according to any one of claims 1 to 6, characterized in that: Before the legal document and the evidence to be associated are vectorized to obtain the text vector and the evidence vector, the following step is also included: Acquire a recognition model to be trained, an analysis model to be trained, and the correlation training data; Vectorize the sample documents and sample evidence to obtain sample text vectors and sample evidence vectors; Using the recognition model to be trained, performing information extraction and feature recognition processing on the sample text vector to output a corresponding sample proposition vector; Using the analysis model to be trained, performing attention weight update processing on the relationship between the sample evidence vector and the sample claim vector to obtain a sample attention weight, and performing correlation scoring processing on the sample evidence vector and the sample claim vector according to the sample attention weight to obtain a sample correlation scoring result; According to the model loss information, the gradient descent method is used to adjust the weight parameters of the recognition model to be trained and the weight parameters of the analysis model to be trained. When the model loss information meets the termination conditions, the pre-trained recognition model and the pre-trained analysis model are obtained; the model loss information characterizes the degree of matching between the correlation training data and the sample correlation scoring results for the correlation description of the sample evidence vector and the sample claim vector.

8. A device for constructing a graph of evidence and fact association, characterized in that: include: The first module is used to vectorize the legal documents and the evidence to be associated to obtain the text vector and the evidence vector; The second module is used to perform information extraction and feature recognition processing on the text vector using a pre-trained recognition model to output a corresponding claim vector; the claim vector includes legal claim information or factual claim information; The third module is used to use the pre-trained analysis model to perform attention weight update processing on the litigation relationship graph to obtain the attention weight, and perform correlation scoring processing on the evidence vector and the claim vector according to the attention weight to obtain the correlation scoring result; the litigation relationship graph is composed of the evidence vector and the claim vector, and the pre-trained recognition model and the pre-trained analysis model are obtained by joint training through the gradient descent method and correlation training data, and the correlation training data is training data with label information representing the correlation between the evidence vector and the claim vector; The fourth module is used to generate a corresponding association map based on the correlation scoring result.

9. An electronic device, characterized in that: The electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the method for constructing an evidence and fact association map as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for constructing a graph of evidence and facts as described in any one of claims 1 to 7 is implemented.