Case information processing method, device and equipment

Abstract

The application provides a case information processing method, device and equipment. The method of the application obtains a plurality of case description files of a to-be-processed case, extracts case description texts from each case description file, divides each case description text into text segments according to the relevance of sentences in the case description text, and the sentences in the same text segment correspond to the same event. The text segments divided by the plurality of case description files are clustered according to the relevance between the text segments, a plurality of clusters are obtained, each cluster contains at least one text segment corresponding to the same event. Event information is extracted from the text segments contained in each cluster, the event information is sorted according to time elements to generate case fact information, and the case fact information is output. The case fact information can be automatically sorted from the multiple and complex case description files, the time consumed by relevant personnel in sorting the case fact is reduced, the efficiency of generating the case fact is improved, and thus the case handling efficiency can be improved.

Description

Technical Field

[0001] This application relates to data processing technology, and more particularly to a method, apparatus and equipment for processing case information. Background Technology

[0002] In real life, there are various legal cases with highly complex facts. For example, criminal cases often involve multiple events of different types occurring at different times and locations. The details of these cases are typically recorded in documents such as written statements, indictments, and official complaints.

[0003] In the investigation of complex criminal cases, oral evidence such as transcripts contains a large amount of case descriptions. The analysis of such oral evidence is a very important step. However, the case descriptions in transcripts are often complicated and disordered. Relevant staff need to extract information, understand the content, and cross-verify it from the transcripts in order to sort out the factual chain of the case. This process consumes a lot of the staff's time and is very inefficient. Summary of the Invention

[0004] This application provides a case information processing method, apparatus, and equipment for automatically generating case fact information from numerous and complex case description documents, thereby improving the efficiency of case fact generation and thus enhancing case handling efficiency.

[0005] On the one hand, this application provides a method for processing case information, including:

[0006] Obtain multiple case description files for the cases to be processed, and extract the case description text from each of the case description files;

[0007] Based on the relevance of sentences in the case description text, the case description text is divided into text segments, wherein sentences in the same text segment correspond to the same event;

[0008] Based on the correlation between the text fragments, the text fragments divided into the multiple case description files are clustered to obtain multiple clusters, and each cluster contains at least one text fragment corresponding to the same event;

[0009] The corresponding event information is determined based on the text fragments contained in each cluster, and the event information includes at least the time element of the event;

[0010] The event information is sorted according to the time element to generate case fact information, and the case fact information is output.

[0011] On the other hand, this application provides a case information processing device, comprising:

[0012] The case description text acquisition module is used to acquire multiple case description files of the case to be processed, and extract the case description text from each of the case description files;

[0013] The event segmentation module is used to divide the case description text into text segments based on the relevance of sentences in the case description text, wherein sentences in the same text segment correspond to the same event;

[0014] The event clustering module is used to cluster the text fragments divided into multiple case description files according to the correlation between the text fragments, to obtain multiple clusters, and each cluster contains at least one text fragment corresponding to the same event;

[0015] An event information extraction module is used to determine the corresponding event information based on the text fragments contained in each cluster, wherein the event information includes at least the time element of the event;

[0016] The case fact generation module is used to sort the event information according to time elements, generate case fact information, and output the case fact information.

[0017] On the other hand, this application provides an electronic device, including:

[0018] A processor, and a memory communicatively connected to the processor;

[0019] The memory stores computer-executed instructions;

[0020] The processor executes computer execution instructions stored in the memory to implement the method described in any of the above aspects.

[0021] On the other hand, this application provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the method described in any of the above aspects.

[0022] The case information processing method, apparatus, and equipment provided in this application acquire multiple case description files of a case to be processed and extract case description text from each file; divide the case description text into text segments based on the relevance of sentences within the text, where sentences in the same text segment correspond to the same event; cluster the text segments from the multiple case description files based on the relevance between the text segments to obtain multiple clusters, where each cluster contains at least one text segment corresponding to the same event; extract event information from the text segments contained in each cluster, where the event information includes at least the time element of the event; sort the event information according to the time element to generate case fact information and output the case fact information. This method can automatically extract case fact information from numerous and complex case description files, reducing the time spent by relevant staff in sorting case facts and improving the efficiency of generating case facts, thereby improving case handling efficiency. Attached Figure Description

[0023] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0024] Figure 1 Example scenario diagram of the case information processing system provided in the embodiments of this application;

[0025] Figure 2 A flowchart of a case information processing method provided for an exemplary embodiment of this application;

[0026] Figure 3 A detailed flowchart of a case information processing method provided for an exemplary embodiment of this application;

[0027] Figure 4 A flowchart of the training of a contrastive learning encoding model provided for an exemplary embodiment of this application;

[0028] Figure 5 A flowchart illustrating an exemplary embodiment of this application for interactive viewing of a case description document;

[0029] Figure 6 A schematic diagram illustrating the application of a case information processing method provided in an exemplary embodiment of this application;

[0030] Figure 7 A schematic diagram of the structure of a case information processing apparatus provided in an exemplary embodiment of this application;

[0031] Figure 8 This is a schematic diagram of the structure of an electronic device provided in an example embodiment of this application.

[0032] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concepts of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0033] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0034] First, let me explain the terms used in this application:

[0035] Interrogation record: A written record made by judicial personnel during an investigation, in order to prove a crime and ascertain the facts of the crime, of the interrogation of a criminal suspect.

[0036] Interrogation record: also known as "witness interrogation record", is a written record made by judicial personnel in criminal proceedings of the process and content of interrogating witnesses and victims.

[0037] Event segmentation: The process of extracting text describing the legal facts of a case from the transcript.

[0038] Fact Chain: Organize the case details mentioned by multiple people in chronological order of events.

[0039] To address the problem of time-consuming and inefficient processes involving information extraction, content understanding, and cross-verification from transcripts to establish the factual chain of a case, this application provides a case information processing method. This method involves acquiring multiple case description documents for a case to be processed and extracting case description text from each document. Based on the relevance of sentences within the case description text, the text is divided into text segments, where sentences within the same segment correspond to the same event. The text segments are then clustered based on their relevance to obtain multiple clusters, each containing at least one text segment corresponding to the same event. Event information is extracted from the text segments within each cluster, including at least the time element of the event. The event information is then sorted according to the time element to generate and output the case fact information. This method automatically extracts case fact information from numerous and complex case description documents, reducing the time spent by relevant staff in processing case facts and improving the efficiency of generating case facts, thereby enhancing case-handling efficiency.

[0040] The method provided in this application can be implemented by an electronic device with computing capabilities, specifically a server deployed in the cloud or a locally deployed computing device. It can be applied to artificial intelligence (AI) judge assistants, intelligent case trial systems / platforms, etc. The electronic device can be a cloud server for the AI ​​judge assistant, a server for the intelligent case trial system / platform, etc., to assist relevant staff in automatically extracting case factual information from numerous and complex case description documents, thereby saving staff time spent on sorting out case facts, improving the efficiency of case fact sorting, and thus improving case handling efficiency.

[0041] Figure 1 Example scenario diagram of the case information processing system provided in the embodiments of this application, such as Figure 1 As shown, the case information processing system runs on an electronic device. Based on a user-specified case to be processed, this electronic device can obtain multiple case description files, such as interrogation records, questioning records, indictment opinions, and indictments. It automatically extracts the case description text from these multiple files, segments the text to identify text fragments belonging to different events, clusters all text fragments to determine the text fragments contained in each event within the case, extracts the event types and elements, obtains event information, and sorts the event information to obtain the factual information of the case. The events in the case's factual information are arranged chronologically, forming the factual chain of the case.

[0042] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0043] Figure 2 This is a flowchart illustrating a case information processing method provided in an exemplary embodiment of this application. The executing entity in this embodiment is an electronic device used for case information processing, which can be a cloud server implementing functions such as AI judge assistants and intelligent case adjudication, or a computing device deployed locally by various departments.

[0044] like Figure 2 As shown, the specific steps of this method are as follows:

[0045] Step S201: Obtain multiple case description files for the case to be processed, and extract the case description text from each case description file.

[0046] The cases to be processed can be criminal cases, civil cases, economic cases, or other types of cases. This embodiment does not make specific limitations here.

[0047] Case description documents refer to documents that contain information describing the facts of the case, or information describing the events in the case. They typically include various records, indictment opinions, indictments, etc.

[0048] For example, case description documents in criminal cases typically include: interrogation records, questioning records, indictment opinions, and indictments.

[0049] The method in this embodiment can automatically extract the content text of each case description file from multiple case description files to obtain the case description text.

[0050] For example, in one possible application scenario, an electronic device can receive a case fact generation request submitted by a user through a front-end page, the request carrying identification information of the case to be processed. Based on the identification information, the electronic device obtains a case description file for the case to be processed. This case description file can be stored on a local storage device or other devices.

[0051] Step S202: Based on the relevance of sentences in the case description text, divide the case description text into text segments, where sentences in the same text segment correspond to the same event.

[0052] After extracting the case description text, for each case description text, based on the semantic relevance of the sentences within the text, each case description text is divided into multiple text segments. Each text segment is a content description fragment describing one event or multiple events combined (which can be considered as one event), and can be called an event segment. Thus, by segmenting the text according to the events described in the sentences, the case description text is split into different event segments.

[0053] Step S203: Based on the correlation between text fragments, cluster the text fragments divided into multiple case description documents to obtain multiple clusters, and each cluster contains at least one text fragment corresponding to the same event.

[0054] Based on the text fragments divided into various case description documents, all text fragments are clustered to obtain multiple clusters. Each cluster contains one or more text fragments, and the text fragments within the same cluster describe the same event, corresponding to the same event. Text fragments within different clusters describe different events; that is, different clusters correspond to different events.

[0055] Step S204: Determine the corresponding event information based on the text fragments contained in each cluster. The event information shall include at least the time element of the event.

[0056] After obtaining multiple clusters corresponding to an event, information such as the type and elements of the corresponding event is extracted from the file fragments contained in the same cluster to obtain the event information.

[0057] The type of event can be set and adjusted according to the type of case in the actual application scenario, such as drunk driving incidents, assault on police officers incidents, intentional injury incidents, etc. This embodiment does not make specific limitations here.

[0058] The elements of an event include: the perpetrator, the victim, the time, the place, the means, the consequences, the amount of money involved, etc.

[0059] Step S205: Sort the event information according to the time element, generate case fact information, and output the case fact information.

[0060] After determining the event information corresponding to each cluster, the event information contained in the case is obtained. In this step, the event information contained in the case is sorted according to the time element to obtain the case fact information. This case fact information includes all the event information of the case, and the event information is arranged in chronological order of occurrence, clearly presenting the case fact chain.

[0061] Optionally, when outputting case fact information, it can be output in various document formats, or a visual interface for displaying the case fact information can be generated. To facilitate the differentiation of different events, blank areas, dividing lines, etc., can be used in this visual interface to separate different events, or the event information of different events can be organized in a certain format and the beginning of each event can be highlighted, or other methods can be used to make the display of case fact information more in line with the user's browsing habits and to provide a more user-friendly presentation.

[0062] In this embodiment, multiple case description files of the case to be processed are obtained, and case description text is extracted from each case description file. Based on the relevance of sentences in each case description text, each case description text is divided into text fragments, where sentences in the same text fragment correspond to the same event. Based on the relevance between text fragments, the text fragments from multiple case description files are clustered to obtain multiple clusters, where at least one text fragment in each cluster corresponds to the same event. Event information is extracted from the text fragments in each cluster, and the event information includes at least the time element of the event. The event information is sorted according to the time element to generate case fact information, and the case fact information is output. This method can automatically extract case fact information from numerous and complex case description files, reducing the time spent by relevant staff in sorting case facts and improving the efficiency of generating case facts, thereby improving case handling efficiency.

[0063] Figure 3 This is a detailed flowchart of a case information processing method provided for an exemplary embodiment of this application. Based on the above embodiments, in this embodiment, after obtaining the case description text, referential resolution is performed on each case description text to replace the referential pronouns of the parties appearing in the case description text with the names of the same parties. This can improve the accuracy of event segmentation and event clustering, thereby improving the accuracy of event segmentation and the extracted event information.

[0064] like Figure 3 As shown, the specific steps of this method are as follows:

[0065] Step S301: In response to the case fact generation request of the case to be processed, obtain multiple case description files of the case to be processed.

[0066] For example, a front-end visual interface is provided through a display device. Through this visual interface, the user can select a case to be processed and trigger a case fact generation operation to send a case fact generation request to an electronic device. The case fact generation request contains the identification information of the case to be processed selected by the user, such as the case number and case name.

[0067] In response to a case fact generation request, the electronic device extracts the identification information of the case to be processed from the request and retrieves the case description file of the case to be processed from the case database based on the identification information.

[0068] The case database is a file repository containing various types of case documents, with at least five case description documents. Case description documents refer to files containing descriptive information about the case or events within it, typically including various transcripts, indictment opinions, and indictments. For example, case description documents in criminal cases usually include: interrogation transcripts, questioning transcripts, indictment opinions, and indictments.

[0069] Optionally, the case database can be stored on the electronic device, which can directly read one or more case description files of the case to be processed from the locally stored case database.

[0070] 0. Optionally, the case database can be stored on other devices, such as servers in the judicial system. Electronic devices obtain...

[0071] Once authorized, case description files for pending cases in the case database can be retrieved from other devices.

[0072] Step S302: Extract the case description text from each case description file.

[0073] In practical applications, the case description documents obtained can be scanned images of paper documents or electronic documents. Electronic case description documents can be in image format or text format.

[0074] 5. In this step, for image files in the case description documents (such as scanned images, image-format documents, etc.), each...

[0075] Each image file undergoes text recognition to obtain a text document for each image file, which is also the text document of the case description file. For each text document of the case description file to be processed, structural analysis is performed, and the case description text is extracted from the text document to obtain the case description text.

[0076] This involves performing text recognition on each image file to obtain its text document. This can be achieved using Optical Character Recognition (OCR) technology, which identifies the text within the image file.

[0077] Given text, you can obtain a text document. This can be achieved using existing OCR tools / software, which will not be elaborated upon here.

[0078] To perform structural analysis on the text document of each case description file for the case being processed, Intelligence Document Processing (IDP) technology can be used to analyze the document, extract hierarchical structure, text content, style information, etc., and extract the text content to obtain the case description text.

[0079] 5. Specifically, case description documents may contain content unrelated to the main body of the case description, such as seals, signatures, and fingerprints. By performing structural analysis on the text document of the case description, style information (such as pagination, page numbers, headers, and footers) and content unrelated to the main body of the case description, such as seals, signatures, and fingerprints, can be removed. The main text of the document can then be extracted to obtain the case description text, thus avoiding the influence of other irrelevant content on the division of events.

[0080] The effects of clustering can improve the accuracy of event segmentation and the extracted event information, thereby improving the accuracy of the final case fact information.

[0081] Step S303: Dereference each case description text to replace the referential persons of the parties appearing in the case description text with the names of the same parties.

[0082] In practical applications, case description documents such as interrogation records and questioning transcripts are truthful records of the questioning and interrogation process of the parties involved (criminal suspects, victims, witnesses), and often use colloquial descriptions. They contain many pronouns such as "I," "he," "she," nicknames, former names, aliases, and pet names. These pronouns can refer to people or organizational structures. Using a large number of identical pronouns across different events is detrimental to event segmentation and clustering.

[0083] In this embodiment, a referential resolution algorithm is used to resolve the referential references of each case description text obtained, so as to replace the referential persons of the parties appearing in the case description text with the names of the same parties.

[0084] Specifically, the identification information of the parties involved, such as names (formal names, organization names, etc.) and identification numbers, is extracted from the case description text. Based on the identification information, personal pronouns used by the parties are retrieved from the database, including their former names, nicknames, aliases, and aliases. Using a pronoun resolution algorithm, the personal pronouns and referents used by the parties in each case description text are replaced with the party's formal name, thus replacing colloquial expressions of personal pronouns in the case description text with formal names. For example, "I" in the interrogation transcript is replaced with the name of the person being interrogated, nicknames are replaced with formal names, and other pronouns are replaced with the actual names of the people / organizations they refer to.

[0085] For each case description text after the referential resolution, through steps S304-S306, each case description text is divided into text fragments based on the relevance of sentences in each case description text.

[0086] Step S304: Use a contrastive learning coding model to encode the sentences in the case description text and determine the sentence representation sequence of the case description text.

[0087] In this embodiment, a pre-trained contrastive learning encoding model is used to encode each sentence in each case description text, resulting in a sentence representation for each sentence. The sentence representations of the sentences in each case description text are arranged in chronological order to form a sequence of sentence representations for the case description text.

[0088] The contrastive learning coding model is obtained through contrastive learning training, and the training process of the contrastive learning coding model will be described in detail in subsequent embodiments.

[0089] For the same case, the factual text and case description text encoded by this contrastive learning coding model show a relatively close distance between them. However, for text information from different cases, the distance between the text representations encoded by the same model is significant. For the description text of the same event, the distance between the text representations encoded by this model is very close, even closer than the distance between the text representations of different events within the same case. For the description text of different events, the distance between the text representations encoded by this model is relatively large. Sentence representations encoded by this contrastive learning coding model can accurately segment events, improving the accuracy of the segmented event fragments.

[0090] Step S305: Use a sequence labeling model to perform sequence labeling on the sentence representation sequence of the case description text to obtain multiple segmented sequences, each segmented sequence containing at least one sentence representation.

[0091] After obtaining the sentence representation sequence of each case description text, a sequence labeling model can be used to perform sequence labeling on the sentence representation sequence of each case description text, resulting in multiple segmented sequences. Each segmented sequence contains at least one sentence representation. Each segmented sequence can contain one sentence representation or multiple consecutive sentence representations.

[0092] This sequence labeling model is trained on a document-level sequence labeling model using the training set in the current scenario. It can label the sentence representations corresponding to the same event in the sentence representation sequence of document-level case description text, resulting in multiple segmented sequences. The sentence representations in each segmented sequence describe one event, or describe multiple events mixed together (which can be regarded as one event).

[0093] For example, the training set used to train the sequence labeling model includes: sentence sequences of case description texts from historical cases, and event labeling information for the sentence sequences. The event labeling information for the sentence sequences refers to the labeling information of the events to which the sentence sequences belong. Based on the event labeling information of the events to which the sentence sequences belong, the sentence sequences can be divided into multiple segmented sequences.

[0094] When training the sequence labeling model, the sentence sequence of the case description text of historical cases is input into the sequence labeling model to be trained for sequence labeling. Based on the sequence labeling results and the segmentation labeling information of the sentence sequence, the loss is calculated and the model parameters of the sequence labeling model are optimized. After multiple rounds of iteration until the model converges, the trained sequence labeling model can be determined.

[0095] In this step, the sentence representation sequence of each case description text is input into the trained sequence labeling model for sequence labeling, which yields the sequence labeling results, which determine multiple segmented sequences.

[0096] Step S306: Determine a text segment based on the sentence representations contained in each segment sequence.

[0097] After obtaining the segmented sequence of sentence representations for each case description text, each segmented sequence contains corresponding sentence representations, forming a text fragment of the case description text. Typically, a text fragment describes the process of an event, or the process of multiple events combined (which can be considered as one event). This text fragment is also called an event fragment.

[0098] By performing steps S304-S306, each case description text is divided into multiple event fragments (text fragments).

[0099] It should be noted that in practical applications, some case description texts contain descriptions unrelated to the event. When dividing the event, the descriptions unrelated to the event and the text describing the event will be separated into different text segments. In other words, some text segments may only contain content unrelated to the event, that is, they do not describe any event. Therefore, when extracting event information from the text segments later, the event information cannot be extracted.

[0100] Furthermore, for the text fragments of multiple case description texts obtained from event segmentation, through steps S307-S308, based on the correlation between the text fragments, the text fragments divided into all case description files for the case are clustered to obtain multiple clusters. Each cluster corresponds to one event, and the corresponding event information can be extracted.

[0101] Step S307: Use a contrastive learning coding model to encode the text fragment and determine the text representation of the text fragment.

[0102] In this step, each text segment of the case description is input into a contrastive learning encoding model for encoding, resulting in a text representation for each segment. The text representation of each segment is a continuous, distributed vector representation that includes information such as the type and content of the event.

[0103] The contrastive learning coding model used in this step is the same model used in step S304 above. For details, please refer to the description of the contrastive learning coding model in the relevant content of step S304 above. It will not be repeated here.

[0104] Step S308: Based on the text representation of the text fragments, cluster the text fragments divided into multiple case description documents to obtain multiple clusters.

[0105] For all text fragments describing the multiple facts of the case, clustering is performed based on the similarity between the text representations of the fragments, resulting in multiple clusters. Each cluster contains one or more text fragments; text fragments within the same cluster describe the same event and correspond to the same event. Text fragments within different clusters describe different events; that is, different clusters correspond to different events.

[0106] In the process of clustering text segments, the similarity between the text representations of the text segments can be cosine similarity, Euclidean distance, Manhattan distance, etc. Common similarity measures used in clustering algorithms can be used, and this embodiment does not make specific limitations here.

[0107] In addition, in this embodiment, any existing clustering algorithm can be used to cluster the text fragments to obtain multiple clusters. This example does not make any specific limitation here.

[0108] Optionally, hierarchical clustering algorithms can be used to cluster text fragments, resulting in multiple clusters. By adjusting the similarity threshold during the clustering process, the clustering hierarchy can be flexibly controlled, thereby obtaining the most suitable clustering result for the current case.

[0109] For example, the range of similarity threshold values ​​can be provided to the user. For instance, for cosine similarity, the range of similarity threshold values ​​can be set to [0,1]. This range can be provided to the user through a front-end visual interface. Users can specify multiple different similarity thresholds through the visual interface or commands.

[0110] Electronic devices perform hierarchical clustering of text segments based on different similarity thresholds specified by the user. Text segments with a similarity greater than or equal to the specified threshold are grouped into the same cluster, resulting in multiple clusters. The front-end page displays the text segments contained in each cluster, allowing the user to view and confirm the degree of clustering. By performing hierarchical clustering at different levels based on multiple user-specified similarity thresholds and displaying the clustering results, users can select and confirm the most accurate clustering result as the final clustering result. This improves the accuracy of event clustering results, thereby enhancing the accuracy of event information extracted based on the clustering results.

[0111] Through steps S307-S308 above, based on the correlation between text fragments, hierarchical clustering is performed on the text fragments divided from multiple case description documents to obtain multiple clusters. This allows text fragments describing event content distributed across multiple different case description documents to be aggregated into text fragment clusters corresponding to the same event. The text fragments in each cluster are descriptions of the same event, achieving accurate event segmentation.

[0112] Step S309: Perform semantic analysis on the text fragments contained in each cluster to determine the type of the corresponding event, and extract the element information of the event from the text fragments contained in each cluster. The event type and element information constitute the event information.

[0113] After obtaining the clusters corresponding to each event in the case, semantic analysis is performed on the text fragments contained in each cluster to determine the type of event and extract the element information of the event from the text fragments contained in each cluster.

[0114] The type of event can be set and adjusted according to the type of case in the actual application scenario, such as drunk driving incidents, assault on police officers incidents, intentional injury incidents, etc. This embodiment does not make specific limitations here.

[0115] The elements of an event include: the perpetrator, the victim, the time, the place, the means, the consequences, the amount of money involved, etc.

[0116] Optionally, a trained event extraction model can be obtained, the text fragments contained in each cluster can be concatenated into a text paragraph, the text paragraph can be input into the event extraction model, the event extraction model can classify and identify the event type corresponding to the text paragraph, and extract the event element information from the text paragraph.

[0117] Specifically, the event extraction model can utilize a natural language processing (NLP) neural network model. This model is trained using a training set specifically designed for event extraction. The training set includes descriptive text and annotation information for the events. The annotation information includes the event type and its features. Additionally, during training, a set of event types is defined, encompassing multiple event types and an "other" type. The "other" type indicates that no event information was extracted, meaning the input text paragraph does not describe an event. The training set can also include text paragraphs describing content unrelated to the event; these unrelated paragraphs are labeled as "other." Using this training set, the NLP neural network model is trained to obtain the trained event extraction model. This model is capable of extracting the event type and feature information corresponding to the input text paragraph.

[0118] Optionally, in this step, a text classification model can be trained using a dataset for event classification to obtain a trained event classification model. In this step, the text fragments contained in each cluster are concatenated into a text paragraph, which is then input into the event classification model. The model classifies and identifies the event type corresponding to the text paragraph, thus determining the event type. Further, semantic analysis is performed on the text paragraph to extract element information describing various elements of the event.

[0119] In this embodiment, by performing event fragment clustering and then extracting the event information corresponding to each cluster based on the clusters obtained from the clustering, the event type and element information can be obtained more accurately.

[0120] Step S310: Sort the event information according to the time element, generate case fact information, and output the case fact information.

[0121] After determining the event information corresponding to each cluster, the event information contained in the case is obtained. In this step, the event information contained in the case is sorted according to the time element to obtain the case fact information. This case fact information includes all the event information of the case, and the event information is arranged in chronological order of occurrence, clearly presenting the case fact chain.

[0122] Optionally, when outputting case fact information, it can be output in various document formats, or a visual interface for displaying the case fact information can be generated. To facilitate the differentiation of different events, blank areas, dividing lines, etc., can be used in this visual interface to separate different events, or the event information of different events can be organized in a certain format and the beginning of each event can be highlighted, or other methods can be used to make the display of case fact information more in line with the user's browsing habits and to provide a more user-friendly presentation.

[0123] In this embodiment, image files of multiple case description documents for the case to be processed are acquired, and text recognition is performed on each image file to obtain a text document of each case description document. The structure of each text document of the case description documents is analyzed, and the case description text is extracted from the text document, which can automatically extract the case description text. Furthermore, each case description text undergoes referential resolution to replace the referential pronouns of the parties appearing in the case description text with the names of the same parties. This replaces colloquial pronouns in the case description text with formal names, improving the accuracy of event segmentation and event clustering, thereby improving the accuracy of event segmentation and the extracted event information. Further, based on... Based on the relevance of sentences in each case description text, each case description text is divided into text fragments of the event; based on the relevance between text fragments, the text fragments are clustered into multiple clusters, with each cluster containing at least one text fragment corresponding to the same event; event information is extracted from the text fragments contained in each cluster; the event information is sorted according to the time element to generate case fact information, and the case fact information is output. This method can automatically sort out case fact information from numerous and complex case description files, assisting users in reading and understanding transcripts and evidence, reducing the time spent by relevant staff in sorting out case facts, improving the efficiency and accuracy of generating case facts, and is applicable to the sorting of case facts for various types of cases, demonstrating strong versatility.

[0124] In an optional embodiment, the contrastive learning encoding model used in steps S304 and S307 above can be determined by training a deep learning model using contrastive learning. Figure 4 As shown, the contrastive learning encoding model can be trained through the following steps:

[0125] Step S401: Obtain the judgment documents of historical cases, and extract the case fact information and case description text of historical cases from the judgment documents.

[0126] Among them, the judgment documents of historical cases clearly and systematically describe the facts of the case, as well as the case description texts in the adopted transcripts. By extracting the case fact information and case description texts of historical cases from the judgment documents, multiple authentic and clear case fact information and case description texts of historical cases can be obtained, and the historical cases to which the case fact information and case description texts belong can be identified.

[0127] Step S402: Based on the case facts and case description text of historical cases, the neural network model is compared and learned to obtain the initial encoding model.

[0128] The neural network model can be a deep learning model, which can better learn the representation ability of case-related descriptive text.

[0129] In this step, the neural network model to be trained is used to encode the case fact information and case description text of historical cases respectively, so as to obtain the case fact text representation and case description text representation of historical cases; positive sample pairs are constructed based on the case fact text representation and case description text representation of the same historical case, and negative sample pairs are constructed based on the text representations of different historical cases; the neural network model is compared and learned using the positive sample pairs and negative sample pairs to obtain the initial encoding model.

[0130] Specifically, the neural network model to be trained encodes each factual information of each historical case, obtaining a textual representation of the case facts. The neural network model also encodes each textual description of the case details for each historical case, obtaining a textual representation of the case details. For each textual representation of the case facts of each historical case, a positive sample pair is constructed with each textual representation of the case details of the same historical case. For each textual representation of each historical case (including both the case factual and case detail textual representations), a negative sample pair is constructed with each textual representation of different historical cases (including both the case factual and case detail textual representations). Based on the constructed positive and negative sample pairs, a first contrastive loss and a first gradient are calculated, and the model parameters of the neural network model are updated through backpropagation based on the first gradient. By minimizing the first contrastive loss, the distance between the textual representations of the case facts and case detail textual representations of the same historical case in the positive sample pair can be narrowed, while the distance between the textual representations of different historical cases in the negative sample pair can be widened. After the contrastive learning training is completed, an initial encoding model is obtained, enabling the textual representations encoded by the initial encoding model to correspond case factual information with case detail textual representations.

[0131] Step S403: Using the case description text with event annotation information, perform comparative learning on the initial coding model to obtain a comparative learning coding model. The comparative learning coding model is used to encode sentences and event fragments in the case description text.

[0132] After obtaining the initial encoding model, this step involves acquiring event annotation information from the case description texts of historical cases. This event annotation information indicates the event to which the case description text belongs. Using the initial encoding model, the case description texts with event annotation information are encoded to obtain text representations of the case description texts. Based on the event annotation information from the case description texts of historical cases, positive sample pairs are constructed using the text representations of case description texts of the same event, and negative sample pairs are constructed using the text representations of case description texts of different events. The initial encoding model is then compared and learned using the positive and negative sample pairs to obtain a contrastive learning encoding model.

[0133] Specifically, event annotation can be performed on the case description texts of historical cases to identify the event information to which each text fragment (or sentence) belongs. Using an initial encoding model, the text fragment (or sentence) of each event in the case description text of each historical case is encoded, resulting in a text representation of the text fragment (or sentence). For each text fragment (or sentence) of each event in each historical case, a positive sample pair is constructed with the text representations of other text fragments (or sentences) of the same event; and a negative sample pair is constructed with the text representations of text fragments (or sentences) of different events. Using the constructed positive and negative sample pairs, the initial encoding model is subjected to contrastive learning, calculating a second contrastive loss and a second gradient. The model parameters of the initial encoding model are updated based on backpropagation using the second gradient. By minimizing the second contrastive loss, the distance between the text representations of text fragments (or sentences) of the same event in positive sample pairs can be reduced, while the distance between the text representations of text fragments (or sentences) of different events in negative sample pairs can be increased. After the contrastive learning training is completed, a trained contrastive learning encoding model is obtained.

[0134] The contrastive learning coding model, trained by comparing case factual information with case description text, is more effective than traditional retrieval and text similarity comparison models for event segmentation and clustering in cases. It can improve the accuracy of event segmentation and clustering, thereby improving the accuracy of event segmentation and the extracted event information.

[0135] In this embodiment, an initial encoding model is obtained by comparing and learning a neural network model using case fact information and case description text from historical cases. This enables the text representation encoded by the initial encoding model to correspond the case fact information with the case description text. Furthermore, the initial encoding model is compared and learned using case description text with event annotation information to obtain a comparative learning encoding model. This enables the text representation encoded by the comparative learning encoding model to reflect the relevance characteristics of whether the text content belongs to the same event. Event segmentation and clustering based on the text representation encoded by the comparative learning encoding model can improve the accuracy of event segmentation and clustering, thereby improving the accuracy of event segmentation and the extracted event information, as well as the accuracy of case fact information.

[0136] Based on any of the above method embodiments, this embodiment can support interactive viewing of case description text. When a user views any case description text of a case, and the user selects a specified text segment, the system automatically locates text paragraphs in other case description files of the current case that describe the same event as the specified text. Since they describe the same event, the automatically located text paragraphs can corroborate the specified text selected by the user, making it convenient for the user to interactively view the descriptions of the same event distributed across different files.

[0137] Figure 5 A flowchart illustrating an exemplary embodiment of this application, showing an interactive viewing of a case description document. Figure 5 As shown, the specific steps to achieve interactive viewing of the case description document are as follows:

[0138] Step S501: In response to the association query operation of the first text in the first case description file, search for the second text associated with the first text in the second case description file.

[0139] In practical applications, when a user views any case description document, they can select a portion of the text (i.e., the first text) and trigger a related query operation. For example, in response to a user selecting a portion of the text (i.e., the first text) of the currently viewed case description document, the system can display a related query operation control on the currently displayed page. When the user clicks this control, a related query operation on the selected first text is triggered.

[0140] The first case description document can be any case description document for the case. The second case description document refers to any case description document for the current case other than the first case description document. In a special case, the second case description document can be the second text describing the same event in the same case description document, which is associated with the first text.

[0141] Optionally, when searching for the second text associated with the first text in the second case description file, this can be achieved in the following way:

[0142] Identify the first event segment containing the first text; based on the semantic similarity between each event segment in the second description file and the first event segment, identify the second event segment associated with the first event segment, and use the second event segment as the second text associated with the first text.

[0143] Specifically, the first text is encoded using a contrastive learning coding model to obtain a text representation of the first text. The similarity between the text representation of the first text and the text representations of each event segment in the second description file is calculated as the semantic similarity between the first text and each event segment in the second description file. Event segments in the second description file whose semantic similarity to the first text is greater than or equal to a preset semantic similarity threshold are designated as second event segments associated with the first event segments and as second texts associated with the first text.

[0144] Optionally, when searching for the second text associated with the first text in the second case description file, the first event segment containing the first text and the cluster containing the first event segment (the cluster can determine the corresponding event) can be determined based on the text segments (i.e., event segments) determined by event segmentation in the foregoing embodiments and the clusters obtained by event clustering. Other event segments contained in the cluster containing the first event segment are taken as second event segments and as the second text associated with the first text.

[0145] For example, when a user views an interrogation record of a case, they select the first text describing event A and trigger a related query operation. In response to this related query operation, the system searches for text fragments belonging to the same event A in other interrogation records and interrogation records of the same case. These fragments are then displayed as second texts associated with the first text. The system also locates the area containing the second text to show the second text associated with the first text for the user to view and refer to.

[0146] Optionally, when searching for a second text associated with the first text in the second case description file, the sentences associated with the first text can be determined based on the semantic similarity between the sentences contained in the first text and the sentences contained in the first text. For example, the semantic similarity between the sentences in the second description file and the sentences contained in the first text can be determined, and one or more sentences with high semantic similarity can be selected as the second text associated with the first text.

[0147] Step S502: Output related query prompt information. The related query prompt information is used to indicate that there is a second text in the second case description file that is related to the first text.

[0148] After finding the second text associated with the first text in the second case description file, a related query prompt message can be output to indicate to the user that there is a second text associated with the first text in the second case description file, so that the user can confirm whether to display the second text.

[0149] Optionally, when searching for a second text, the search can be performed on the second text with the highest semantic similarity. While displaying the related query prompt on the display page, a confirmation control for displaying the second text can also be shown on the display page. Clicking this control triggers the display of the second text.

[0150] Optionally, when searching for secondary text, multiple secondary texts with high semantic similarity can be found. While displaying related query suggestions on the page, key information (such as name, title, etc.) of the secondary case description documents containing the secondary text can be provided for the user to select. The user can click on any of the secondary case description documents to trigger the display of the secondary text within that document.

[0151] Step S503: In response to the display operation of the second text, display the area where the second text is located in the second case description file.

[0152] In response to any display operation of the second text, the second case description file containing the second text is opened, and the area where the second text is located in the second case description file is located is positioned and displayed so that the user can directly view the second text without having to manually locate the area where the second text is located.

[0153] The solution provided in this embodiment, in response to the association query operation of the first text in the first case description file, searches for the second text associated with the first text in the second case description file, outputs association query prompt information, and in response to the display operation of the second text, displays the area where the second text is located in the second case description file. It can realize interactive viewing of semantically related texts in different case description files, which is convenient for users to view, improves the humanization and customization functions of the case information processing system, reduces the operation of users manually flipping through and searching for related content, and assists users in reading and understanding transcripts and evidence.

[0154] The case information processing solution provided in this application, based on artificial intelligence technology, organizes, associates, analyzes, and visualizes case description documents such as written records, automatically extracts case fact information, and can assist relevant personnel in analyzing the case, allowing them to focus more on the analysis of details. This can alleviate the problem of insufficient police officers and judges to a certain extent, improve case handling efficiency, and achieve the separation of simple and complex cases.

[0155] The case information processing method provided in this application embodiment can be executed in the cloud, where several computing nodes (cloud servers) can be deployed. Each computing node has processing resources such as computing and storage. In the cloud, multiple computing nodes can be organized to provide a certain service; of course, a single computing node can also provide one or more services. The cloud provides the service by providing a service interface to the outside world, which users call to use the corresponding service.

[0156] According to the solution provided in this application embodiment, the cloud can provide a service interface with a defined service (case information processing service). Users call this service interface through their terminal devices to send a case information processing request to the cloud, which includes the identification information of the case to be processed. The cloud determines the computing node that responds to the request and uses the processing resources in that computing node to perform the following steps:

[0157] Obtain multiple case description files for the cases to be processed, and extract the case description text from the case description files;

[0158] Based on the relevance of sentences in each case description text, each case description text is divided into text segments, where sentences in the same text segment correspond to the same event;

[0159] Based on the correlation between text fragments, the text fragments divided into multiple case description documents are clustered to obtain multiple clusters, and each cluster contains at least one text fragment corresponding to the same event;

[0160] The corresponding event information is determined based on the text fragments contained in each cluster, and the event information includes at least the time element of the event;

[0161] The event information is sorted according to the time element to generate case fact information;

[0162] The case facts information will be output to the terminal device for display.

[0163] The above execution process can be referred to the relevant descriptions in the other embodiments mentioned above, and will not be repeated here.

[0164] For ease of understanding, combined with Figure 6 To illustrate with an example. Users can... Figure 6 The terminal device E1 illustrated in the diagram calls the case information processing service to upload the identification information of the cases to be processed. The service interface for users to call this service includes software development kits (SDKs) and application programming interfaces (APIs). Figure 6The diagram illustrates an API interface scenario. In the cloud, as shown, assume that service cluster E2 provides case information processing services, and service cluster E2 includes at least one computing node. Upon receiving the request, service cluster E2 executes the steps described in the previous embodiment to obtain the case's factual information and feeds it back to terminal device E1.

[0165] The terminal device E1 displays the received case fact information on the interface. It can also receive user interactions and respond accordingly.

[0166] Figure 7 This is a schematic diagram of a case information processing apparatus provided as an exemplary embodiment of this application. The case information processing apparatus provided in this embodiment can execute the processing flow provided in the case information processing method embodiment. Figure 7 As shown, the case information processing device 60 includes: a case description text acquisition module 61, an event division module 62, an event clustering module 63, an event information extraction module 64, and a case fact generation module 65.

[0167] The case description text acquisition module 61 is used to acquire multiple case description files of the case to be processed and extract the case description text from each case description file.

[0168] The event segmentation module 62 is used to divide the case description text into text segments based on the relevance of sentences in the case description text, where sentences in the same text segment correspond to the same event.

[0169] The event clustering module 63 is used to cluster the text fragments divided into multiple case description documents based on the correlation between text fragments, and obtain multiple clusters, each cluster containing at least one text fragment corresponding to the same event.

[0170] The event information extraction module 64 is used to determine the corresponding event information based on the text fragments contained in each cluster. The event information includes at least the time element of the event.

[0171] The case fact generation module 65 is used to sort event information according to time elements, generate case fact information, and output the case fact information.

[0172] In an optional embodiment, when acquiring multiple case description files of a case to be processed and extracting case description text from each case description file, the case description text acquisition module 61 is further configured to:

[0173] The system acquires image files of multiple case description documents for the case to be processed; performs text recognition on each image file to obtain a text document of the case description document; performs structural analysis on the text document and extracts the case description text from the text document.

[0174] In an optional embodiment, after extracting the case description text from the text document, the case description text acquisition module 61 is further configured to:

[0175] The denotation of the case description text is performed to replace the referential persons of the parties involved in the case description text with the names of the same parties.

[0176] In an optional embodiment, when dividing the case description text into text segments based on the relevance of sentences in the case description text, the event segmentation module 62 is further configured to:

[0177] A contrastive learning encoding model is used to encode sentences in the case description text, determining the sentence representation sequence of the case description text; a sequence labeling model is used to sequence label the sentence representation sequence of the case description text, obtaining multiple...

[0178] Five segmented sequences, each containing at least one sentence representation; a text segment is determined based on the sentence representations contained in each segmented sequence and the corresponding sentences.

[0179] In an optional embodiment, when clustering text fragments from multiple case description documents based on the correlation between text fragments to obtain multiple clusters, the event clustering module 63 is further configured to:

[0180] A contrastive learning coding model is used to encode text fragments and determine their text representations. Based on these text representations, the text fragments from multiple case description documents are clustered to obtain multiple clusters.

[0181] 0 In an optional embodiment, when determining the corresponding event information based on the text fragments contained in each cluster, the event...

[0182] The information extraction module 64 is also used for:

[0183] Semantic analysis is performed on the text fragments contained in each cluster to determine the type of the corresponding event, and the element information of the event is extracted from the text fragments contained in each cluster. The event type and element information constitute the event information.

[0184] In an optional embodiment, the case information processing device 60 further includes: a correlation query module, used for:

[0185] In response to a correlation query operation on the first text in the first case description file, search for a second text in the second case description file that is related to the first text; output correlation query prompt information, which indicates that there is a second text in the second case description file that is related to the first text; in response to a display operation on the second text, display the area where the second text is located in the second case description file.

[0186] 0 In an optional embodiment, the second text associated with the first text is located in the second case description file.

[0187] At the same time, the related query module is also used for:

[0188] Identify the first event segment containing the first text; based on the semantic similarity between each event segment in the second case description document and the first event segment, identify the second event segment associated with the first event segment, and use the second event segment as the second text associated with the first text.

[0189] 5. In an optional embodiment, the case information processing device 60 further includes:

[0190] The model training module is used to train the contrastive learning encoding model in the following way:

[0191] Obtain historical case judgments and extract factual information and case descriptions from them; based on this information and descriptions, perform comparative learning on a neural network model to obtain...

[0192] An initial encoding model was developed; using case description text with event-annotated information, the initial encoding model was subjected to comparative learning to obtain a comparatively learned encoding model. This model was then used to process sentences and event fragments in the case description text.

[0193] coding.

[0194] In an optional embodiment, when performing comparative learning on the neural network model based on the factual information and case description text of historical cases to obtain an initial encoded model, the model training module is further used for:

[0195] Using a neural network model, the factual information and description text of historical cases are encoded separately to obtain factual text representations and description text representations of 5 historical cases. Positive sample pairs are constructed based on the factual text representations and description text representations of the same historical case, and negative sample pairs are constructed based on the text representations of different historical cases. The neural network model is compared and learned using the positive and negative sample pairs to obtain the initial encoding model.

[0196] In an optional embodiment, when implementing the comparative learning of the initial encoding model using descriptive text with event annotation information to obtain a comparative learning encoding model, the model training module is further configured to:

[0197] The process involves: acquiring event annotation information from historical case description texts, which indicates the event to which the description text belongs; using an initial encoding model to encode the case description texts with event annotation information to obtain text representations of the case description texts; constructing positive sample pairs using the text representations of case description texts from the same event and negative sample pairs using the text representations of case description texts from different events based on the event annotation information from historical case description texts; and using the positive and negative sample pairs to perform comparative learning on the initial encoding model to obtain a comparative learning encoding model.

[0198] The apparatus provided in this application embodiment can be specifically used to execute the method provided in any of the above method embodiments, and the specific functions and technical effects achieved will not be elaborated here.

[0199] Figure 8 This is a schematic diagram of the structure of an electronic device provided in an example embodiment of this application. Figure 8 As shown, the electronic device 70 includes a processor 701 and a memory 702 communicatively connected to the processor 701, the memory 702 storing computer execution instructions.

[0200] The processor executes computer execution instructions stored in the memory to implement the solution provided in any of the above method embodiments. The specific functions and technical effects that can be achieved will not be elaborated here.

[0201] This application also provides a computer-readable storage medium storing computer-executable instructions. When executed by a processor, the computer-executable instructions are used to implement the solution provided in any of the above method embodiments. The specific functions and technical effects to be achieved are not described here.

[0202] This application also provides a computer program product, which includes a computer program stored in a readable storage medium. At least one processor of the electronic device can read the computer program from the readable storage medium. The at least one processor executes the computer program to cause the electronic device to perform the solution provided in any of the above method embodiments. The specific functions and technical effects that can be achieved are not described here.

[0203] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation portals are provided for users to choose to authorize or refuse.

[0204] Furthermore, in some of the processes described in the above embodiments and accompanying drawings, multiple operations appear in a specific order. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The sequence numbers are merely used to distinguish different operations, and the sequence number itself does not represent any execution order. Additionally, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to different types. "Multiple" means two or more, unless otherwise explicitly specified.

[0205] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the following claims.

[0206] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A method for processing case information, characterized in that, include: Obtain multiple case description files for the cases to be processed, and extract the case description text from each of the case description files; Based on the relevance of sentences in the case description text, the case description text is divided into text segments, wherein sentences in the same text segment correspond to the same event; Based on the correlation between the text fragments, the text fragments divided into the multiple case description files are clustered to obtain multiple clusters, and each cluster contains at least one text fragment corresponding to the same event; The corresponding event information is determined based on the text fragments contained in each cluster, and the event information includes at least the time element of the event; The event information is sorted according to the time element to generate case fact information, and the case fact information is output. The step of determining the corresponding event information based on the text fragments contained in each cluster includes: Semantic analysis is performed on the text fragments contained in each cluster to determine the type of the corresponding event, and the element information of the event is extracted from the text fragments contained in each cluster. The event type and element information constitute the event information.

2. The method according to claim 1, characterized in that, The process of obtaining multiple case description files for the case to be processed and extracting case description text from each of the case description files includes: Obtain image files from multiple case description documents for the case to be processed; Perform text recognition on each of the image files to obtain the text document of the case description file; The text document is structurally parsed, and the case description text in the text document is extracted.

3. The method according to claim 2, characterized in that, After extracting the case description text from the text document, the process further includes: The case description text is subjected to referential resolution, in which the referential pronouns of the parties appearing in the case description text are replaced with the names of the same parties.

4. The method according to claim 1, characterized in that, The step of dividing the case description text into text segments based on the relevance of sentences in the case description text includes: A contrastive learning coding model is used to encode sentences in the case description text to determine the sentence representation sequence of the case description text; Using a sequence labeling model, the sentence representation sequence of the case description text is sequence labeled to obtain multiple segmented sequences, each segmented sequence containing at least one sentence representation; A text segment is determined based on the sentence representations of each segmented sequence.

5. The method according to claim 1, characterized in that, The step involves clustering the text fragments from the multiple case description documents based on the correlation between the text fragments, resulting in multiple clusters, including: The text segment is encoded using a contrastive learning encoding model to determine the text representation of the text segment; Based on the text representation of the text fragments, the text fragments divided into multiple case description documents are clustered to obtain multiple clusters.

6. The method according to any one of claims 1-5, characterized in that, Also includes: In response to an association query operation on the first text in the first case description file of the case, a second text associated with the first text is found in the second case description file; Output related query prompt information, which is used to indicate that there is a second text in the second case description file that is related to the first text; In response to the display operation of the second text, the area where the second text is located in the second case description file is displayed.

7. The method according to claim 6, characterized in that, The step of searching for a second text associated with the first text in the second case description file includes: Determine the first event segment containing the first text; Based on the semantic similarity between each event fragment in the second case description document and the first event fragment, a second event fragment associated with the first event fragment is determined, and the second event fragment is used as the second text associated with the first text.

8. The method according to claim 3 or 4, characterized in that, The contrastive learning encoding model is trained in the following manner: Obtain historical case judgments and extract factual information and case descriptions from the judgments. Based on the factual information and case description text of the historical cases, the neural network model is compared and learned to obtain the initial encoding model; Using case description text with event annotation information, the initial encoding model is subjected to comparative learning to obtain the comparative learning encoding model, which is used to encode sentences and event fragments in the case description text.

9. The method according to claim 8, characterized in that, The process of comparing and learning the neural network model based on the case facts and case description text of the historical cases to obtain an initial encoding model includes: Using a neural network model, the case fact information and case description text of the historical cases are encoded respectively to obtain the case fact text representation and case description text representation of the historical cases; Positive sample pairs are constructed based on the textual representations of case facts and case descriptions of the same historical case, and negative sample pairs are constructed based on the textual representations of different historical cases. The neural network model is compared and learned using the positive and negative sample pairs to obtain an initial encoding model.

10. The method according to claim 8, characterized in that, The step of using case description text with event annotation information to perform comparative learning on the initial coding model to obtain the comparative learning coding model includes: Obtain event annotation information from the case description text of the historical case, wherein the event annotation information is used to indicate the event to which the case description text belongs; Using the initial encoding model, the case description text with event annotation information is encoded to obtain a text representation of the case description text; Based on the event annotation information of the case description text of the historical cases, positive sample pairs are constructed using the text representation of the case description text of the same event, and negative sample pairs are constructed using the text representation of the case description text of different events. The initial coding model is subjected to comparative learning using the positive and negative sample pairs to obtain the comparative learning coding model.

11. A case information processing device, characterized in that, include: The case description text acquisition module is used to acquire multiple case description files of the case to be processed, and extract the case description text from each of the case description files; The event segmentation module is used to divide the case description text into text segments based on the relevance of sentences in the case description text, wherein sentences in the same text segment correspond to the same event; The event clustering module is used to cluster the text fragments divided into multiple case description files according to the correlation between the text fragments, to obtain multiple clusters, and each cluster contains at least one text fragment corresponding to the same event; An event information extraction module is used to determine the corresponding event information based on the text fragments contained in each cluster, wherein the event information includes at least the time element of the event; The case fact generation module is used to sort the event information according to time elements, generate case fact information, and output the case fact information; The event information extraction module is specifically used for: Semantic analysis is performed on the text fragments contained in each cluster to determine the type of the corresponding event, and the element information of the event is extracted from the text fragments contained in each cluster. The event type and element information constitute the event information.

12. An electronic device, characterized in that, include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory to implement the method as described in any one of claims 1-10.

13. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1-10.

14. A computer program product, characterized in that, The computer program product stores computer execution instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1-10.

Images