A method, system and medium for predicting a judgment of a judicial case

Abstract

The application discloses a kind of judicial case decision prediction method, system and medium;It is related to natural language processing technical field;The present scheme is improved on the basis of traditional prediction technology, and the fact description feature representation model of the extracted fact description part is obtained based on judicial decision document and content, the content and structural integrity of judicial adjudication document are guaranteed to the greatest extent, and the length of text is reduced, the problem that the original text is not fully loaded due to the length limitation of prediction model input is avoided, on the other hand, the fact description feature representation model is constructed based on xsBert model and BiGRU model, which is suitable for real case scene, and the fact description feature representation is mapped to the charge and law by the full connection layer, to solve the judgment prediction problem existing in multiple charges and law.

Description

Technical Field

[0001] This invention relates to the field of natural language processing technology, specifically to a method, system, and medium for predicting judgments in judicial cases. Background Technology

[0002] In today's era, Natural Language Processing (NLP) technology is experiencing rapid development and is widely used in text analysis, machine translation, chatbots, and other fields. With the rise of deep learning, models such as BERT and GPT have become standard, capable of understanding complex language structures and contexts, significantly improving text generation and comprehension capabilities. Meanwhile, pre-training and fine-tuning strategies allow these models to quickly adapt to various specific tasks, lowering the development threshold. Furthermore, emerging research directions such as cross-language processing and multimodal learning have further expanded the application scenarios of NLP technology, driving advancements in technologies such as intelligent voice assistants and content recommendation systems.

[0003] In the judicial field, natural language processing (NLP) also plays a crucial role. First, NLP technology can automate the processing of large volumes of legal documents and data, significantly reducing the time judges and lawyers spend on document analysis, retrieval, and organization, thus improving case handling efficiency. Second, through intelligent algorithms, NLP technology ensures more accurate analysis of legal provisions and cases, reducing the risk of human error and inconsistencies, and enhancing judicial fairness. Third, NLP technology can transform complex legal language into easily understandable content, making legal information more accessible to the public and enhancing the popularization and transparency of legal knowledge. Furthermore, NLP technology can analyze historical cases and judgments, providing data support and trend predictions to help judges and lawyers make more evidence-based decisions in case hearings and strategy formulation.

[0004] Currently, most of the better models in the field of judicial judgment prediction are designed for single-crime judgment prediction, treating it as a single-label multi-classification problem. Cases involving multiple crimes and legal provisions are directly removed during corpus preprocessing, resulting in insufficient research on cases involving multiple crimes and legal provisions. However, in real-world trials, some cases often involve multiple crimes and may be convicted and sentenced according to multiple legal provisions, even involving concurrent sentencing. Due to the complexity of cases involving multiple crimes and legal provisions, there has been limited research on this topic. Furthermore, current methods using neural networks such as CNNs and RNNs to extract textual features of case facts as input to predict corresponding legal provisions and crime labels have drawbacks. RNNs may face the vanishing gradient problem when processing long texts, leading to insufficient understanding of long-range dependencies; CNNs often struggle to capture global contextual information, focusing primarily on local features. Both are susceptible to input noise, resulting in inaccurate feature extraction and impacting the performance of downstream tasks. Judicial texts fall into the category of long texts, and the aforementioned methods are ineffective for feature extraction from long texts, being highly sensitive to noise and leading to inaccurate judgments. Summary of the Invention

[0005] The technical problem this invention aims to solve is that existing judicial judgment prediction methods lack research on cases involving multiple charges and legal provisions, and the feature extraction of judicial texts is incomplete, resulting in information loss. This invention aims to provide a judicial judgment prediction method, system, and medium. It improves upon traditional prediction techniques by extracting factual description feature representation models from judicial judgment documents and content, maximizing the integrity of the content and structure of judicial judgment documents while reducing text length and avoiding the problem of incomplete original text loading due to input length limitations of prediction models. Furthermore, it constructs factual description feature representation models that fit real-world judgment scenarios based on xsBert and BiGRU models, and maps the factual description feature representations to charges and legal provisions using fully connected layers, thus solving the problem of predicting judgments involving multiple charges and legal provisions.

[0006] This invention is achieved through the following technical solution:

[0007] This solution provides a method for predicting judgments in judicial cases, including:

[0008] The original data of judicial judgment documents is obtained, and the original data is preprocessed to obtain a judgment dataset; the judgment dataset includes a factual description section, a crime determination section, and a legal provision determination section;

[0009] Based on the xsBert model, feature encoding is performed on the fact description part to obtain the fact description semantic encoding information. Based on the BiGRU model, bidirectional feature extraction is performed on the fact description semantic encoding information to obtain the fact description feature representation model.

[0010] A case description label dataset of the cases to be predicted is constructed. The case description label dataset is input into the fact description feature representation model. The fully connected layer maps the fact description feature representation to the crime and legal provisions to obtain the probability value of the judicial case to be predicted on each crime and legal provision.

[0011] The probability value is used to make a threshold judgment to obtain the decision prediction result.

[0012] A further optimized solution involves obtaining the original data of judicial judgment documents and preprocessing the original data to obtain a judgment dataset, including the following methods:

[0013] Obtain the original data of judicial judgment documents;

[0014] Based on the keywords of fact description, crime and judgment, and legal provisions and judgment, the content of fact description, crime and judgment, and legal provisions and judgment are extracted respectively.

[0015] Extract crime tags and legal provision tags from the content of the crime-related judgment and the content of the legal provision judgment respectively to form a tag list;

[0016] Set a specific crime label and a specific legal provision label, and filter out the raw data that does not contain the specified crime label and the specified legal provision label;

[0017] Construct a judgment dataset consisting of <fact description content, crime label set, and legal provision label set>.

[0018] A further optimization scheme involves using the xsBert model to encode the fact description part to obtain semantic encoding information for the fact description, and then using the BiGRU model to perform bidirectional feature extraction on the semantic encoding information for the fact description to obtain a feature representation model for the fact description, including the following methods:

[0019] Obtain the fact description content of the judgment dataset, and input the first insertion mark [CLS] of each fact description sentence into the xsBert model for encoding;

[0020] Semantic feature vectors for each fact description sentence are obtained based on the xsBert model;

[0021] Based on the BiGRU network, the semantic feature vectors are mapped to the same hidden space to extract fine-grained text features, and a fact description feature representation model is obtained based on the fine-grained text features.

[0022] A further optimization scheme is that the network structure of the xsBert model includes an input layer, a hidden layer, and an output layer;

[0023] The hidden layer consists of 12 layers, including an attention layer, a fully connected layer, an intermediate layer, and a hidden output layer.

[0024] The input layer obtains the word vectors of the factual description sentences as input vectors;

[0025] The attention layer is a multi-head attention mechanism with 12 heads; after obtaining the attention score of the fact description sentence, the attention layer superimposes the input vector as the output.

[0026] The output of the attention layer is fed into a fully connected layer, and then the output of the intermediate layer is obtained through the GELU activation function;

[0027] The hidden output layer obtains its output by performing a full connection and Dropout operation on the output of the intermediate layer.

[0028] The outputs of all hidden output layers are used as the outputs of the xsBert model.

[0029] A further optimization scheme involves mapping the semantic feature vectors to the same hidden space using a BiGRU network, extracting fine-grained text features, and obtaining a factual description feature representation model based on these fine-grained text features, including the following methods:

[0030] The semantic feature vectors are fused to obtain the fused vector;

[0031] Initialize a bidirectional GRU network and input the fused vector into the bidirectional GRU network to capture fine-grained features of the text;

[0032] By concatenating the output representations of the last two time steps in the bidirectional GRU network, the final fact description feature representation model is obtained.

[0033] A further optimization scheme is that the fine-grained text feature output is represented as:

[0034] output=GRU(fact_all,num_layer,Dim,bidirectional,dropout)

[0035] Where Dim is the model encoding dimension, num_layer is the number of network layers; fact_all represents the fusion vector; bidirectional indicates whether it is bidirectional encoding; dropout represents the probability of random dropout to prevent overfitting.

[0036] A further optimized solution involves constructing a case description label dataset for the cases to be predicted, inputting the case description label dataset into a factual description feature representation model, and using a fully connected layer to map the factual description feature representations onto crimes and legal provisions to obtain the probability values ​​of the judicial cases to be predicted for each crime and legal provision; including the following methods:

[0037] Obtain the raw data of the cases to be predicted and construct a case description tag dataset consisting of <fact description content, crime tag set, and legal provision tag set>.

[0038] The case description label dataset is input into the fact description feature representation model. The features are mapped to legal provisions and crimes through a fully connected layer to obtain the distribution probability of the case to be predicted in each legal provision label and crime label.

[0039] The probability distribution of the case to be predicted under each legal provision label and crime label is input into the Sigmoid activation function to calculate the probability value of the judicial case to be predicted under each crime and legal provision.

[0040] A further optimization scheme is that the step of obtaining the judgment prediction result by thresholding the probability value includes the following method: mapping the probability value of the judicial case to be predicted for each crime and legal provision to the real crime and legal provision through a hash table.

[0041] This solution provides a judicial case judgment prediction system to implement the aforementioned judicial case judgment prediction method. The system includes:

[0042] The preprocessing module is used to acquire the raw data of judicial judgment documents and preprocess the raw data to obtain a judgment dataset; the judgment dataset includes a factual description section, a crime determination section, and a legal provision determination section;

[0043] The feature extraction module is used to encode the fact description part based on the xsBert model to obtain the fact description semantic encoding information, and to perform bidirectional feature extraction based on the fact description semantic encoding information to obtain the fact description feature representation model.

[0044] The prediction module is used to construct a case description label dataset of the cases to be predicted. The case description label dataset is input into the fact description feature representation model, and the fully connected layer maps the fact description feature representation to the crime and legal provisions to obtain the probability value of the judicial case to be predicted on each crime and legal provision.

[0045] The output module is used to perform threshold judgment on the probability value to obtain the decision prediction result.

[0046] This solution also provides a computer-readable medium having a computer program stored thereon, which, when executed by a processor, can implement the aforementioned method for predicting the judgment of a judicial case.

[0047] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0048] 1. This invention provides a method, system, and medium for predicting judgments in judicial cases. This solution improves upon traditional prediction techniques by extracting factual description features from judicial judgment documents and content to ensure the maximum integrity of the content and structure of judicial judgment documents, while reducing text length and avoiding the problem of incomplete original text loading due to input length limitations of the prediction model. Furthermore, it constructs a factual description feature representation model that fits real-world judgment scenarios based on the xsBert and BiGRU models. A fully connected layer maps the factual description feature representation to the crime and legal provisions, solving the problem of predicting judgments involving multiple crimes and legal provisions.

[0049] 2. This invention provides a method, system, and medium for predicting judgments in judicial cases. This solution extracts content from judicial judgment documents using regular expression matching based on the document structure, obtaining the factual description and final judgment portion as determined by the court, thus shortening the text length and improving feature extraction efficiency. Features are extracted from the factual description portion using the xsBert model. The pre-trained model enables sufficient semantic encoding of the factual description portion, making it more relevant to the judicial field. A BiGRU model network maps the semantic information feature vectors to the same latent space, further capturing fine-grained feature information in the factual description. A linear layer is used as the prediction output, and a multi-label classification prediction method is used to predict the extracted features, obtaining the final prediction result, which better reflects real-world judgment prediction scenarios, including scenarios with multiple charges and legal provisions.

[0050] 3. The present invention provides a method, system and medium for predicting judgments in judicial cases; this solution extracts the factual description part and the judgment prediction part of judicial judgment documents by regular matching according to the document structure, which can maximize the integrity of the content and structure of judicial judgment documents, reduce the text length, and avoid the problem of incomplete loading of the original text due to the limitation of model input length.

[0051] 4. This invention provides a method, system, and medium for predicting judgments in judicial cases. This solution combines the xsBert model and the BiGRU model to address the impact of model input or output limitations on feature extraction from long texts. The xsBert model is used for feature extraction of single sentences, which effectively avoids the limitation of model input length since the length of a single sentence is often limited. The BiGRU model is used for fine-grained feature extraction, which can compensate for the deficiency of the xsBert model in capturing the contextual semantics between sentences due to extracting features from single sentences. A linear layer is used for multi-label classification prediction, making the model more consistent with real-world case scenarios and solving the problem of predicting judgments involving multiple charges and legal provisions. Attached Figure Description

[0052] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort. In the drawings:

[0053] Figure 1 A flowchart illustrating the process of predicting judgments in judicial cases;

[0054] Figure 2 A schematic diagram illustrating the judgment prediction process in judicial cases;

[0055] Figure 3 This is a schematic diagram of the structure of a judicial case judgment prediction system. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.

[0057] Existing methods for predicting judicial judgments lack research on cases involving multiple charges and legal provisions, and their feature extraction from judicial texts is incomplete, resulting in information loss. To address these shortcomings, this solution provides the following embodiments to resolve the aforementioned technical problems:

[0058] Example 1

[0059] This embodiment provides a method for predicting judgments in judicial cases, such as... Figure 1 and Figure 2 As shown, it includes:

[0060] Step one involves obtaining the raw data of judicial judgment documents and preprocessing the raw data to obtain a judgment dataset. The judgment dataset includes a factual description section, a crime determination section, and a legal provision determination section. Step one specifically includes the following sub-steps:

[0061] S11, Obtain the original data of judicial judgment documents;

[0062] S12, based on the keywords of fact description, crime judgment, and legal provisions judgment, extract the content of the fact description, crime judgment, and legal provisions judgment respectively; in the specific implementation process, the keywords of crime judgment and legal provisions judgment include "the court found that...", "this court holds...", "the judgment is as follows...", "the ruling is as follows...", "according to the provisions of the law...", etc. in judicial judgment documents.

[0063] S13, extract crime tags and legal provision tags from the content of the crime judgment and the content of the legal provision judgment respectively to form a tag list;

[0064] S14, set a specified crime label and a specified legal provision label, and filter out the original data that does not contain the specified crime label and the specified legal provision label;

[0065] S15, construct a judgment dataset consisting of <fact description content, crime label set, and legal provision label set>.

[0066] The process involves identifying and omitting basic information in judicial judgments, such as personnel and location information, that does not relate to the judgment content. It then reads the judicial judgments sentence by sentence, identifying the factual descriptions, charges, and legal provisions determined by the court. If no corresponding content is identified in the judicial document, it is discarded as incomplete. The identified content is further processed, converting charge and legal provision labels into corresponding numerical representations. For content containing multiple labels, a label list is created. Data that does not contain the specified labels is filtered out, resulting in a true and complete judgment dataset.

[0067] Step two involves using the xsBert model to encode the fact description part to obtain semantic encoding information for the fact description, and then using the BiGRU model to perform bidirectional feature extraction on the semantic encoding information for the fact description to obtain a feature representation model for the fact description. This step specifically includes the following sub-steps:

[0068] S21, Obtain the fact description content of the judgment dataset: Fact = {fact1, fact2, fact3, ... fact} n}, where n represents the total number of factual sentences in the factual description content; fact n This represents the nth factual description sentence.

[0069] The network structure of the xsBert model described in this scheme includes an input layer, hidden layers, and an output layer. The hidden layer has a total of 12 layers, including an attention layer, a fully connected layer, an intermediate layer, and a hidden output layer. The input layer obtains the word vectors of the factual description sentence as the input vector. The attention layer uses a 12-head multi-head attention mechanism. After obtaining the attention score of the factual description sentence, the attention layer superimposes it on the input vector as the output. The output of the attention layer is connected to a fully connected layer, and then the output of the intermediate layer is obtained through the GELU activation function. The hidden output layer performs fully connected and Dropout operations on the output of the intermediate layer to obtain the output of the hidden output layer. The outputs of all hidden output layers are used as the output of the xsBert model.

[0070] S22, obtain the semantic feature vectors of each fact description sentence based on the xsBert model;

[0071] According to the xsBert model, the maximum length of the input text is N, where N = 380. For each fact description sentence... i ={w1, w2, w3, ... w L}, where L is the length of the fact description sentence, L∈N; insert the [CLS] symbol before w1 of each fact description sentence, and then insert the [CLS] symbol into the fact description sentence. i，cls The input is fed into the xsBert model for encoding, where a [CLS] flag is added before the corpus. The vector corresponding to this flag in the last layer can serve as the semantic representation of the entire sentence, thus being used for downstream classification tasks. The xsBert model extracts features from each factual description sentence to obtain the output vector. Among them, fact i，cls This represents the output after inserting the [CLS] symbol into the i-th sentence of the fact description;

[0072] S23, Based on the BiGRU network, the semantic feature vectors are mapped to the same hidden space to extract fine-grained text features, and a factual description feature representation model is obtained based on these fine-grained text features. This step specifically includes the following methods:

[0073] S231, fuse the semantic feature vectors to obtain the fusion vector fact all:

[0074]

[0075] Where n represents the number of sentences in a factual description;

[0076] S232, initialize a bidirectional GRU network, input the fused vector into the bidirectional GRU network to capture fine-grained text features; the text fine-grained feature output is represented as:

[0077] output=GRU(fact_all,num_layer,Dim,bidirectional,dropout)

[0078] Where Dim represents the model encoding dimension, num_layer represents the number of network layers; fact_all represents the fusion vector; bidirectional indicates whether it is bidirectional encoding; dropout represents the probability of random dropout to prevent overfitting.

[0079] S233, concatenate the output representations of the last two time steps in the bidirectional GRU network to obtain the final fact description feature representation model f: f = Concate(output[-2,:,:], output[-1,:,:]);

[0080] Where output[-2,:,:] and output[-1,:,:] represent the outputs of the last two time steps of the GRU network.

[0081] Step 3: Construct a case description label dataset for the cases to be predicted. Input the case description label dataset into a factual description feature representation model. A fully connected layer maps the factual description feature representations to crimes and legal provisions, obtaining the probability values ​​of the judicial cases to be predicted for each crime and legal provision. This step specifically includes the following methods:

[0082] S31. Obtain the raw data of the cases to be predicted and construct a case description tag dataset of <fact description content, crime tag set, and legal provision tag set>.

[0083] S32, input the case description label dataset into the fact description feature representation model, and map the features to legal provisions and crimes through a fully connected layer to obtain the distribution probability of the case to be predicted in each legal provision label and crime label; the distribution probability of the case to be predicted i is: R i =Linear i (f); where f represents the final factual description feature expression.

[0084] S33, input the probability distribution of the case to be predicted under each legal provision label and crime label into the Sigmoid activation function to calculate the probability value of the judicial case to be predicted under each crime and legal provision: R out，i =Sigmoid(R) i ); where R i Let represent the probability distribution of case i to be predicted.

[0085] Step four: Perform a threshold judgment on the probability value to obtain the decision prediction result.

[0086] This step specifically includes the following method: mapping the probability values ​​of the judicial cases to be predicted for each crime and legal provision to the actual crime and legal provision using a hash table.

[0087] This solution extracts content from judicial judgments using regular expression matching based on the document structure, obtaining the factual description and final judgment as determined by the court, thus shortening the text length and improving feature extraction efficiency. It then uses the xsBert model to extract features from the factual description, leveraging a pre-trained model to fully semantically encode the factual description, making it more relevant to the judicial field. A BiGRU model network maps the semantic information feature vectors to the same latent space, further capturing fine-grained features from the factual description. Finally, a linear layer is used as the prediction output, and a multi-label classification prediction method is employed to predict the extracted features, obtaining the final prediction result that better reflects real-world judgment prediction scenarios, including those involving multiple charges and legal provisions.

[0088] Example 2

[0089] This embodiment provides a judgment prediction system for judicial cases, such as... Figure 3 As shown, the system for implementing the judgment prediction method for judicial cases described in Embodiment 1 includes:

[0090] The preprocessing module is used to acquire the raw data of judicial judgment documents and preprocess the raw data to obtain a judgment dataset; the judgment dataset includes a factual description section, a crime determination section, and a legal provision determination section;

[0091] The feature extraction module is used to encode the fact description part based on the xsBert model to obtain the fact description semantic encoding information, and to perform bidirectional feature extraction based on the fact description semantic encoding information to obtain the fact description feature representation model.

[0092] The prediction module is used to construct a case description label dataset of the cases to be predicted. The case description label dataset is input into the fact description feature representation model, and the fully connected layer maps the fact description feature representation to the crime and legal provisions to obtain the probability value of the judicial case to be predicted on each crime and legal provision.

[0093] The output module is used to perform threshold judgment on the probability value to obtain the decision prediction result.

[0094] Example 3

[0095] This embodiment provides a computer-readable medium storing a computer program thereon. The computer program, when executed by a processor, can implement a judgment prediction method for judicial cases as described in Embodiment 1, specifically including the following steps:

[0096] Step 1: Obtain the raw data of the judicial judgment documents and preprocess the raw data to obtain the judgment dataset; the judgment dataset includes the factual description part, the crime judgment part, and the legal provision judgment part;

[0097] Step 2: Based on the xsBert model, feature encoding is performed on the fact description part to obtain the fact description semantic encoding information. Based on the BiGRU model, bidirectional feature extraction is performed on the fact description semantic encoding information to obtain the fact description feature representation model.

[0098] Step 3: Construct a case description label dataset of the cases to be predicted. Input the case description label dataset into the fact description feature representation model. The fully connected layer maps the fact description feature representation to the crime and legal provisions to obtain the probability value of the judicial case to be predicted on each crime and legal provision.

[0099] Step four: Perform a threshold judgment on the probability value to obtain the decision prediction result.

[0100] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting judgments in judicial cases, characterized in that, include: The original data of judicial judgment documents is obtained, and the original data is preprocessed to obtain a judgment dataset; the judgment dataset includes a factual description section, a crime determination section, and a legal provision determination section; Based on the xsBert model, feature encoding is performed on the fact description part to obtain the fact description semantic encoding information. Based on the BiGRU model, bidirectional feature extraction is performed on the fact description semantic encoding information to obtain the fact description feature representation model. A case description label dataset of the cases to be predicted is constructed. The case description label dataset is input into the fact description feature representation model. The fully connected layer maps the fact description feature representation to the crime and legal provisions to obtain the probability value of the judicial case to be predicted on each crime and legal provision. The decision prediction result is obtained by applying a threshold to the probability value; The method for obtaining the raw data of judicial judgment documents and preprocessing the raw data to obtain the judgment dataset includes: Obtain the original data of judicial judgment documents; Based on the keywords of fact description, crime and judgment, and legal provisions and judgment, the content of fact description, crime and judgment, and legal provisions and judgment are extracted respectively. Extract crime tags and legal provision tags from the content of the crime-related judgment and the content of the legal provision judgment respectively to form a tag list; Set a specific crime label and a specific legal provision label, and filter out the raw data that does not contain the specified crime label and the specified legal provision label; Construct a judgment dataset consisting of <fact description content, crime label set, and legal provision label set>. The method involves encoding the fact description part using the xsBert model to obtain semantic encoding information for the fact description, and then performing bidirectional feature extraction on the semantic encoding information for the fact description using the BiGRU model to obtain a feature representation model for the fact description, including the following methods: Obtain the fact description content of the judgment dataset, and input the first insertion mark [CLS] of each fact description sentence into the xsBert model for encoding; Semantic feature vectors for each fact description sentence are obtained based on the xsBert model; Based on the BiGRU network, the semantic feature vectors are mapped to the same hidden space to extract fine-grained text features, and a fact description feature representation model is obtained based on the fine-grained text features. The process involves constructing a case description label dataset for the cases to be predicted, inputting the case description label dataset into a factual description feature representation model, and using a fully connected layer to map the factual description feature representations onto crimes and legal provisions, thereby obtaining the probability values ​​of the judicial cases to be predicted for each crime and legal provision; including method: Obtain the raw data of the cases to be predicted and construct a case description tag dataset consisting of <fact description content, crime tag set, and legal provision tag set>. The case description label dataset is input into the fact description feature representation model. The features are mapped to legal provisions and crimes through a fully connected layer to obtain the distribution probability of the case to be predicted in each legal provision label and crime label. The probability distribution of the case to be predicted under each legal provision label and crime label is input into the Sigmoid activation function to calculate the probability value of the judicial case to be predicted under each crime and legal provision.

2. The method for predicting the judgment of a judicial case according to claim 1, characterized in that, The network structure of the xsBert model includes an input layer, a hidden layer, and an output layer; The hidden layer consists of 12 layers, including an attention layer, a fully connected layer, an intermediate layer, and a hidden output layer. The input layer obtains the word vectors of the factual description sentences as input vectors; The attention layer is a multi-head attention mechanism with 12 heads; After the attention layer obtains the attention score of the fact description sentence, it superimposes the input vector as the output. The output of the attention layer is fed into a fully connected layer, and then the output of the intermediate layer is obtained through the GELU activation function; The hidden output layer obtains its output by performing a full connection and Dropout operation on the output of the intermediate layer. The outputs of all hidden output layers are used as the outputs of the xsBert model.

3. The method for predicting the judgment of a judicial case according to claim 1, characterized in that, The method involves mapping the semantic feature vectors to the same hidden space using a BiGRU network to extract fine-grained text features, and obtaining a factual description feature representation model based on these features, including the following steps: The semantic feature vectors are fused to obtain the fused vector; Initialize a bidirectional GRU network and input the fused vector into the bidirectional GRU network to capture fine-grained features of the text; By concatenating the output representations of the last two time steps in the bidirectional GRU network, the final fact description feature representation model is obtained.

4. The method for predicting the judgment of a judicial case according to claim 3, characterized in that, The fine-grained text feature output is represented as follows: ； Where Dim is the model encoding dimension, num_layer is the number of network layers; fact_all represents the fusion vector; bidirectional indicates whether it is bidirectional encoding; dropout represents the probability of random dropout to prevent overfitting.

5. The method for predicting the judgment of a judicial case according to claim 1, characterized in that, The method of obtaining the judgment prediction result by thresholding the probability value includes: mapping the probability value of the judicial case to be predicted for each crime and legal provision to the real crime and legal provision through a hash table.

6. A judgment prediction system for judicial cases, characterized in that, The system is used to implement the judgment prediction method for judicial cases according to any one of claims 1-5, the system comprising: The preprocessing module is used to acquire the raw data of judicial judgment documents and preprocess the raw data to obtain a judgment dataset; the judgment dataset includes a factual description section, a crime determination section, and a legal provision determination section; The feature extraction module is used to encode the fact description part based on the xsBert model to obtain the fact description semantic encoding information, and to perform bidirectional feature extraction based on the fact description semantic encoding information to obtain the fact description feature representation model. The prediction module is used to construct a case description label dataset of the cases to be predicted. The case description label dataset is input into the fact description feature representation model, and the fully connected layer maps the fact description feature representation to the crime and legal provisions to obtain the probability value of the judicial case to be predicted on each crime and legal provision. The output module is used to perform threshold judgment on the probability value to obtain the decision prediction result.

7. A computer-readable medium having a computer program stored thereon, characterized in that, The computer program, when executed by a processor, can implement a method for predicting the judgment of a judicial case as described in any one of claims 1-5.

Images