Case information processing method and system based on document generation model

Abstract

The invention discloses a case information processing method and system based on a document generation model, and the method comprises the steps: carrying out the data analysis and processing of case information through employing a semantic understanding and data analysis model based on a deep learning technology, extracting a named entity from the case information, determining the type of the entity, and carrying out the processing of the named entity. Meanwhile, key events are extracted from the case information, corresponding legal terms are intelligently matched based on semantic features of the key events, and then the obtained legal terms, key events, entities and entity category information are integrated into a designed case related document template through a condition generation model to generate case related documents. According to the invention, the efficiency and accuracy of case information processing can be significantly improved through an automation technology, the burden of law enforcement officers is effectively reduced, and the timeliness of case processing is ensured.

Description

Technical Field

[0001] The present application relates to the field of intelligent matching, and more specifically, to a case information processing method and system based on a document generation model. Background Art

[0002] In judicial practice, case information, as an important legal document that records the process of case investigation, questioning, and trial, carries a large amount of key information and evidence, and plays a vital role in the subsequent handling of the case, the application of law, and the final judgment.

[0003] The traditional way of processing case information is highly dependent on manual reading, understanding and summarizing, and then manually writing case-related documents. This process is not only time-consuming and labor-intensive, inefficient, but also easily affected by human factors, resulting in inconsistent standards and increasing the workload of law enforcement personnel.

[0004] In recent years, with the rapid development of artificial intelligence technology, automated case information processing methods based on natural language processing and machine learning have gradually become a research hotspot. Therefore, in order to overcome the shortcomings of traditional case information processing methods, a case information processing method based on document generation models is needed. Summary of the invention

[0005] In order to solve the above technical problems, this application is proposed. The embodiment of the present application provides a case information processing method and system based on a document generation model, which uses a semantic understanding and data analysis model based on deep learning technology to perform data analysis and processing on case information, extract named entities from the case information, and determine entity categories, while extracting key events from the case information, and intelligently matching corresponding legal clauses based on the semantic features of key events, and then using a conditional generation model to integrate the acquired legal clauses, key events, entities, and entity category information into the designed case-related document template to generate case-related documents, which can significantly improve the efficiency and accuracy of case information processing through automation technology, effectively reduce the burden on law enforcement personnel, and ensure the timeliness of case processing.

[0006] According to one aspect of the present application, a case information processing method based on a document generation model is provided, which includes: Extracting case information to be processed from a case information data set; Performing text cleaning on the case information to be processed to obtain cleaned case information; Performing named entity recognition on the cleaned case information to identify entities in the cleaned case information to obtain a set of entities; Passing each entity in the entity set through an entity classification model to obtain category information of each entity in the entity set; Extracting key events from the pending case information; Performing semantic coding on the key event to obtain a key event semantic coding feature vector; Determine the legal clauses matching the case based on the semantic coding feature vector of the key event; Designing a template for case-related documents, wherein the template includes a fixed format and a dynamic fill-in part; The conditional generation model is used to integrate the legal clauses matching the case, the key events, the set of entities and the category information of each entity in the set of entities into the template of the case-related document to obtain the case-related document.

[0007] According to another aspect of the present application, a case information processing system based on a document generation model is provided, which includes: An information extraction module, used for extracting case information to be processed from a case information data set; A text cleaning module, used for performing text cleaning on the case information to be processed to obtain cleaned case information; a named entity recognition module, used for performing named entity recognition on the cleaned case information to identify entities in the cleaned case information to obtain a set of entities; An entity classification module, used for passing each entity in the entity set through an entity classification model to obtain category information of each entity in the entity set; A key event extraction module, used to extract key events from the case information to be processed; A key event semantic coding module, used for semantically coding the key event to obtain a key event semantic coding feature vector; A legal clause determination module, used to determine the legal clauses matching the case based on the key event semantic encoding feature vector; A template design module, used to design a template for case-related documents, wherein the template includes a fixed format and a dynamic filling part; The case-related document generation module is used to use the conditional generation model to integrate the legal terms matching the case, the key events, the set of entities and the category information of each entity in the set of entities into the template of the case-related document to obtain the case-related document.

[0008] The present application has at least the following technical effects: Compared with the prior art, the present application provides a case information processing method and system based on a document generation model, which performs data analysis and processing on case information by adopting a semantic understanding and data analysis model based on deep learning technology, extracts named entities from the case information, and determines the entity category, while extracting key events from the case information, and intelligently matches the corresponding legal clauses based on the semantic features of the key events, and then uses the conditional generation model to integrate the acquired legal clauses, key events, entities and entity category information into the designed case-related document template to generate case-related documents. It can significantly improve the efficiency and accuracy of case information processing through automation technology, effectively reduce the burden on law enforcement personnel, and at the same time ensure the timeliness of case processing and improve the quality and efficiency of law enforcement work. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0010] Figure 1 is a flowchart of a case information processing method based on a document generation model according to an embodiment of the present application; Figure 2 A data flow diagram of a case information processing method based on a document generation model according to an embodiment of the present application; Figure 3 It is a flowchart of sub-step S7 of the case information processing method based on the document generation model according to an embodiment of the present application; Figure 4 It is a block diagram of a case information processing system based on a document generation model according to an embodiment of the present application. DETAILED DESCRIPTION

[0011] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0012] As shown in this application and claims, unless the context clearly indicates an exception, the words "a", "an", "an" and / or "the" do not refer to the singular and may also include the plural. Generally speaking, the terms "include" and "comprise" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0013] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are only illustrative, and different aspects of the system and method can use different modules.

[0014] Flowcharts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, various steps may be processed in reverse order or simultaneously as required. Meanwhile, other operations may also be added to these processes, or a certain step or several steps of operations may be removed from these processes.

[0015] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0016] The traditional case information processing method is highly dependent on manual reading, understanding and summarizing, and then manually writing relevant documents. This process is not only time-consuming and labor-intensive, but also inefficient and easily affected by human factors, making it difficult to ensure the accuracy and efficiency of information extraction. This not only increases the workload of law enforcement personnel, but may also affect the timeliness of case processing. In recent years, with the rapid development of artificial intelligence technology, automated case information processing methods based on natural language processing and machine learning have gradually become a research hotspot. Therefore, in order to overcome the shortcomings of traditional case information processing methods, a case information processing method based on a document generation model is proposed.

[0017] In the technical solution of the present application, a case information processing method based on a document generation model is proposed. Figure 1 It is a flowchart of a case information processing method based on a document generation model according to an embodiment of the present application. Figure 2 Schematic diagram of data flow of the case information processing method based on the document generation model according to the embodiment of the present application. Figure 1 and Figure 2As shown, according to the embodiment of the present application, the case information processing method based on the document generation model includes the following steps: S1, extracting the case information to be processed from the case information data set; S2, performing text cleaning on the case information to be processed to obtain the cleaned case information; S3, performing named entity recognition on the cleaned case information to identify the entities in the cleaned case information to obtain a set of entities; S4, respectively passing each entity in the set of entities through the entity classification model to obtain the category information of each entity in the set of entities; S5, extracting key events from the case information to be processed; S6, performing semantic encoding on the key events to obtain a key event semantic encoding feature vector; S7, determining the legal clauses matching the case based on the key event semantic encoding feature vector; S8, designing a template for case-related documents, the template including a fixed format and a dynamic filling part; S9, using the conditional generation model to integrate the legal clauses matching the case, the key events, the set of entities and the category information of each entity in the set of entities into the template of the case-related document to obtain the case-related document.

[0018] In particular, S1 and S2 extract the case information to be processed from the case information data set; and perform text cleaning on the case information to be processed to obtain the cleaned case information. The case information data set may be electronic documents from various sources, such as court records, police reports, witness statements, etc. Since case information may come from different sources or different recorders, it may contain various noises, such as irrelevant characters, punctuation errors, spelling errors, inconsistent formats, etc. In order to improve the accuracy and efficiency of subsequent data processing, this application further performs text cleaning on the case information to be processed, removes spelling errors, and unifies the information format to obtain more standardized cleaned case information.

[0019] In particular, the S3 performs named entity recognition on the cleaned case information to identify entities in the cleaned case information to obtain a set of entities. Considering that when generating case-related documents, it is necessary to accurately quote relevant entities in the case information. Therefore, the present application further uses named entity recognition technology to process the cleaned case information to identify entities in the cleaned case information, such as the name of the person being questioned, the ID number of the person being questioned, the police interrogator, the police recorder, the cause of the case, the identity of the person being questioned, and other key information of all relevant personnel to obtain a set of entities. Through automated named entity recognition technology, the workload of manual reading and analysis of case information can be greatly reduced, and the efficiency and accuracy of information extraction can be improved.

[0020] In particular, the S4, respectively passes each entity in the set of entities through the entity classification model to obtain the category information of each entity in the set of entities. In the technical solution of the present application, in order to further subdivide the identified entities into specific categories to clarify the role and attributes of each entity in the case, such as distinguishing names of people, places, names of organizations, etc., the present application further adopts an entity classification model to classify each entity in the set of entities to obtain the category information of each entity in the set of entities, thereby providing more accurate basic data for subsequent document generation and improving the accuracy and professionalism of case-related document generation. In an embodiment of the present application, the entity classification model is trained and implemented using a conditional random field CRF model.

[0021] In particular, the S5 extracts key events from the pending case information. It should be understood that in order to further focus on the core information and important plots of the case, thereby providing strong information support for the legal analysis and document writing of the case, in the technical solution of the present application, key events are further extracted from the pending case information. Among them, key events represent important plots or behaviors in the case and are the core information for case handling. By extracting and analyzing key events, the main facts and development context of the case can be quickly located, and the applicable legal basis can be found more accurately, providing a solid legal basis for the generation of case-related documents.

[0022] In particular, the S6 performs semantic encoding on the key events to obtain a key event semantic encoding feature vector. Among them, in order to further determine the legal terms related to the key events, the present application first performs semantic encoding processing on the key events, adopts a pre-trained language model to capture the intrinsic meaning and contextual relationship of the key events, and converts it into a numerical form understandable by a computer through word embedding coding technology and self-attention mechanism to obtain a key event semantic encoding feature vector, thereby providing data support for subsequent legal terms query matching and achieving a deep understanding and feature expression of key events. In an embodiment of the present application, the BERT (Bidirectional EncoderRepresentations from Transformers) model is used to perform semantic encoding processing on the key events. It is worth mentioning that BERT uses unsupervised learning to perform pre-training on a large amount of text data. During the pre-training process, BERT learns to encode words and their context in the text, thereby capturing the relationship between words and the overall meaning of the text.

[0023] In particular, the S7 determines the legal clauses matching the case based on the key event semantic encoding feature vector. Figure 3As shown, S7 includes: S71, obtaining alternative legal clauses; S72, semantically encoding the alternative legal clauses to obtain a semantic encoding feature vector of the alternative legal clauses; S73, performing feature fine-grained interactive matching on the key event semantic encoding feature vector and the alternative legal clause semantic encoding feature vector to obtain a key event-alternative legal clause semantic fine-grained optimal alignment fusion feature vector; S74, determining whether to use the alternative legal clause as a legal clause matching the case based on the key event-alternative legal clause semantic fine-grained optimal alignment fusion feature vector.

[0024] Specifically, the S71 and S72 obtain alternative legal clauses; and semantically encode the alternative legal clauses to obtain a semantically encoded feature vector of the alternative legal clauses. In the technical solution of the present application, in order to query and match the key event with the applicable legal clauses, the alternative legal clauses are first obtained, and then the alternative legal clauses are semantically encoded using a pre-trained language model to fully explore the legal logic and content meaning of the alternative legal clauses, convert them into a numerical form consistent with the semantically encoded feature vector of the key event, and generate a semantically encoded feature vector of the alternative legal clauses to facilitate subsequent matching calculations.

[0025] Specifically, the S73 performs fine-grained interactive feature matching on the key event semantic encoding feature vector and the alternative legal clause semantic encoding feature vector to obtain a key event-alternative legal clause semantic fine-grained optimal alignment and fusion feature vector. In order to improve the accuracy of semantic feature matching between the key events and the alternative legal clauses, the present application proposes a fine-grained feature interactive matching method, which performs fine-grained decoupling and interactive matching calculations on the key event semantic encoding feature vector and the key event semantic encoding feature vector to achieve optimal alignment and fusion between key events and alternative legal clauses.

[0026] In an embodiment of the present application, the key event semantic coding feature vector and the alternative legal clause semantic coding feature vector are subjected to feature fine-grained interactive matching to obtain a key event-alternative legal clause semantic fine-grained optimal alignment fusion feature vector, including: first, based on a predetermined scale, fine-grained feature decoupling is performed on the key event semantic coding feature vector and the alternative legal clause semantic coding feature vector to obtain a set of key event sub-component semantic feature vectors and a set of alternative legal clause sub-component semantic feature vectors; through fine-grained feature decoupling, the key event semantic coding feature vector and the alternative legal clause semantic coding feature vector are decomposed into multiple sub-components to form a set of key event sub-component semantic feature vectors and a set of alternative legal clause sub-component semantic feature vectors, so as to facilitate feature comparison at a lower level, thereby capturing more subtle feature differences and improving the accuracy of feature matching and interaction. Next, the set of the key event sub-component semantic feature vectors and the set of the alternative legal clause sub-component semantic feature vectors are optimally paired and multi-dimensionally interacted to obtain a set of key event-alternative legal clause optimal sub-component semantic feature pairing interaction fusion vectors; specifically, the hyperbolic space distance measurement factor between any key event sub-component semantic feature vector and any alternative legal clause sub-component semantic feature vector in the set of the key event sub-component semantic feature vectors and the set of the alternative legal clause sub-component semantic feature vectors is first calculated; then, based on the hyperbolic space distance measurement factor, the key event sub-component semantic feature vector is optimally paired and multi-dimensionally interacted to obtain a set of key event-alternative legal clause optimal sub-component semantic feature pairing interaction fusion vectors. The set of feature vectors and the set of semantic feature vectors of the alternative legal clause sub-components are screened for the best feature sub-component pairing to obtain the set of optimal {key event sub-component semantic feature vector, alternative legal clause sub-component semantic feature vector} pairs; then each optimal {key event sub-component semantic feature vector, alternative legal clause sub-component semantic feature vector} pair in the set of optimal {key event sub-component semantic feature vector, alternative legal clause sub-component semantic feature vector} pairs is input into the feature multi-dimensional interaction module to obtain the set of interactive fusion vectors of the key event-alternative legal clause optimal sub-component semantic feature pairings. Specifically, after feature decoupling, the hyperbolic space distance measurement method is used to calculate the hyperbolic space distance measurement factor between any key event sub-component semantic feature vector and the alternative legal clause sub-component semantic feature vector to capture the semantic similarity between the key event and the alternative legal clause, thereby providing guidance for subsequent feature optimal matching queries.Then, based on the hyperbolic space distance measurement factor, the set of key event sub-component semantic feature vectors and the set of alternative legal clause sub-component semantic feature vectors are screened for the best feature sub-component pairing to screen out a pair of feature vectors with the smallest distance in the hyperbolic space, thereby ensuring that the subsequent interactive fusion process retains the most valuable feature information and achieves the best match. Furthermore, by performing multi-dimensional interaction on each of the selected best pairs, a deeper feature association between the two is learned to enhance the expressive power of the features. Finally, the set of key event-alternative legal clause optimal sub-component semantic feature pairing interactive fusion vectors is subjected to sub-component feature semantic fusion to obtain the key event-alternative legal clause semantic fine-grained optimal alignment fusion feature vector. In a specific example, the set of key event-alternative legal clause optimal sub-component semantic feature pairing interactive fusion vectors is cascaded to obtain the key event-alternative legal clause semantic fine-grained optimal alignment fusion feature vector. Among them, by cascading all the best paired interaction results to integrate the interaction information of all sub-components, a global key event-alternative legal clause semantic fine-grained optimal alignment fusion feature vector is generated to more comprehensively characterize the semantic relationship between key events and alternative legal clauses.

[0027] Among them, based on the hyperbolic space distance measurement factor, the set of key event sub-component semantic feature vectors and the set of alternative legal clause sub-component semantic feature vectors are screened for the best feature sub-component pairing to obtain the optimal set of {key event sub-component semantic feature vectors, alternative legal clause sub-component semantic feature vectors} pairs, and the process includes: for each key event sub-component semantic feature vector in the set of key event sub-component semantic feature vectors, selecting the alternative legal clause sub-component semantic feature vector with the smallest hyperbolic space distance measurement factor between it and the set of alternative legal clause sub-component semantic feature vectors as the pairing object to obtain the optimal set of {key event sub-component semantic feature vectors, alternative legal clause sub-component semantic feature vectors} pairs.

[0028] More specifically, the process of inputting each optimal {key event sub-component semantic feature vector, alternative legal clause sub-component semantic feature vector} pair in the set of the optimal {key event sub-component semantic feature vector, alternative legal clause sub-component semantic feature vector} pairs into the feature multi-dimensional interaction module to obtain the set of the key event-alternative legal clause optimal sub-component semantic feature pairing interaction fusion vectors includes: respectively calculating the position point addition, position point subtraction and position point multiplication between the key event sub-component semantic feature vector and the alternative legal clause sub-component semantic feature vector in the optimal {key event sub-component semantic feature vector, alternative legal clause sub-component semantic feature vector} pairing to obtain the first semantic interaction result, the second semantic interaction result and the third semantic interaction result; calculating the position weighted sum between the first semantic interaction result, the second semantic interaction result and the third semantic interaction result to obtain the key event-alternative legal clause optimal sub-component semantic feature pairing interaction fusion vector.

[0029] In summary, in the above embodiment, the key event semantic coding feature vector and the alternative legal clause semantic coding feature vector are subjected to feature fine-grained interactive matching to obtain a key event-alternative legal clause semantic fine-grained optimal alignment fusion feature vector, including: processing the key event semantic coding feature vector and the alternative legal clause semantic coding feature vector using the following feature fine-grained matching interactive formula to obtain the key event-alternative legal clause semantic fine-grained optimal alignment fusion feature vector, wherein the feature fine-grained matching interactive formula is:

[0030]

[0031]

[0032]

[0033]

[0034]

[0035] in, represents the semantic encoding feature vector of the key event, represents the semantic encoding feature vector of the alternative legal clause, Indicates decoupling processing, , , and Respectively represent the first, second, and and Key event sub-component semantic feature vectors, The value of is the number of semantic feature vectors of the key event sub-components, , , , Respectively represent the first, second, and and The semantic feature vectors of the alternative legal clause sub-components, represents the inverse hyperbolic cosine function in hyperbolic space, represents the norm of the eigenvector, Indicates the The semantic feature vector of the key event subcomponent and the The hyperbolic space distance measurement factor between the semantic feature vectors of the alternative legal clause sub-components, represents the search for the index of the semantic feature vector of the alternative legal clause subcomponent that minimizes the hyperbolic space distance metric factor, represents the index of the semantic feature vector of the alternative legal clause subcomponent corresponding to the minimum hyperbolic space distance metric factor, , and For different weight parameters, represents dot product, Indicates point addition, Indicates point reduction, , , and Respectively represent the first, second, and and The interactive fusion vector of the optimal sub-component semantic feature pairing of key events and alternative legal clauses, Indicates cascade operation, Represents the key event-alternative legal clause semantic fine-grained optimal alignment fusion feature vector.

[0036] Specifically, the S74 determines whether to use the alternative legal clause as a legal clause matching the case based on the key event-alternative legal clause semantic fine-grained optimal alignment fusion feature vector. In a specific example of the present application, the key event-alternative legal clause semantic fine-grained optimal alignment fusion feature vector is input into a classifier-based retrieval result generator to obtain a retrieval result, and the detection result is used to indicate whether the alternative legal clause is determined as a legal clause matching the case. Among them, the classifier is optimized through training, and can effectively evaluate and classify the matching relationship between the key events and the alternative legal clauses contained in the key event-alternative legal clause semantic fine-grained optimal alignment fusion feature vector, and generate corresponding retrieval results, thereby determining whether to use the alternative legal clause as a legal clause matching the case.

[0037] That is, considering that the key event semantic encoding feature vector and the key event semantic encoding feature vector represent the encoded semantic features of the key event and the alternative legal clause respectively, when they are input into the feature interaction network based on feature fine-grained optimal matching, the semantic difference of the encoded features based on the source semantic difference will cause the feature fine-grained optimal matching difference, so that the key event-alternative legal clause semantic fine-grained optimal alignment fusion feature vector will also have a long-range class regression understanding representation imbalance due to the fusion significant difference based on the local matching degree difference, affecting the accuracy of the classification result. Preferably, inputting the key event-alternative legal clause semantic fine-grained optimal alignment fusion feature vector into a classifier-based retrieval result generator to obtain retrieval results includes: Determine the number of zero eigenvalues ​​in the key event-alternative legal clause semantic fine-grained optimal alignment fusion feature vector, and respectively calculate the reciprocal of the logarithm of the number of zero eigenvalues ​​with base two and the exponential value of the reciprocal of the number of zero eigenvalues ​​with base natural constants to obtain a first key event-alternative legal clause semantic fine-grained optimal alignment fusion representation value and a second key event-alternative legal clause semantic fine-grained optimal alignment fusion representation value; Calculate a power function of each eigenvalue of the key event-alternative legal clause semantic fine-grained optimal alignment fusion feature vector with the inverse of the number of zero eigenvalues ​​as the exponent to obtain a key event-alternative legal clause semantic fine-grained optimal alignment fusion structure state vector; Performing a dot multiplication of the key event-alternative legal clause semantic fine-grained optimal alignment fusion structure state vector and the first key event-alternative legal clause semantic fine-grained optimal alignment fusion representation value to obtain a key event-alternative legal clause semantic fine-grained optimal alignment fusion information representation vector; Performing dot multiplication of the autocorrelation matrix of the key event-alternative legal clause semantic fine-grained optimal alignment fusion structure state vector with the second key event-alternative legal clause semantic fine-grained optimal alignment fusion representation value and weight hyperparameter respectively to obtain the key event-alternative legal clause semantic fine-grained optimal alignment fusion regression understanding matrix; Performing matrix multiplication of the key event-alternative legal clause semantic fine-grained optimal alignment fusion information representation vector and the key event-alternative legal clause semantic fine-grained optimal alignment fusion regression understanding matrix to obtain an optimized key event-alternative legal clause semantic fine-grained optimal alignment fusion feature vector; The optimized key event-alternative legal clause semantic fine-grained optimal alignment fusion feature vector is input into a classifier-based retrieval result generator to obtain a retrieval result.

[0038] That is, the first key event-alternative legal clause semantic fine-grained optimal alignment fusion representation value and the second key event-alternative legal clause semantic fine-grained optimal alignment fusion representation value are expressed as:

[0039]

[0040] in, is the number of zero eigenvalues ​​in the key event-alternative legal clause semantic fine-grained optimal alignment fusion feature vector, is the first key event-alternative legal clause semantic fine-grained optimal alignment fusion representation value, is the second key event-alternative legal clause semantic fine-grained optimal alignment fusion representation value, It is an exponential function with a natural constant as its base.

[0041] And the optimized key event-alternative legal clause semantic fine-grained optimal alignment fusion feature vector is expressed as:

[0042]

[0043]

[0044] in, is the key event-alternative legal clause semantic fine-grained optimal alignment fusion feature vector, is the weight hyperparameter, is the key event-alternative legal clause semantic fine-grained optimal alignment fusion regression understanding matrix, is the key event-alternative legal clause semantic fine-grained optimal alignment fusion information representation vector, is the number of zero eigenvalues ​​in the key event-alternative legal clause semantic fine-grained optimal alignment fusion feature vector, is the first key event-alternative legal clause semantic fine-grained optimal alignment fusion representation value, is the second key event-alternative legal clause semantic fine-grained optimal alignment fusion representation value, represents a power function in which each eigenvalue of the key event-alternative legal clause semantic fine-grained optimal alignment fusion feature vector is calculated with the inverse of the number of zero eigenvalues ​​as the exponent, represents the transpose of a vector, It is the point multiplication by position. is matrix multiplication, It is the optimized key event-alternative legal clause semantic fine-grained optimal alignment fusion feature vector.

[0045] Based on this, the present application realizes effective long-range modeling of the key event-alternative legal clause semantic fine-grained optimal alignment fusion feature vector under field linear complexity by modeling the vector field zero dimension of the key event-alternative legal clause semantic fine-grained optimal alignment fusion feature vector as a global field redundant dependency. In this way, in the feature regression process of the key event-alternative legal clause semantic fine-grained optimal alignment fusion feature vector, the structured correlation self-distillation between information representation and class understanding is performed based on the selective state space of feature distribution, thereby maintaining the overall information complexity of the feature distribution of the key event-alternative legal clause semantic fine-grained optimal alignment fusion feature vector, while promoting the long-range dynamic class understanding balance of its feature convergence, so as to improve the class understanding consistency between the classification regression and feature extraction processes, and improve the accuracy of the retrieval results obtained by inputting the key event-alternative legal clause semantic fine-grained optimal alignment fusion feature vector into the classifier-based retrieval result generator.

[0046] In other specific examples of the present application, the legal clauses matching the case can also be determined based on the key event semantic encoding feature vector in other ways, for example: using natural language processing (NLP) technology to extract key events from the case text; encoding the key events into semantic vectors to represent the meaning and semantic relationship of the events; obtaining legal clauses related to the case; encoding the legal clauses into semantic vectors to represent the meaning and semantic relationship of the clauses; using cosine similarity or other similarity metrics to calculate the similarity between the key event semantic vector and the legal clause semantic vector; identifying the legal clause with the highest similarity as the legal clause matching the case.

[0047] In particular, the S8 designs a template for case-related documents, and the template includes a fixed format and a dynamic filling part. After the above processing, the entities, entity category information, key events and relevant legal clauses in the case information are obtained. Then, case-related documents can be generated based on this information. In order to ensure the standardization and accuracy of the documents, it is necessary to first design a template for case-related documents. Specifically, the template should include a fixed format and a dynamic filling part. Among them, the fixed format part of the template (such as heading, signature, date, etc.) can be pre-set, and only relevant information needs to be filled in the dynamic filling part according to the specific case situation, thereby improving the efficiency of document production. As the needs are updated and changed, the fixed format part in the template can remain unchanged, while the dynamic filling part can be adjusted accordingly according to the new needs to adapt to the law enforcement needs of different periods.

[0048] In particular, the S9, utilizes the conditional generation model to integrate the legal clauses matching the case, the key events, the set of entities, and the category information of each entity in the set of entities into the template of the case-related document to obtain the case-related document. That is, the conditional generation model is utilized to integrate the legal clauses matching the case, the key events, the set of entities, and the category information of each entity in the set of entities into the template of the case-related document to obtain the case-related document. It should be understood that the conditional generation model is a text generation technology based on deep learning, which can automatically generate coherent and contextual text content based on given conditional information. In the technical solution of the present application, the conditional generation model can understand the structure and semantics of case-related documents through training and learning, thereby generating standardized and accurate document content according to the specific circumstances of the case.

[0049] In summary, the case information processing method based on the document generation model according to the embodiment of the present application is explained, which uses a semantic understanding and data analysis model based on deep learning technology to perform data analysis and processing on the case information, extracts named entities from the case information, and determines the entity category, and extracts key events from the case information, and intelligently matches the corresponding legal terms based on the semantic features of the key events, and then uses the conditional generation model to integrate the acquired legal terms, key events, entities and entity category information into the designed case-related document template to generate case-related documents. In this way, the efficiency and accuracy of case information processing can be significantly improved through automation technology, effectively reducing the burden on law enforcement personnel, and ensuring the timeliness of case processing.

[0050] Furthermore, a case information processing system based on a document generation model is also provided.

[0051] Figure 4 FIG. 1 is a block diagram of a case information processing system based on a document generation model according to an embodiment of the present application. Figure 4As shown, according to the case information processing system 300 based on the document generation model of the embodiment of the present application, it includes: an information extraction module 310, which is used to extract the case information to be processed from the case information data set; a text cleaning module 320, which is used to perform text cleaning on the case information to be processed to obtain the cleaned case information; a named entity recognition module 330, which is used to perform named entity recognition on the cleaned case information to identify the entities in the cleaned case information to obtain a set of entities; an entity classification module 340, which is used to pass each entity in the set of entities through an entity classification model to obtain the category information of each entity in the set of entities; a key event extraction module 350, which is used to extract the case information from the case information to be processed. Process case information to extract key events; a key event semantic encoding module 360 ​​is used to semantically encode the key events to obtain key event semantic encoding feature vectors; a legal clause determination module 370 is used to determine the legal clauses that match the case based on the key event semantic encoding feature vectors; a template design module 380 is used to design a template for case-related documents, wherein the template includes a fixed format and a dynamic fill-in part; a case-related document generation module 390 is used to utilize a conditional generation model to integrate the legal clauses that match the case, the key events, the set of entities, and the category information of each entity in the set of entities into the template of the case-related document to obtain a case-related document.

[0052] As described above, the case information processing system 300 based on the document generation model according to the embodiment of the present application can be implemented in various wireless terminals, such as a server with a case information processing algorithm based on the document generation model. In a possible implementation, the case information processing system 300 based on the document generation model according to the embodiment of the present application can be integrated into the wireless terminal as a software module and / or a hardware module. For example, the case information processing system 300 based on the document generation model can be a software module in the operating system of the wireless terminal, or can be an application developed for the wireless terminal; of course, the case information processing system 300 based on the document generation model can also be one of the many hardware modules of the wireless terminal.

[0053] Alternatively, in another example, the case information processing system 300 based on the document generation model and the wireless terminal may also be separate devices, and the case information processing system 300 based on the document generation model may be connected to the wireless terminal via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0054] The embodiments of the present disclosure have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A case information processing method based on a document generation model, characterized in that: include: Extracting case information to be processed from a case information data set; Performing text cleaning on the case information to be processed to obtain cleaned case information; Performing named entity recognition on the cleaned case information to identify entities in the cleaned case information to obtain a set of entities; Passing each entity in the entity set through an entity classification model to obtain category information of each entity in the entity set; Extracting key events from the pending case information; Performing semantic coding on the key event to obtain a key event semantic coding feature vector; Determine the legal clauses matching the case based on the semantic coding feature vector of the key event; Designing a template for case-related documents, wherein the template includes a fixed format and a dynamic fill-in part; The conditional generation model is used to integrate the legal clauses matching the case, the key events, the set of entities and the category information of each entity in the set of entities into the template of the case-related document to obtain the case-related document.

2. The case information processing method based on the document generation model according to claim 1 is characterized in that: Based on the semantic coding feature vector of the key event, the legal clauses matching the case are determined, including: Obtain alternative legal terms; Performing semantic encoding on the alternative legal clauses to obtain a semantic encoding feature vector of the alternative legal clauses; Performing feature fine-grained interactive matching on the key event semantic encoding feature vector and the alternative legal clause semantic encoding feature vector to obtain a key event-alternative legal clause semantic fine-grained optimal alignment fusion feature vector; Based on the key event-alternative legal clause semantic fine-grained optimal alignment fusion feature vector, determine whether to use the alternative legal clause as a legal clause matching the case.

3. The case information processing method based on the document generation model according to claim 2 is characterized in that: Performing fine-grained interactive feature matching on the key event semantic encoding feature vector and the alternative legal clause semantic encoding feature vector to obtain a key event-alternative legal clause semantic fine-grained optimal alignment fusion feature vector, including: Based on a predetermined scale, fine-grained feature decoupling is performed on the key event semantic coding feature vector and the alternative legal clause semantic coding feature vector to obtain a set of key event sub-component semantic feature vectors and a set of alternative legal clause sub-component semantic feature vectors; Performing optimal pairing multi-dimensional interaction on the set of key event sub-component semantic feature vectors and the set of alternative legal clause sub-component semantic feature vectors to obtain a set of key event-alternative legal clause optimal sub-component semantic feature pairing interaction fusion vectors; The sub-component feature semantic fusion is performed on the set of the key event-alternative legal clause optimal sub-component semantic feature pairing interactive fusion vectors to obtain the key event-alternative legal clause semantic fine-grained optimal alignment fusion feature vector.

4. The case information processing method based on the document generation model according to claim 3 is characterized in that: The set of the key event sub-component semantic feature vectors and the set of the alternative legal clause sub-component semantic feature vectors are optimally paired and multi-dimensionally interacted to obtain a set of key event-alternative legal clause optimal sub-component semantic feature paired interaction fusion vectors, including: Calculating a hyperbolic space distance measurement factor between any key event sub-component semantic feature vector and any alternative legal clause sub-component semantic feature vector in the set of key event sub-component semantic feature vectors and the set of alternative legal clause sub-component semantic feature vectors; Based on the hyperbolic space distance measurement factor, the set of the key event sub-component semantic feature vectors and the set of the alternative legal clause sub-component semantic feature vectors are screened for optimal feature sub-component pairing to obtain an optimal set of {key event sub-component semantic feature vectors, alternative legal clause sub-component semantic feature vectors} pairs; Each optimal {key event sub-component semantic feature vector, alternative legal clause sub-component semantic feature vector} pair in the set of optimal {key event sub-component semantic feature vector, alternative legal clause sub-component semantic feature vector} pairs is input into the feature multi-dimensional interaction module to obtain a set of key event-alternative legal clause optimal sub-component semantic feature pairing interaction fusion vectors.

5. The case information processing method based on the document generation model according to claim 4 is characterized in that: Based on the hyperbolic space distance measurement factor, the set of the key event sub-component semantic feature vectors and the set of the alternative legal clause sub-component semantic feature vectors are screened for optimal feature sub-component pairing to obtain an optimal set of {key event sub-component semantic feature vectors, alternative legal clause sub-component semantic feature vectors} pairs, including: For each key event sub-component semantic feature vector in the set of key event sub-component semantic feature vectors, the alternative legal clause sub-component semantic feature vector with the smallest hyperbolic space distance measurement factor is selected from the set of alternative legal clause sub-component semantic feature vectors as a pairing object to obtain the optimal set of {key event sub-component semantic feature vector, alternative legal clause sub-component semantic feature vector} pairs.

6. The case information processing method based on the document generation model according to claim 5 is characterized in that: Inputting each optimal {key event sub-component semantic feature vector, alternative legal clause sub-component semantic feature vector} pair in the set of the optimal {key event sub-component semantic feature vector, alternative legal clause sub-component semantic feature vector} pairs into the feature multi-dimensional interaction module to obtain a set of the key event-alternative legal clause optimal sub-component semantic feature pairing interaction fusion vectors, including: Respectively calculating the position point addition, position point subtraction and position point multiplication between the key event subcomponent semantic feature vector and the alternative legal clause subcomponent semantic feature vector in the optimal {key event subcomponent semantic feature vector, alternative legal clause subcomponent semantic feature vector} pair to obtain a first semantic interaction result, a second semantic interaction result and a third semantic interaction result; The position-weighted sum of the first semantic interaction result, the second semantic interaction result and the third semantic interaction result is calculated to obtain the key event-alternative legal clause optimal sub-component semantic feature pairing interaction fusion vector.

7. The case information processing method based on the document generation model according to claim 6 is characterized in that: The sub-component feature semantic fusion is performed on the set of the key event-alternative legal clause optimal sub-component semantic feature pairing interactive fusion vectors to obtain the key event-alternative legal clause semantic fine-grained optimal alignment fusion feature vector, including: The set of the key event-alternative legal clause optimal sub-component semantic feature pairing interaction fusion vectors is cascaded to obtain the key event-alternative legal clause semantic fine-grained optimal alignment fusion feature vector.

8. The case information processing method based on the document generation model according to claim 7 is characterized in that: Based on the semantic fine-grained optimal alignment fusion feature vector of the key event-alternative legal clause, determining whether to use the alternative legal clause as a legal clause matching the case includes: The key event-alternative legal clause semantic fine-grained optimal alignment fusion feature vector is input into a classifier-based retrieval result generator to obtain a retrieval result, and the detection result is used to indicate whether the alternative legal clause is determined as a legal clause matching the case.

9. A case information processing system based on a document generation model, characterized in that: include: An information extraction module, used for extracting case information to be processed from a case information data set; A text cleaning module, used for performing text cleaning on the case information to be processed to obtain cleaned case information; a named entity recognition module, used for performing named entity recognition on the cleaned case information to identify entities in the cleaned case information to obtain a set of entities; An entity classification module, used for passing each entity in the entity set through an entity classification model to obtain category information of each entity in the entity set; A key event extraction module, used to extract key events from the case information to be processed; A key event semantic coding module, used for semantically coding the key event to obtain a key event semantic coding feature vector; A legal clause determination module, used to determine the legal clauses matching the case based on the key event semantic encoding feature vector; The template design module is used to design templates for case-related documents, wherein the templates include a fixed format and a dynamic filling part.

10. The case information processing system based on the document generation model according to claim 9 is characterized in that: Also includes: The case-related document generation module is used to use the conditional generation model to integrate the legal terms matching the case, the key events, the set of entities and the category information of each entity in the set of entities into the template of the case-related document to obtain the case-related document.

Images