Case extraction system for judicial documents

Abstract

The invention relates to the technical field of natural language processing, in particular to a cause extraction system for judicial documents, which comprises a historical information input module used for inputting historical judicial documents, a cause statistics module used for generating a corresponding relation table, a language data acquisition module used for acquiring stop word tables and text vector files of the judicial documents, and a result extraction module used for extracting the stop word tables and the text vector files of the judicial documents. The system comprises a historical information processing module for preprocessing a historical judicial document to obtain historical fact phrases and historical law article phrases, a text vector marking module for analyzing the relevancy between fact words and law articles, a core feature extraction module for extracting the relevancy between the target judicial document and the historical law article phrases, and a text feature extraction module for extracting the target judicial document. The judicial cause extraction module is used for extracting core characteristics of each historical judicial document and analyzing cause characteristics, and the judicial cause extraction module is used for carrying out law article weight sorting on a target judicial document and extracting cause types of the target judicial document. The judicial document cause extraction efficiency is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of natural language processing, and in particular to a case cause extraction system for judicial documents. Background Art

[0002] In judicial practice, a vast number of judicial documents require manual reading and processing, a crucial step in which is determining the cause of action. This process is not only time-consuming and labor-intensive, but also prone to errors, especially when faced with a large number of similar but subtly different cases. With the development of big data and artificial intelligence technologies, it has become possible to use automated tools to assist or replace manual case cause extraction. However, existing automated methods often suffer from low accuracy and poor adaptability, making them difficult to meet practical application needs.

[0003] Chinese Patent Publication No. CN108021545A discloses a method and device for extracting causes of action from judicial documents, relating to the field of computer technology. The main purpose is to extract possible causes of action from judicial documents by matching cause of action keywords according to word order, thereby avoiding erroneous extraction due to word order issues. The main technical solution of the invention is as follows: using cause of action information in a judicial document sample to obtain a set of cause of action keyword samples, wherein the set of cause of action keyword samples includes cause of action keyword samples from different types of judicial document samples; segmenting a batch of judicial documents based on the cause of action keyword samples to obtain a cause of action keyword lexicon for the batch of judicial documents, wherein the cause of action keyword lexicon contains cause of action keywords and corresponding total word frequencies; extracting cause of action keywords from the target judicial document according to word order; and determining the cause of action in the target judicial document based on the co-occurrence frequency of adjacent cause of action keywords and the total word frequency. Therefore, the invention does not analyze the judgment provisions in the judicial document, but only analyzes the factual text of the judicial document, which cannot achieve efficient extraction of judicial causes of action and has the problem of low efficiency in extracting cause of action from judicial documents. Summary of the Invention

[0004] The object of the present invention is to provide a case cause extraction system for judicial documents to solve at least one of the problems existing in the prior art.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] Historical information entry module, used to enter historical judicial documents;

[0007] Case statistics module, used to generate corresponding relationship tables based on historical judicial documents;

[0008] A language data acquisition module is used to obtain a stop word list and text vector file of a judicial document, and is also used to obtain a target judicial document;

[0009] The historical information processing module is used to pre-process historical judicial documents to obtain historical fact phrases and historical legal clause phrases;

[0010] A text vector tagging module is used to perform word vector tagging on historical fact phrases and historical legal clause phrases based on the text vector file, and to analyze the relevance between factual words and legal clauses based on the word vector tagging results;

[0011] The core feature extraction module is used to extract the core features of each historical judicial document based on historical fact phrases and historical legal phrases, and analyze the characteristics of each case based on the extraction results;

[0012] The judicial case cause extraction module is used to sort the target judicial documents by legal article weights based on the correlation analysis results between fact words and legal articles, and to extract judicial case causes based on the legal article weight sorting results and the feature analysis results of each case cause.

[0013] Furthermore, the case cause statistics module collects statistics on the corresponding types of historical case causes and historical judgment laws based on historical judicial documents;

[0014] The case cause statistics module counts the historical case cause types corresponding to each historical judgment law clause in each historical judicial document, and stores the statistical results.

[0015] Furthermore, the text vector marking module includes a vector marking unit, which performs word vector marking on the historical fact phrases and the historical legal provision phrases according to the text vector file;

[0016] The vector marking unit further calculates a mapping vector group c(i) based on the vector group a of the historical fact phrases and the vector group b(i) of the historical legal clause phrases.

[0017] Furthermore, the text vector marking module is further provided with a weight analysis unit, which is used to analyze the mapping weights according to the calculation results of each mapping vector group;

[0018] The weight analysis unit calculates the mapping index α(i) between the vector group a of the historical fact phrases and each historical legal clause phrase;

[0019] The weight analysis unit compares the mapping index α(i) with each preset mapping index, and analyzes the mapping weight according to the comparison result. The analysis results of the mapping weight include: low, normal and high.

[0020] Furthermore, the relevance analysis unit analyzes the relevance between the fact words and the legal provisions based on the analysis results of the mapping vectors; the relevance analysis unit counts the number n of historical fact phrases that appear in the same historical judgment legal provision in each historical judicial document, and calculates the occurrence frequency v(j,i) of each fact word based on the statistical results of the historical fact phrases;

[0021] The relevance analysis unit calculates the relevance β(i, j) between the fact words and the legal provisions according to the occurrence frequency of each fact word.

[0022] Furthermore, the core feature extraction module includes an event extraction unit, which divides fact words in the historical fact phrase, and the division results include entity words and relationship words;

[0023] The event extraction unit divides the historical fact phrases into minimum event phrases according to predefined event patterns.

[0024] Furthermore, the core feature extraction module further includes a core feature extraction unit, which is used to extract core features of historical judicial documents based on the analysis results of each minimum event phrase and the historical legal terms phrase;

[0025] The core feature extraction unit iteratively divides each minimum event phrase according to a predefined pattern to obtain each iterative event phrase, and counts the number of occurrences of related words in each iterative event phrase;

[0026] The core feature extraction unit uses the relational word with the largest number of occurrences in each iterative event phrase as the core feature of the historical judicial document.

[0027] Furthermore, the core feature extraction module is further provided with a case feature analysis unit, which is used to analyze each case feature based on the core feature extraction results of each historical judicial document;

[0028] The case feature analysis unit counts the number of occurrences n3(k) of the core features of the same historical case type in each historical judicial document, and analyzes each case feature based on the statistical results to determine whether the core feature is a case feature of the case type.

[0029] Furthermore, the target weight analysis unit is used to sort the target judicial documents by legal article weight according to the relevance analysis results between the fact words and the legal articles;

[0030] The target weight analysis unit pre-processes the target judicial document to obtain the target fact phrase;

[0031] The target weight analysis unit calculates the relevance β(p)' between the target fact phrase and each target judgment legal provision according to the relevance analysis result between the fact words and the legal provision;

[0032] The target weight analysis unit sorts each judgment law in descending order according to the relevance β(p)' between the fact phrase and each target law phrase, and takes the first ranked target judgment law as the target case law.

[0033] Furthermore, the judicial cause of action extraction module further includes a cause of action matching unit, which is used to extract the cause of action type of the target judicial document based on the result of the legal article weight ranking, the result of the feature analysis of each cause of action, and the corresponding relationship table;

[0034] The case cause matching unit extracts the core features of the target judicial document based on the target fact phrase;

[0035] The case cause matching unit matches the case cause type of the target judicial document according to the core features of the target judicial document, the result of the weight ranking of the legal provisions and the correspondence table: if the core features of the target judicial document successfully match the case cause features and the target case cause legal provision and the case cause type corresponding to the case cause features successfully match in the correspondence table, the case cause matching unit determines that the match is successful and stores the target judicial document in the case cause type that matches successfully; if the core features of the target judicial document successfully match the case cause features and the target case cause legal provision and the case cause type corresponding to the case cause features fail to match in the correspondence table, the case cause matching unit determines that the match fails and adds the target case cause legal provision of the target judicial document to the case cause type corresponding to the correspondence table; if the core features of the target judicial document fail to match the case cause features and the target case cause legal provision and the case cause type corresponding to the case cause features successfully match in the correspondence table, the case cause matching unit determines that the match fails; if the core features of the target judicial document fail to match the case cause features and the target case cause legal provision and the case cause type corresponding to the case cause features fail to match in the correspondence table, the case cause matching unit determines that the match fails.

[0036] Compared with the prior art, the beneficial effects of the present invention are that historical judicial documents are entered through the historical information entry module, thereby improving the completeness and accuracy of the acquisition of historical judicial documents; historical judicial documents are processed through the historical information processing module, thereby improving the accuracy of the extraction of the relevance of factual words to legal provisions and the core features of historical judicial texts; the case cause statistics module generates a corresponding relationship table based on historical judicial documents, thereby improving the accuracy of case cause extraction; the language data acquisition module obtains the stop word list and text vector file of the judicial document, which are also used to obtain target judicial documents, thereby improving the accuracy of data acquisition; the text vector labeling module analyzes the relevance of factual words to legal provisions, thereby improving the accuracy of the relevance analysis; the core feature extraction module analyzes the features of each case cause, thereby improving the accuracy of the case cause feature analysis; the judicial case cause extraction module matches and judges the case cause type, thereby improving the accuracy of the judicial case cause extraction. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0038] Figure 1 Schematic diagram of the structure of the case cause extraction system used for judicial documents in this embodiment.

[0039] Figure 2 Schematic diagram of the text vector marking module structure of this embodiment.

[0040] Figure 3 This is a schematic diagram of the core feature extraction module structure of this embodiment.

[0041] Figure 4 This is a structural diagram of the judicial case cause extraction module of this embodiment. DETAILED DESCRIPTION

[0042] In order to more clearly illustrate the present invention, the present invention is further described below in conjunction with preferred embodiments and accompanying drawings. Similar components in the accompanying drawings are represented by the same reference numerals. It should be understood by those skilled in the art that the following detailed description is illustrative rather than restrictive and should not be used to limit the scope of protection of the present invention.

[0043] It should be noted that, although the terms "first," "second," and "third" may be used to describe the embodiments of the present application, the description should not be limited to these terms. These terms are merely used to distinguish the descriptions. For example, without departing from the scope of the embodiments of the present application, "first" may also be referred to as "second," and similarly, "second" may also be referred to as "first."

[0044] Specifically, the case cause extraction system for judicial documents described in this embodiment is applied to the preliminary stage of case cause extraction and classification of judicial documents, and is used for the extraction and classification of multiple case cause types; the judicial document is specifically a judgment; the system described in this embodiment is installed in the data processing server of judicial documents of the judicial department; the case cause extraction system for judicial documents described in this embodiment strictly abides by the guiding documents related to case cause extraction issued by the Supreme People's Court, and the guiding documents include but are not limited to the "Regulations on Case Causes of Civil Cases", etc., and the collection, use and processing of relevant data in this embodiment need to comply with relevant national and regional laws, regulations and standards; the case cause types described in this embodiment include contracts, gratuitous management, unjust enrichment disputes, tort liability, action types, inaction types, administrative compensation types and various criminal charges.

[0045] See also Figure 1 As shown, it is a structural diagram of the case cause extraction system for judicial documents in this embodiment, including a document information input module, which is used to input historical judicial documents; the historical judicial documents are judicial documents that have passed the final trial, and the historical judicial documents include historical case cause types, historical fact texts and historical judgment articles; it can be understood that: the number of historical judgment articles is not unique; the historical case cause type is the case cause type confirmed by the final trial of the historical judicial document, and the historical judgment article is the content of the article involved in the historical judicial document.

[0046] Specifically, the historical judicial documents are obtained through the official website of judicial decisions; the judicial documents described in this embodiment are specifically judicial decisions; it can be understood that this embodiment does not specifically limit the number of historical judicial documents obtained, and those skilled in the art can set it freely, as long as the number of historical judicial documents is met. The number of historical judicial documents in this embodiment is 10,000. At the same time, the more historical judicial documents there are, the higher the accuracy of the case extraction in this embodiment and the longer it takes.

[0047] Please continue reading Figure 1 As shown, the system further includes a case cause statistics module, which is connected to the historical information entry module and is used to collect statistics on the corresponding types of historical case causes and historical judgment laws based on historical judicial documents;

[0048] The case cause statistics module counts the historical case cause types corresponding to each historical judgment law article in each historical judicial document, and arranges the statistical results into a corresponding relationship table. The case cause statistics module is also used to store the corresponding relationship table.

[0049] Specifically, the format of the correspondence table described in this embodiment is a two-column table, one column for the case type and the other column for the legal article entry.

[0050] Please continue reading Figure 1 As shown, the system also includes a language data acquisition module, which is used to obtain the stop word list and text vector file of the judicial document, and is also used to obtain the target judicial document; the stop word list of the judicial document can obtain the Chinese stop word list as the stop word list of the judicial document through the NLTK library of the Python language; the text vector file is a pre-trained word vector file; this embodiment does not make specific limitations on the method of obtaining the stop word list and the text vector file, and those skilled in the art can set it freely, and only need to meet the requirements for obtaining the stop word list and the text vector file. This embodiment is obtained by downloading from GitHub; at the same time, the structure of the target judicial document is the same as the structure of the historical judicial document; it can be understood that this embodiment does not make specific limitations on the word vector dimension of the text vector file, and those skilled in the art can set it freely, and only need to meet the requirements for the word vector dimension of the text vector file. This embodiment selects a pre-trained word vector file with a vector dimension of 128 as the text vector file.

[0051] See also Figure 1 As shown, the system further includes a historical information processing module, which is connected to the text information input module and is used to pre-process historical judicial documents to obtain historical fact phrases and related historical legal provisions;

[0052] The historical information processing module performs word segmentation on each historical fact text according to the stop word list to obtain each historical fact phrase;

[0053] The historical information processing module segments each historical judgment law article according to the stop word list to obtain each historical law article phrase.

[0054] Specifically, this embodiment does not specifically limit the specific processes of "segmenting each historical fact text according to a stop word list to obtain each historical fact phrase" and "segmenting each historical judgment law article according to a stop word list to obtain each historical law article phrase". Those skilled in the art can freely set them up, such as segmenting each historical fact text and each historical judgment law article through Python language, NLTK library and stop word list; the historical fact phrase includes multiple fact words; the historical law article phrase includes multiple law article words.

[0055] Please continue reading Figure 1 As shown, the system also includes a text vector marking module, which is connected to the historical information processing module and the language data acquisition module. The text vector marking module is used to mark the historical fact phrases and historical legal phrases with word vectors according to the text vector file, and analyze the mapping vectors of the historical fact phrases and the historical legal phrases according to the word vector marking results, and also analyze the relevance between fact words and legal provisions according to the analysis results of the mapping vectors.

[0056] See also Figure 2 As shown, the text vector marking module includes a vector marking unit, which performs word vector marking on historical fact phrases and historical legal phrases according to the text vector file;

[0057] The vector marking unit further calculates a mapping vector group c(i) based on the vector group a of the historical fact phrases and the vector group b(i) of the historical legal clause phrases, and sets c(i)=b(i) / a;

[0058] Among them, c(i) represents the mapping vector group a of historical fact phrases and the i-th historical legal provision phrase.

[0059] Specifically, the calculation process of the mapping vector c(i) is: a=[a1,a2,a3...], b(i)=[b1,b2,b3...bN], c(i)={c1,c2,c3...cN}, c1=min{b1 / a1,b1 / a2,b1 / a3...}, c2=min{b2 / a1,b2 / a2,b2 / a3...}, c3=min{b3 / a1,b3 / a2,b3 / a3...}.. .cN=min{bN / a1,bN / a2,bN / a3...}; N is the number of legal words in the vector group b(i) of the i-th historical legal word group, a1 is the word vector of the first historical fact word, a2 is the word vector of the second historical fact word, a3 is the word vector of the third historical fact word, b1 is the word vector of the first historical legal word, b2 is the word vector of the second historical legal word, b3 is the word vector of the third historical legal word, and bN is the word vector of the N-th historical legal word.

[0060] Please continue reading Figure 2 As shown, the text vector marking module is further provided with a weight analysis unit, which is connected to the vector marking unit and is used to analyze the mapping weight according to the calculation results of each mapping vector group;

[0061] The weight analysis unit calculates the mapping index of the vector group a of the historical fact phrases and each historical legal provision phrase, and sets the mapping index of the vector group a of the historical fact phrases and the i-th historical legal provision phrase to α(i), setting α(i)=[c1 2 +c2 2 +...cN 2 ] 1 / 2 ;

[0062] The weight analysis unit compares the mapping index α(i) with each preset mapping index, and analyzes the mapping weight according to the comparison result: if α(i) < A1, the weight analysis unit determines that the mapping weight of the historical fact phrase to the legal article is high; if A1 ≤ α(i) < A2, the weight analysis unit determines that the mapping weight of the historical fact phrase to the legal article is normal; if α(i) ≥ A2, the weight analysis unit determines that the mapping weight of the historical fact phrase to the legal article is low;

[0063] Wherein, A1 is a first preset mapping index, A2 is a second preset mapping index, and A1<A2;

[0064] The weight analysis unit sorts each historical judgment article in descending order according to the mapping index α(i), and takes the historical judgment article ranked first as the core article of the historical judicial document, and takes the historical article phrase corresponding to the historical judgment article ranked first as the core article phrase of the historical judicial document.

[0065] Specifically, this embodiment does not impose any specific restrictions on the values ​​of the first preset mapping index A1 and the second preset mapping index A2. Those skilled in the art can set them freely, as long as the value requirements of the first preset mapping index A1 and the second preset mapping index A2 are met. For example, the first preset mapping index A1 can be set to 70% of the average value of each mapping index, and the second preset mapping index A2 can be set to 130% of the average value of each mapping index.

[0066] Please continue reading Figure 2 As shown, the text vector marking module is further provided with a relevance analysis unit, which is connected to the weight analysis unit and analyzes the relevance between the fact words and the legal provisions according to the analysis result of the mapping vector;

[0067] The relevance analysis unit counts the number n of historical fact phrases that appear in the same historical judgment article in each historical judicial document, and calculates the frequency of occurrence v(j,i) of each fact word based on the statistical result of the historical fact phrases, setting v(j,i)=n(j,i) / n, wherein n(j,i) is the number of occurrences of the j-th fact word in the historical judicial document to which the i-th historical judgment article belongs, and v(j,i) is the frequency of occurrence of the j-th fact word in the historical judicial document to which the i-th historical judgment article belongs; the relevance analysis unit determines the relevance of each fact word to each judgment article by counting the historical fact texts judged by the same article, thereby improving the accuracy of the relevance analysis;

[0068] The relevance analysis unit calculates the relevance β(i, j) between the fact word and the legal provision according to the frequency of occurrence of each fact word, and sets β(i, j) = v(j, i) 1 / 2 .

[0069] Please continue reading Figure 1 As shown, the system also includes a core feature extraction module, which is connected to the historical information processing module. The core feature extraction module is used to extract the core features of historical judicial documents based on historical fact phrases and historical legal phrases.

[0070] See also Figure 3 As shown, the core feature extraction module includes an event extraction unit, which extracts event content from historical judicial documents based on historical fact phrases;

[0071] The event extraction unit divides the fact words in the historical fact phrase: if |aj|<B, the event extraction unit divides the fact words into entity words; if |aj|≥B, the event extraction unit divides the fact words into relation words;

[0072] Where aj is the jth fact word, and B is the modulus of the preset partition vector;

[0073] The event extraction unit divides the historical fact phrases into minimum event phrases according to predefined event patterns.

[0074] Specifically, the predefined pattern is a matching pattern set by means of a regular expression, specifically (entity set, relationship word, entity set), and the entity set is a set of multiple entity words; at the same time, this embodiment does not make specific restrictions on the value of the modulus B of the preset partitioning vector. Those skilled in the art can set it freely, and only need to meet the value requirements of the modulus B of the preset partitioning vector. For example, in this embodiment, the modulus B of the preset partitioning vector can be set to 1. It is worth noting that the modulus B of the preset partitioning vector needs to be set according to the pre-trained word vector file used.

[0075] Please continue reading Figure 3 As shown, the core feature extraction module further includes a core feature extraction unit, which is connected to the event extraction unit and is used to extract core features of historical judicial documents based on the analysis results of each minimum event phrase and the historical legal terms phrase;

[0076] The core feature extraction unit iteratively divides each minimum event phrase according to a predefined pattern to obtain each iterative event phrase, and counts the number of occurrences of related words in each iterative event phrase;

[0077] The core feature extraction unit uses the relational word with the largest number of occurrences in each iterative event phrase as the core feature of the historical judicial document.

[0078] Specifically, this embodiment does not perform an iterative division of each minimum event phrase according to a predefined pattern to obtain each iterative event phrase. Instead, the minimum event group is exhaustively enumerated as an entity set, and all pattern matching results are listed as iterative event groups. The format of each iterative event group is the same as that of the minimum event group, i.e., (entity set, relation word, entity set), and the "relation word" is the statistical target.

[0079] Please continue reading Figure 3 As shown, the core feature extraction module is further provided with a case feature analysis unit, which is connected to the core feature extraction unit and is used to analyze each case feature according to the core feature extraction results of each historical judicial document;

[0080] The case feature analysis unit counts the number of occurrences n3(k) of the core features of the same historical case type in each historical judicial document, and analyzes each case feature according to the sorting result: if n3(k) / N1<H, the case feature analysis unit determines that the core feature is not the case feature of the case type; if n3(k) / N1≥H, the case feature analysis unit determines that the core feature is the case feature of the case type; wherein H is a preset number ratio, and N1 is the number of historical judicial documents of the same historical case type in each historical judicial document.

[0081] Specifically, the "number of occurrences of core features of the same historical case type n3(k)" refers to the historical judicial documents with the same historical case type in each historical judicial document. The number of occurrences of the same core features of this type of historical judicial documents is counted, and the statistical result is recorded as n3(k), where n3(k) represents the number of occurrences of the core features of the kth type. At the same time, this embodiment does not make specific restrictions on the value of the preset quantity ratio H. Those skilled in the art can set it freely, as long as the value requirements of the preset quantity ratio H are met. For example, the preset quantity ratio H can be set to 20%.

[0082] Please continue reading Figure 1 As shown, the system also includes a judicial cause of action extraction module, which is connected to the language data acquisition module, the text vector labeling module and the core feature extraction module. The judicial cause of action extraction module is used to sort the target judicial documents by legal article weights based on the correlation analysis results between fact words and legal articles, and to extract judicial causes of action based on the legal article weight sorting results and the feature analysis results of each cause of action.

[0083] See also Figure 4 As shown, the judicial case extraction module includes a target weight analysis unit, which is used to analyze the relevance between fact words and legal provisions and sort the target judicial documents by legal provision weights;

[0084] The target weight analysis unit pre-processes the target judicial document to obtain the target fact phrase;

[0085] The target weight analysis unit calculates the relevance β(p)' between the target fact phrase and each target judgment legal provision based on the relevance analysis result between the fact words and the legal provision, and sets

[0086]

[0087] Among them, β(g, p) is the correlation between the g-th fact word in the target fact phrase and the p-th target judgment law;

[0088] The target weight analysis unit sorts the judgment articles in descending order according to the relevance β(p)' between the fact phrases and the target legal article phrases, and takes the target judgment article ranked first as the target case cause article; the target case cause article is determined by the target weight analysis unit, thereby improving the accuracy of the target case cause article analysis, and thereby improving the accuracy of the case cause extraction.

[0089] Specifically, the target legal article is defined as the legal article that determines the type of the target judicial document's legal article.

[0090] Please continue reading Figure 4As shown, the judicial cause of action extraction module further includes a cause of action matching unit, which is connected to the target weight analysis unit. The cause of action matching unit is used to extract the cause of action type of the target judicial document based on the legal article weight ranking result, the result of the feature analysis of each cause of action, and the corresponding relationship table;

[0091] The case cause matching unit extracts the core features of the target judicial document based on the target fact phrase;

[0092] The cause of action matching unit matches the cause of action type of the target judicial document according to the core features of the target judicial document, the result of the weight ranking of the legal provisions and the corresponding relationship table: if the core features of the target judicial document successfully match the cause of action features and the cause of action types corresponding to the target cause of action legal provisions and the cause of action features are successfully matched in the corresponding relationship table, the cause of action matching unit determines that the match is successful and stores the target judicial document in the successfully matched cause of action type; if the core features of the target judicial document successfully match the cause of action features and the cause of action types corresponding to the target cause of action legal provisions and the cause of action features are unmatched in the corresponding relationship table, the cause of action matching unit determines that the match is unsuccessful and stores the target judicial document in the target cause of action type. The legal article is added to the case type corresponding to the correspondence table; if the core feature of the target judicial document fails to match the case feature and the case type corresponding to the target case legal article and the case feature is successfully matched in the correspondence table, the case matching unit determines that the matching fails; if the core feature of the target judicial document fails to match the case feature and the case type corresponding to the target case legal article and the case feature is failed to match in the correspondence table, the case matching unit determines that the matching fails; the case matching unit preliminarily confirms the case type through the core feature of the target judicial document, and then further determines the case type through the case relationship table and the target case legal article, thereby improving the accuracy of case extraction.

[0093] Specifically, the specific implementation process of "extracting the core features of the target judicial document based on the target fact phrase" in this embodiment is the same as the extraction process of "the core features of each historical judicial document", and this embodiment will not elaborate on it; it can be understood that when the match is successful in this embodiment, the case type of the target judicial document is the case type corresponding to the case feature. The more specific matching logic is: the case type corresponding to the case feature exists, and at the same time, the target case law corresponds to the case type in the corresponding relationship table. At this time, the match is successful.

[0094] It is understandable that if there is more than one case type that is successfully matched, the target judicial text will be stored and marked in all case types, and will be finally confirmed manually in the subsequent final review.

[0095] Thus far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art may make equivalent changes or substitutions to the relevant technical features, and the technical solutions after such changes or substitutions will fall within the scope of protection of the present invention.

Claims

1. A case cause extraction system for judicial documents, characterized in that: include: Historical information entry module, used to enter historical judicial documents; Case statistics module, used to generate corresponding relationship tables based on historical judicial documents; A language data acquisition module is used to obtain a stop word list and text vector file of a judicial document, and is also used to obtain a target judicial document; The historical information processing module is used to pre-process historical judicial documents to obtain historical fact phrases and historical legal clause phrases; A text vector tagging module is used to perform word vector tagging on historical fact phrases and historical legal clause phrases based on the text vector file, and to analyze the relevance between factual words and legal clauses based on the word vector tagging results; The core feature extraction module is used to extract the core features of each historical judicial document based on historical fact phrases and historical legal phrases, and analyze the characteristics of each case based on the extraction results; The judicial case cause extraction module is used to sort the target judicial documents by legal article weights based on the correlation analysis results between fact words and legal articles, and to extract judicial case causes based on the legal article weight sorting results and the feature analysis results of each case cause.

2. The case cause extraction system for judicial documents according to claim 1 is characterized in that: The case cause statistics module collects statistics on the corresponding types of historical case causes and historical judgment laws based on historical judicial documents; The case cause statistics module counts the historical case cause types corresponding to each historical judgment law clause in each historical judicial document, and stores the statistical results.

3. The case extraction system for judicial documents according to claim 2 is characterized in that: The text vector marking module includes a vector marking unit, which performs word vector marking on historical fact phrases and historical legal clause phrases according to the text vector file; The vector marking unit further calculates a mapping vector group c(i) based on the vector group a of the historical fact phrases and the vector group b(i) of the historical legal clause phrases.

4. The case extraction system for judicial documents according to claim 3 is characterized in that: The text vector marking module is further provided with a weight analysis unit, which is used to analyze the mapping weight according to the calculation results of each mapping vector group; The weight analysis unit calculates the mapping index α(i) between the vector group a of the historical fact phrases and each historical legal clause phrase; The weight analysis unit compares the mapping index α(i) with each preset mapping index, and analyzes the mapping weight according to the comparison result. The analysis results of the mapping weight include: low, normal and high.

5. The case extraction system for judicial documents according to claim 4 is characterized in that: The relevance analysis unit analyzes the relevance between the fact words and the legal provisions according to the analysis results of the mapping vectors; the relevance analysis unit counts the number n of historical fact phrases that appear in the same historical judgment legal provision in each historical judicial document, and calculates the occurrence frequency v(j,i) of each fact word based on the statistical results of the historical fact phrases; The relevance analysis unit calculates the relevance β(i, j) between the fact words and the legal provisions according to the occurrence frequency of each fact word.

6. The case cause extraction system for judicial documents according to claim 5, characterized in that: The core feature extraction module includes an event extraction unit, which divides fact words in historical fact phrases, and the division results include entity words and relationship words; The event extraction unit divides the historical fact phrases into minimum event phrases according to predefined event patterns.

7. The case extraction system for judicial documents according to claim 6 is characterized in that: The core feature extraction module further includes a core feature extraction unit, which is used to extract core features of historical judicial documents based on the analysis results of each minimum event phrase and the historical legal terms phrase; The core feature extraction unit iteratively divides each minimum event phrase according to a predefined pattern to obtain each iterative event phrase, and counts the number of occurrences of related words in each iterative event phrase; The core feature extraction unit uses the relational word with the largest number of occurrences in each iterative event phrase as the core feature of the historical judicial document.

8. The case cause extraction system for judicial documents according to claim 7, characterized in that: The core feature extraction module is further provided with a case feature analysis unit, which is used to analyze the features of each case based on the core feature extraction results of each historical judicial document; The case feature analysis unit counts the number of occurrences n3(k) of the core features of the same historical case type in each historical judicial document, and analyzes each case feature based on the statistical results to determine whether the core feature is a case feature of the case type.

9. The case cause extraction system for judicial documents according to claim 8, characterized in that: The target weight analysis unit is used to analyze the relevance between fact words and legal provisions and sort the target judicial documents by legal provision weights; The target weight analysis unit pre-processes the target judicial document to obtain the target fact phrase; The target weight analysis unit calculates the relevance β(p)' between the target fact phrase and each target judgment legal provision according to the relevance analysis result between the fact words and the legal provision; The target weight analysis unit sorts each judgment law in descending order according to the relevance β(p)' between the fact phrase and each target law phrase, and takes the first ranked target judgment law as the target case law.

10. The case cause extraction system for judicial documents according to claim 9, characterized in that: The judicial cause of action extraction module also includes a cause of action matching unit, which is used to extract the cause of action type of the target judicial document based on the result of the legal article weight ranking, the result of the feature analysis of each cause of action, and the corresponding relationship table; The case cause matching unit extracts the core features of the target judicial document based on the target fact phrase; The case cause matching unit matches the case cause type of the target judicial document according to the core features of the target judicial document, the result of the weight ranking of the legal provisions and the correspondence table: if the core features of the target judicial document successfully match the case cause features and the target case cause legal provision and the case cause type corresponding to the case cause features successfully match in the correspondence table, the case cause matching unit determines that the match is successful and stores the target judicial document in the case cause type that matches successfully; if the core features of the target judicial document successfully match the case cause features and the target case cause legal provision and the case cause type corresponding to the case cause features fail to match in the correspondence table, the case cause matching unit determines that the match fails and adds the target case cause legal provision of the target judicial document to the case cause type corresponding to the correspondence table; if the core features of the target judicial document fail to match the case cause features and the target case cause legal provision and the case cause type corresponding to the case cause features successfully match in the correspondence table, the case cause matching unit determines that the match fails; if the core features of the target judicial document fail to match the case cause features and the target case cause legal provision and the case cause type corresponding to the case cause features fail to match in the correspondence table, the case cause matching unit determines that the match fails.

Images