Clue extraction method and system for public interest litigation cases based on multi-feature model

Abstract

The present invention provides a method and system for extracting clues of public interest litigation cases based on a multi-feature model, comprising: step S1: preprocessing the source text of the public interest litigation case; step S2: using the preprocessed source text of the public interest litigation case as the input object of the clue element extraction module; step S3: using the clue element extraction module to process the input source text of the public interest litigation case, and outputting a case clue element annotation sequence; step S4: converting and integrating the case clue element annotation sequence to form a case clue element set corresponding to the source text of the public interest litigation case, and completing the extraction of clue elements; step S5: organizing and integrating the extracted clue elements to obtain clues corresponding to the entire source text of the public interest litigation case. The present invention can improve the efficiency of extracting clue elements of the source text of the public interest litigation case; and solve the problems of low efficiency in finding clues of public interest litigation cases and a relatively single source of clues in the prior art.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method and system for extracting clues from public interest litigation cases based on a multi-feature model. Background Art

[0002] Public interest litigation is a litigation specifically aimed at safeguarding national interests and social public interests, rather than a form of litigation. Specifically, as long as the illegal behavior of the actor infringes upon national interests and social public interests, and infringes upon or has potential damage to the legitimate rights and interests of the state or unspecified persons, any organization or individual has the right to file a lawsuit with the court, requesting the court of jurisdiction to try and impose sanctions in accordance with the law, stop and correct the illegal behavior, and maximize the protection of national interests and social public interests.

[0003] At present, the procuratorial organs discover case clues mainly through manual methods such as accusations and reports. The sources of clues are relatively single and the case clues are limited. Therefore, it is necessary to broaden the source of clues, actively build external cooperation mechanisms, and pay attention to new media such as online forums. Compared with traditional channels, these new media channels have the characteristics of large text volume and low density of case clues. Therefore, making full use of new media channels requires high clue screening capabilities.

[0004] In the existing technology, clues for public interest litigation cases mainly come from manual methods such as accusations and reports. The sources of clues are relatively single and the efficiency of discovering case clues is limited.

[0005] At present, the procuratorial organs discover case clues mainly through manual methods such as accusations and reports. The sources of clues are relatively single and the case clues are limited. Therefore, it is necessary to broaden the source of clues, actively build external cooperation mechanisms, and pay attention to new media such as online forums. Compared with traditional channels, these new media channels have the characteristics of large text volume and low density of case clues. Therefore, making full use of new media channels requires high clue screening capabilities.

[0006] The Chinese patent document CN112711650A (application number: CN201911018650.8) discloses "a method and system for mining clues for public interest litigation", which belongs to the field of data mining technology, including obtaining text data to be mined, calculating the quantitative proportion of the combined words related to the public interest litigation feature word library in the combined words of the text data to be mined, and the correlation mean of the combined words related to the feature words; calculating the sentiment score of the text data to be mined; calculating the text similarity between the text data to be mined and the cases in the public interest litigation case library; calculating the topic similarity between the text data to be mined and the cases in the public interest litigation case library; based on the quantitative proportion, correlation mean, sentiment score, text similarity and topic similarity, obtaining effective public interest litigation clues from the text data to be mined. Although this method of mining clues for public interest litigation can mine clues for public interest litigation from Internet data, it is mainly based on statistical features such as quantitative proportion, correlation mean, text similarity and topic similarity, and the feature extraction granularity is coarse and the accuracy is low. It fails to be used in combination with the more accurate, better and more novel BERT model.

[0007] Patent document CN112711693A (application number: CN201911017853.5) discloses a litigation clue mining method and system based on multi-feature fusion, which belongs to the field of data mining technology, including: obtaining text data to be mined, the storage structure of the text data includes title and content; performing sentiment analysis on the content of the text data, and judging whether the negative score of the sentiment analysis is greater than the set first confidence threshold, if it is, entity recognition is performed, otherwise the data is discarded; performing named entity recognition on the content of the text data, and judging whether the recognition result contains valid entities, if it is similarity calculation, otherwise the data is discarded; performing semantic similarity calculation on the title of the text data and the title of the reference text, and judging whether the semantic similarity is greater than the set second confidence threshold, if it is determined to be a valid public interest litigation clue, otherwise the data is discarded. This invention fails to be used in conjunction with the BERT model, which is more accurate, more effective, and more novel. Summary of the invention

[0008] In view of the defects in the prior art, the purpose of the present invention is to provide a clue extraction method and system for public interest litigation cases based on a multi-feature model.

[0009] A method for extracting clues from public interest litigation cases based on a multi-feature model provided by the present invention includes:

[0010] Step S1: pre-processing the source text of the public interest litigation case;

[0011] Step S2: using the pre-processed public interest litigation case source text as the input object of the clue element extraction module;

[0012] Step S3: Processing the inputted public interest litigation case source text using the clue element extraction module, and outputting a case clue element annotation sequence;

[0013] Step S4: converting and integrating the case clue element annotation sequence to form a case clue element set corresponding to the public interest litigation case source text, and completing the extraction of clue elements;

[0014] Step S5: Organize and integrate the extracted clue elements to obtain clues corresponding to the entire public interest litigation case source text.

[0015] Preferably, the four elements of illegal subject element, illegal behavior element, time element and location element in the source text of the public interest litigation are extracted;

[0016] The illegal subject element indicates which person or company should bear the illegal liability in the public interest litigation case corresponding to the public interest litigation case source text;

[0017] The illegal behavior element indicates what kind of illegal behavior the illegal subject has committed or what kind of consequences it has caused;

[0018] The temporal element indicates when the alleged violation occurred;

[0019] The location element indicates where the violation occurred.

[0020] Preferably, the illegal subject element, the time element and the location element adopt the BI marking method, and the first character in the element is marked with the label B-category, and the subsequent characters are marked with I-category;

[0021] The illegal behavior elements are marked with tail punctuation, and the punctuation after the element sentence is marked with E-category;

[0022] Mark the remaining non-element characters as O.

[0023] Preferably, in step S2:

[0024] The clue element extraction module includes: a Bert network module, a BiLSTM+Transformer+CNN network module and a CRF network module;

[0025] The Bert network module, the BiLSTM+Transformer+CNN network module and the CRF network module are different components of the clue element extraction module, which are in a serial logical relationship;

[0026] When the clue element extraction model extracts case clue elements, it converts this task into a sequence labeling task, so that various clue elements can be labeled simultaneously in a multi-task joint manner; the BiLSTM+Transformer+CNN network adds a self-attention mechanism; and the Chinese Bert model is expanded and pre-trained using public interest litigation case source texts.

[0027] Preferably, the Bert network module includes:

[0028] Bert Chinese pre-training model bert_base_chinese;

[0029] Bert Chinese pre-training model bert_base_chinese includes:

[0030] Use public interest litigation case source text to expand and pre-train bert_base_chinese;

[0031] The BiLSTM+Transformer+CNN network module includes:

[0032] The self-attention mechanism is adopted and the training adopts the sequence labeling task.

[0033] According to the present invention, a clue extraction system for public interest litigation cases based on a multi-feature model is provided, comprising:

[0034] Module M1: Preprocessing of source texts of public interest litigation cases;

[0035] Module M2: using the pre-processed public interest litigation case source text as the input object of the clue element extraction module;

[0036] Module M3: Processing the inputted public interest litigation case source text using the clue element extraction module, and outputting a case clue element annotation sequence;

[0037] Module M4: Convert and integrate the case clue element annotation sequence to form a case clue element set corresponding to the public interest litigation case source text, and complete the extraction of clue elements;

[0038] Module M5: Organize and integrate the extracted clue elements to obtain clues corresponding to the entire public interest litigation case source text.

[0039] Preferably, the four elements of illegal subject element, illegal behavior element, time element and location element in the source text of the public interest litigation are extracted;

[0040] The illegal subject element indicates which person or company should bear the illegal liability in the public interest litigation case corresponding to the public interest litigation case source text;

[0041] The illegal behavior element indicates what kind of illegal behavior the illegal subject has committed or what kind of consequences it has caused;

[0042] The temporal element indicates when the alleged violation occurred;

[0043] The location element indicates where the violation occurred.

[0044] Preferably, the illegal subject element, the time element and the location element adopt the BI marking method, and the first character in the element is marked with the label B-category, and the subsequent characters are marked with I-category;

[0045] The illegal behavior elements are marked with tail punctuation, and the punctuation after the element sentence is marked with E-category;

[0046] Mark the remaining non-element characters as O.

[0047] Preferably, in the module M2:

[0048] The clue element extraction module includes: a Bert network module, a BiLSTM+Transformer+CNN network module and a CRF network module;

[0049] The Bert network module, the BiLSTM+Transformer+CNN network module and the CRF network module are different components of the clue element extraction module, which are in a serial logical relationship;

[0050] When the clue element extraction model extracts case clue elements, it converts this task into a sequence labeling task, so that various clue elements can be labeled simultaneously in a multi-task joint manner; the BiLSTM+Transformer+CNN network adds a self-attention mechanism; and the Chinese Bert model is expanded and pre-trained using public interest litigation case source texts.

[0051] Preferably, the Bert network module includes:

[0052] Bert Chinese pre-training model bert_base_chinese;

[0053] Bert Chinese pre-training model bert_base_chinese includes:

[0054] Use public interest litigation case source text to expand and pre-train bert_base_chinese;

[0055] The BiLSTM+Transformer+CNN network module includes:

[0056] The self-attention mechanism is adopted and the training adopts the sequence labeling task.

[0057] Compared with the prior art, the present invention has the following beneficial effects:

[0058] 1. The present invention extracts clue elements contained in the source text of the public interest litigation case through a clue element extraction module, organizes and integrates the extracted clue elements, and obtains clues corresponding to the entire source text of the public interest litigation case; the mutual correlation between the case clue elements reflected by the semantic information contained in the source text is utilized, so the clue element extraction module can simultaneously mark various clue elements in a multi-task joint manner, thereby improving the efficiency of clue element extraction of the source text of the public interest litigation case;

[0059] 2. The present invention provides a method for extracting clue elements of public interest litigation cases with high accuracy;

[0060] 3. The present invention solves the problems of low efficiency in finding clues for public interest litigation cases and relatively single clue sources in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0062] Figure 1 This is a schematic diagram of the implementation process of the method for extracting clue elements for public interest litigation;

[0063] Figure 2 This is a schematic diagram of the clue element extraction module;

[0064] Figure 3 A sequence diagram to mark clues in the case. DETAILED DESCRIPTION

[0065] The present invention is described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, for those of ordinary skill in the art, several changes and improvements can also be made without departing from the concept of the present invention. These all belong to the protection scope of the present invention.

[0066] Embodiment 1:

[0067] According to a clue extraction method for public interest litigation cases based on a multi-feature model provided by the present invention, Figure 1-Figure 3 As shown, including:

[0068] Step S1: pre-processing the source text of the public interest litigation case;

[0069] Step S2: using the pre-processed public interest litigation case source text as the input object of the clue element extraction module;

[0070] Step S3: Processing the inputted public interest litigation case source text using the clue element extraction module, and outputting a case clue element annotation sequence;

[0071] Step S4: converting and integrating the case clue element annotation sequence to form a case clue element set corresponding to the public interest litigation case source text, and completing the extraction of clue elements;

[0072] Step S5: Organize and integrate the extracted clue elements to obtain clues corresponding to the entire public interest litigation case source text.

[0073] Specifically, the four elements of illegal subject element, illegal behavior element, time element and location element in the source text of public interest litigation are extracted;

[0074] The illegal subject element indicates which person or company should bear the illegal liability in the public interest litigation case corresponding to the public interest litigation case source text;

[0075] The illegal behavior element indicates what kind of illegal behavior the illegal subject has committed or what kind of consequences it has caused;

[0076] The temporal element indicates when the alleged violation occurred;

[0077] The location element indicates where the violation occurred.

[0078] Specifically, the illegal subject element, time element, and location element adopt the BI annotation method. The first character in the element is annotated with the label B-category, and the subsequent characters are annotated with I-category;

[0079] The illegal behavior elements are marked with tail punctuation, and the punctuation after the element sentence is marked with E-category;

[0080] Mark the remaining non-element characters as O.

[0081] Specifically, in step S2:

[0082] The clue element extraction module includes: a Bert network module, a BiLSTM+Transformer+CNN network module and a CRF network module;

[0083] The Bert network module, the BiLSTM+Transformer+CNN network module and the CRF network module are different components of the clue element extraction module, which are in a serial logical relationship;

[0084] When the clue element extraction model extracts case clue elements, it converts this task into a sequence labeling task, so that various clue elements can be labeled simultaneously in a multi-task joint manner; the BiLSTM+Transformer+CNN network adds a self-attention mechanism; and the Chinese Bert model is expanded and pre-trained using public interest litigation case source texts.

[0085] Specifically, the Bert network module includes:

[0086] Bert Chinese pre-training model bert_base_chinese;

[0087] Bert Chinese pre-training model bert_base_chinese includes:

[0088] Use public interest litigation case source text to expand and pre-train bert_base_chinese;

[0089] The BiLSTM+Transformer+CNN network module includes:

[0090] The self-attention mechanism is adopted and the training adopts the sequence labeling task.

[0091] Embodiment 2:

[0092] Embodiment 2 is a preferred example of Embodiment 1, and is used to illustrate the present invention in more detail.

[0093] Those skilled in the art may understand the clue extraction method for public interest litigation cases based on a multi-feature model provided by the present invention as a specific implementation method of the clue extraction system for public interest litigation cases based on a multi-feature model, that is, the clue extraction system for public interest litigation cases based on a multi-feature model may be realized by executing the step flow of the clue extraction method for public interest litigation cases based on a multi-feature model.

[0094] According to the present invention, a clue extraction system for public interest litigation cases based on a multi-feature model is provided, comprising:

[0095] Module M1: Preprocessing of source texts of public interest litigation cases;

[0096] Module M2: using the pre-processed public interest litigation case source text as the input object of the clue element extraction module;

[0097] Module M3: Processing the inputted public interest litigation case source text using the clue element extraction module, and outputting a case clue element annotation sequence;

[0098] Module M4: Convert and integrate the case clue element annotation sequence to form a case clue element set corresponding to the public interest litigation case source text, and complete the extraction of clue elements;

[0099] Module M5: Organize and integrate the extracted clue elements to obtain clues corresponding to the entire public interest litigation case source text.

[0100] Specifically, the four elements of illegal subject element, illegal behavior element, time element and location element in the source text of public interest litigation are extracted;

[0101] The illegal subject element indicates which person or company should bear the illegal liability in the public interest litigation case corresponding to the public interest litigation case source text;

[0102] The illegal behavior element indicates what kind of illegal behavior the illegal subject has committed or what kind of consequences it has caused;

[0103] The temporal element indicates when the alleged violation occurred;

[0104] The location element indicates where the violation occurred.

[0105] Specifically, the illegal subject element, time element, and location element adopt the BI annotation method. The first character in the element is annotated with the label B-category, and the subsequent characters are annotated with I-category;

[0106] The illegal behavior elements are marked with tail punctuation, and the punctuation after the element sentence is marked with E-category;

[0107] Mark the remaining non-element characters as O.

[0108] Specifically, in the module M2:

[0109] The clue element extraction module includes: a Bert network module, a BiLSTM+Transformer+CNN network module and a CRF network module;

[0110] The Bert network module, the BiLSTM+Transformer+CNN network module and the CRF network module are different components of the clue element extraction module, which are in a serial logical relationship;

[0111] When the clue element extraction model extracts case clue elements, it converts this task into a sequence labeling task, so that various clue elements can be labeled simultaneously in a multi-task joint manner; the BiLSTM+Transformer+CNN network adds a self-attention mechanism; and the Chinese Bert model is expanded and pre-trained using public interest litigation case source texts.

[0112] Specifically, the Bert network module includes:

[0113] Bert Chinese pre-training model bert_base_chinese;

[0114] Bert Chinese pre-training model bert_base_chinese includes:

[0115] Use public interest litigation case source text to expand and pre-train bert_base_chinese;

[0116] The BiLSTM+Transformer+CNN network module includes:

[0117] The self-attention mechanism is adopted and the training adopts the sequence labeling task.

[0118] Embodiment 3:

[0119] Embodiment 3 is a preferred example of Embodiment 1, and is used to illustrate the present invention in more detail.

[0120] 1. Refine the invention point

[0121] Based on the BERT model and BiLSTM+Transformer+CNN network, the four elements of "illegal subject elements, illegal behavior elements, time elements and location elements" in the source text of public interest litigation are extracted.

[0122] Specifically, when the clue element extraction model extracts case clue elements, it converts this task into a sequence labeling task, so that various types of clue elements can be labeled simultaneously in a multi-task joint manner; the BiLSTM+Transformer+CNN network adds a self-attention mechanism; and the Chinese Bert model is expanded and pre-trained using public interest litigation case source texts.

[0123] A method for extracting clue elements of public interest litigation cases, comprising:

[0124] Extract clue elements contained in the source text of public interest litigation through the clue element extraction module;

[0125] The extracted clue elements are organized and integrated to obtain clues corresponding to the entire public interest litigation case source text.

[0126] Specifically, pre-process the source text of public interest litigation cases;

[0127] Using the pre-processed public interest litigation case source text as the input object of the clue element extraction module;

[0128] The inputted public interest litigation case source text is processed using the clue element extraction module, and a case clue element annotation sequence is output;

[0129] The case clue element annotation sequence is converted and integrated to form a case clue element set corresponding to the public interest litigation case source text, thereby completing the extraction of clue elements.

[0130] Specifically, the clue element extraction module includes: Bert network module, BiLSTM+Transformer+CNN network module, CRF network module;

[0131] The Bert network module, BiLSTM+Transformer+CNN network module, and CRF network module are different components of the clue element extraction module and are in a serial logical relationship.

[0132] Specifically, the Bert Chinese pre-training model bert_base_chinese;

[0133] The Bert Chinese pre-training model bert_base_chinese includes:

[0134] Use public interest litigation case source text to expand and pre-train bert_base_chinese.

[0135] Specifically, (1) adopt the self-attention mechanism

[0136] (2) Training using sequence tagging tasks

[0137] Specifically, the illegal subject element, time element and location element adopt the BI annotation method, and the first character in the element is annotated with the label "B-category", and the subsequent characters are annotated with "I-category";

[0138] The illegal behavior elements are marked with tail punctuation, and the punctuation after the element sentence is marked with "E-category";

[0139] The remaining non-element characters are marked as "O".

[0140] Embodiment 4:

[0141] Embodiment 4 is a preferred example of Embodiment 1, and is used to illustrate the present invention in more detail.

[0142] The technical problem to be solved by the present invention is to provide a method for extracting clues for public interest litigation cases, so as to solve the problems of low efficiency in finding clues for public interest litigation cases and relatively single clue sources in the prior art.

[0143] In order to solve the above technical problems, an embodiment of the present invention provides a method for extracting clues of public interest litigation cases, including:

[0144] Extract clue elements contained in the source text of public interest litigation through the clue element extraction module;

[0145] The extracted clue elements are organized and integrated to obtain clues corresponding to the entire public interest litigation case source text.

[0146] Furthermore, the extracting of clue elements contained in the source text of the public interest litigation case by the clue element extraction module includes:

[0147] Pre-process the source text of public interest litigation;

[0148] Using the pre-processed public interest litigation case source text as the input object of the clue element extraction module;

[0149] The inputted public interest litigation case source text is processed using the clue element extraction module, and a case clue element annotation sequence is output;

[0150] The case clue element annotation sequence is converted and integrated to form a case clue element set corresponding to the public interest litigation case source text, thereby completing the extraction of clue elements.

[0151] Furthermore, the clue element extraction module includes: Bert network module, BiLSTM+Transformer+CNN network module, CRF network module;

[0152] The Bert network module, BiLSTM+Transformer+CNN network module, and CRF network module are different components of the clue element extraction module and are in a serial logical relationship.

[0153] Furthermore, the case clue element set includes: illegal subject element, illegal behavior element, time element and location element;

[0154] The illegal subject element indicates which person or company should bear the illegal liability in the public interest litigation case corresponding to the public interest litigation case source text;

[0155] The illegal behavior element indicates what kind of illegal behavior the illegal subject has committed or what kind of consequences it has caused;

[0156] The temporal element indicates when the alleged violation occurred;

[0157] The location element indicates where the violation occurred.

[0158] Embodiment 5:

[0159] Embodiment 5 is a preferred example of Embodiment 1, and is used to illustrate the present invention in more detail.

[0160] like Figure 1 As shown, the method for extracting clue elements of public interest litigation provided by the embodiment of the present invention includes:

[0161] Step 1: Extract clue elements contained in the source text of the public interest litigation case through the clue element extraction module;

[0162] In practical applications, there are four types of clue elements that need to be extracted:

[0163] 1. Illegal subject element: Which person or company should bear the illegal responsibility for the public interest litigation case corresponding to the public interest litigation case source text?

[0164] 2. Illegal behavior elements: What illegal behavior did the illegal subject commit or what consequences did it cause?

[0165] 3. Time element: when did the illegal act occur? 4. Location element: where did the illegal act occur?

[0166] In actual applications, the source text of a public interest litigation case may not contain all four clue elements mentioned above, and may not contain any case clues. Therefore, the embodiment of the present invention regards the illegal behavior element as the primary key. If the public interest litigation clue element extraction method does not extract the illegal behavior element, it is considered that the source text of the public interest litigation case does not contain any case clues.

[0167] For example, for the case source text "On the evening of April 27, 2019, Liu Xlun, the husband of Zhang Xrong, the former women's director of Sanjiang Village, Taihe Town, Dongpo District, privately occupied public resources next to the fish pond he contracted, dug a latrine and piled chicken manure; his behavior seriously damaged the collective interests, endangered the safety of pedestrians, and seriously damaged the social environment, which is incompatible with the current vigorous protection of the environment. This person is extremely arrogant and unreasonable, and the village leaders can't control him. Please ask the superior leaders to take care of it.", output:

[0168] 1. Elements of illegal subject: Liu Xulun

[0169] 2. Illegal behavior elements: Liu Xlun, the husband of Zhang Xrong, the former women's director of Sanjiang Village, Taihe Town, Dongpo District, privately occupied public resources next to the fish pond he contracted, dug a pit and piled chicken manure

[0170] 3. Time element: April 27, 2019

[0171] 4. Location elements: Sanjiang Village, Taihe Town, Dongpo District

[0172] Furthermore, the extracting of clue elements contained in the source text of the public interest litigation case by the clue element extraction module includes:

[0173] 1. Pre-process the source text of public interest litigation;

[0174] 2. Using the pre-processed public interest litigation case source text as the input object of the clue element extraction module;

[0175] 3. The input public interest litigation case source text is processed using the clue element extraction module, and a case clue element annotation sequence is output.

[0176] Among them, the public interest litigation case source text is preprocessed, including punctuation normalization. Specifically, the punctuation marks that are repeated continuously in the text are removed, such as changing "..." to "." and changing "..." to "...".

[0177] Step 2: Organize and integrate the extracted clue elements to obtain clues corresponding to the entire public interest litigation case source text;

[0178] The clue elements are organized and integrated, including:

[0179] Remove duplicates from the extracted clue elements.

[0180] like Figure 2 As shown, the clue element extraction module in this embodiment includes: a Bert network module, a BiLSTM+Transformer+CNN network module, and a CRF network module;

[0181] The Bert network module includes:

[0182] Bert Chinese pre-training model bert_base_chinese;

[0183] The Bert Chinese pre-training model bert_base_chinese includes:

[0184] Use public interest litigation case source text to expand and pre-train bert_base_chinese.

[0185] In this embodiment, the BiLSTM+Transformer+CNN network module includes:

[0186] 1. Use self-attention mechanism

[0187] 2. Training uses sequence tagging tasks

[0188] like Figure 3As shown, the key to the case clue element annotation sequence in this embodiment is:

[0189] 1. The illegal subject element, time element and location element adopt the BI marking method. The first character in the element is marked with the label "B-category", and the subsequent characters are marked with "I-category";

[0190] 2. The illegal behavior elements are marked with the tail punctuation method, and the punctuation after the element sentence is marked with "E-category";

[0191] 3. Mark the remaining non-element characters as "O".

[0192] Those skilled in the art know that, in addition to implementing the system, device and its various modules provided by the present invention in a purely computer-readable program code, it is entirely possible to implement the same program in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers and embedded microcontrollers by logically programming the method steps. Therefore, the system, device and its various modules provided by the present invention can be considered as a hardware component, and the modules included therein for implementing various programs can also be considered as structures within the hardware component; the modules for implementing various functions can also be considered as both software programs for implementing the method and structures within the hardware component.

[0193] The above describes the specific embodiments of the present invention. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essence of the present invention. In the absence of conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.

Claims

1. A clue extraction method for public interest litigation cases based on a multi-feature model, characterized in that: include: Step S1: pre-processing the source text of the public interest litigation case; Step S2: using the pre-processed public interest litigation case source text as the input object of the clue element extraction module; Step S3: Processing the inputted public interest litigation case source text using the clue element extraction module, and outputting a case clue element annotation sequence; Step S4: converting and integrating the case clue element annotation sequence to form a case clue element set corresponding to the public interest litigation case source text, and completing the extraction of clue elements; Step S5: Organize and integrate the extracted clue elements to obtain clues corresponding to the entire public interest litigation case source text; In step S2: The clue element extraction module includes: a Bert network module, a BiLSTM+Transformer+CNN network module and a CRF network module; The Bert network module, the BiLSTM+Transformer+CNN network module and the CRF network module are different components of the clue element extraction module, which are in a serial logical relationship; When the clue element extraction model extracts case clue elements, it converts this task into a sequence labeling task, so that various clue elements can be labeled simultaneously in a multi-task joint manner; the BiLSTM+Transformer+CNN network adds a self-attention mechanism; and the Chinese Bert model is expanded and pre-trained using public interest litigation case source texts.

2. The method for extracting clues from public interest litigation cases based on a multi-feature model according to claim 1 is characterized in that: Extract the four elements of illegal subject element, illegal behavior element, time element and location element in the source text of public interest litigation; The illegal subject element indicates which person or company should bear the illegal liability in the public interest litigation case corresponding to the public interest litigation case source text; The illegal behavior element indicates what kind of illegal behavior the illegal subject has committed or what kind of consequences it has caused; The temporal element indicates when the alleged violation occurred; The location element indicates where the violation occurred.

3. The method for extracting clues from public interest litigation cases based on a multi-feature model according to claim 1 is characterized in that: The illegal subject element, time element and location element are annotated using the BI method. The first character in the element is annotated with the label B-category, and the subsequent characters are annotated with I-category. The illegal behavior elements are marked with tail punctuation, and the punctuation after the element sentence is marked with E-category; Mark the remaining non-element characters as O.

4. The method for extracting clues from public interest litigation cases based on a multi-feature model according to claim 1 is characterized in that: Bert network modules include: Bert Chinese pre-training model bert_base_chinese; Bert Chinese pre-training model bert_base_chinese includes: Use public interest litigation case source text to expand and pre-train bert_base_chinese; The BiLSTM+Transformer+CNN network module includes: The self-attention mechanism is adopted and the training adopts the sequence labeling task.

5. A clue extraction system for public interest litigation cases based on a multi-feature model, characterized in that: include: Module M1: Preprocessing of source texts of public interest litigation cases; Module M2: using the pre-processed public interest litigation case source text as the input object of the clue element extraction module; Module M3: Processing the inputted public interest litigation case source text using the clue element extraction module, and outputting a case clue element annotation sequence; Module M4: Convert and integrate the case clue element annotation sequence to form a case clue element set corresponding to the public interest litigation case source text, and complete the extraction of clue elements; Module M5: Organize and integrate the extracted clue elements to obtain clues corresponding to the entire public interest litigation case source text; In the module M2: The clue element extraction module includes: a Bert network module, a BiLSTM+Transformer+CNN network module and a CRF network module; The Bert network module, the BiLSTM+Transformer+CNN network module and the CRF network module are different components of the clue element extraction module, which are in a serial logical relationship; When the clue element extraction model extracts case clue elements, it converts this task into a sequence labeling task, so that various clue elements can be labeled simultaneously in a multi-task joint manner; the BiLSTM+Transformer+CNN network adds a self-attention mechanism; and the Chinese Bert model is expanded and pre-trained using public interest litigation case source texts.

6. The clue extraction system for public interest litigation cases based on a multi-feature model according to claim 5 is characterized by: Extract the four elements of illegal subject element, illegal behavior element, time element and location element in the source text of public interest litigation; The illegal subject element indicates which person or company should bear the illegal liability in the public interest litigation case corresponding to the public interest litigation case source text; The illegal behavior element indicates what kind of illegal behavior the illegal subject has committed or what kind of consequences it has caused; The temporal element indicates when the alleged violation occurred; The location element indicates where the violation occurred.

7. The clue extraction system for public interest litigation cases based on a multi-feature model according to claim 5 is characterized by: The illegal subject element, time element and location element are annotated using the BI method. The first character in the element is annotated with the label B-category, and the subsequent characters are annotated with I-category. The illegal behavior elements are marked with tail punctuation, and the punctuation after the element sentence is marked with E-category; Mark the remaining non-element characters as O.

8. The clue extraction system for public interest litigation cases based on a multi-feature model according to claim 5 is characterized by: Bert network modules include: Bert Chinese pre-training model bert_base_chinese; Bert Chinese pre-training model bert_base_chinese includes: Use public interest litigation case source text to expand and pre-train bert_base_chinese; The BiLSTM+Transformer+CNN network module includes: The self-attention mechanism is adopted and the training adopts the sequence labeling task.

Images