Intelligent identification method of legal document elements based on big data

Abstract

The present invention provides a method for intelligently identifying elements of legal documents based on big data, including: determining the attribute of a first legal document received by a server to obtain a first attribute, selecting a set of first document elements corresponding to the first attribute in a database; determining a first identification feature corresponding to the first document element set, identifying the first legal document based on the first identification feature, and obtaining a first element content corresponding to the first identification feature; determining the first identification feature of the first element content that is not obtained as a second identification feature, and determining the first identification feature of the first element content that does not meet the requirements as a third identification feature; processing the standard element table based on the second and third identification features to obtain a feedback element table and feed it back to a target end, receiving feedback information from the target end based on the feedback element table and filling it into the standard element table, obtaining a total standard element table and sending it to a management end.

Description

Technical Field

[0001] The present invention relates to data processing technology, and in particular to a method for intelligently identifying elements of legal documents based on big data. Background Art

[0002] As a concise and clearly structured document, the element-based judgment document aims to improve judicial efficiency and transparency through standardized and elemental content presentation. This document contains key elements such as basic case information, dispute focus, fact finding, legal basis, judgment result and reasons, reducing lengthy descriptions and making it easier for all parties to understand.

[0003] In the prior art, the generation of element-based judgment documents mostly relies on manual operations. Judicial personnel need to manually extract important elements from various legal documents and fill them into standard formats. This method is time-consuming, labor-intensive and error-prone, and cannot meet the needs of efficiently processing a large number of legal documents. At the same time, due to the complexity and diversity of the types and contents of legal documents, the prior art cannot flexibly cope with the automated processing of different types of documents, which further limits the automatic generation of element-based judgment documents.

[0004] Therefore, how to automatically extract and fill in important elements from legal documents and assist users in improving the efficiency and accuracy of generating element-based judicial documents has become an urgent problem that needs to be solved. Summary of the invention

[0005] The embodiment of the present invention provides a method for intelligently identifying elements of legal documents based on big data, which can automatically extract and fill in important element content from legal documents, and assist users in improving the efficiency and accuracy of generating element-based judicial documents.

[0006] A first aspect of an embodiment of the present invention provides a method for intelligently identifying elements of legal documents based on big data, comprising:

[0007] The server determines the attribute of the first legal document received to obtain a first attribute, and selects a first document element set corresponding to the first attribute in the database;

[0008] Determine a first identification feature corresponding to the first document element set, identify the first legal document based on the first identification feature, and obtain a first element content corresponding to the first identification feature;

[0009] Retrieving a standard element table corresponding to the first attribute, filling the first element content into the standard element table, determining the first identification feature of the first element content that is not obtained as the second identification feature, and determining the first identification feature of the first element content that does not meet the requirement as the third identification feature;

[0010] Based on the second identification feature and the third identification feature, the standard element table is processed to obtain a feedback element table which is fed back to the target end. Based on the feedback element table, feedback information from the target end is received and filled into the standard element table to obtain a total standard element table which is sent to the management end.

[0011] Optionally, in a possible implementation manner of the first aspect, the server determines the attribute of the received first legal document to obtain a first attribute, and selects a first document element set corresponding to the first attribute in the database, including:

[0012] The server identifies attribute keywords of the received first legal document, wherein the attribute keywords include the cause of action, and locates the line corresponding to the first legal document to obtain the first line location;

[0013] Eliminate the text and punctuation text corresponding to the attribute keyword in the first line location, and recognize the remaining text in the first line location to obtain a first recognized text;

[0014] If the remaining text in the first row position is empty, determining the corresponding second row position based on the first row position, and recognizing the remaining text in the second row position to obtain a first recognized text;

[0015] The first recognized text is segmented to determine the corresponding preset standard text, and the attribute corresponding to the preset standard text is obtained to obtain a first attribute, and a first document element set corresponding to the first attribute in the database is selected, and each first attribute has a preset first document element set.

[0016] Optionally, in a possible implementation of the first aspect, the word segmentation processing of the first recognized text determines a corresponding preset standard text, and an attribute corresponding to the preset standard text obtains a first attribute, and a first document element set corresponding to the first attribute in a database is selected, and each first attribute has a preset first document element set, including:

[0017] Recognize and remove the preset removal words in the first recognition text, and perform word segmentation on the remaining first recognition text to obtain a plurality of first recognition word segments;

[0018] Compare the first recognition segmented word with the preset standard segmented word of each preset standard text, and if there is a preset standard segmented word of the preset standard text that is exactly the same as the first recognition segmented word, use the corresponding preset standard text as the determined preset standard text;

[0019] If it is determined that there is no situation where the preset standard word segmentation of the preset standard text is completely identical to the first recognition word segmentation, the recognition similarity between the preset standard text and the first recognition text is calculated to obtain the determined preset standard text.

[0020] Optionally, in a possible implementation manner of the first aspect, if it is determined that there is no situation where the preset standard segmentation of the preset standard text is completely identical to the first recognition segmentation, calculating the recognition similarity between the preset standard text and the first recognition text to obtain the determined preset standard text includes:

[0021] Calculating the number of identical segmented words between the preset standard segmented words and the first recognized segmented words of the preset standard text, and determining a first identical coefficient corresponding to each identical first recognized segmented word, each first recognized segmented word having a preset first identical coefficient;

[0022] Based on the comprehensive calculation of the same number, the different number, and the first same coefficient between the preset standard segmentation and the first recognition segmentation, the recognition similarity between the preset standard text and the corresponding first recognition text is obtained, and the recognition similarity is calculated by the following formula:

[0023]

[0024] in, To identify similarity, The same number of preset standard segmentation words and first recognition segmentation words, is the difference between the preset standard segmentation and the first recognition segmentation, is the quantity weight, For the The first identical coefficient of the first identified word, is the upper limit value of the number of identical standard segmentations and first recognition segmentations, is the coefficient constant value, is the coefficient weight value;

[0025] The determined preset standard text is obtained based on the recognition similarity and interaction with the management terminal, and the preset standard participle in the preset standard text is updated based on the first recognition participle, and the first identical coefficient of the preset standard participle is obtained based on the recognition similarity.

[0026] Optionally, in a possible implementation manner of the first aspect, obtaining the determined preset standard text based on the recognition similarity and interacting with the management terminal, updating the preset standard segmentation in the preset standard text based on the first recognition segmentation, and obtaining a first identical coefficient of the preset standard segmentation based on the recognition similarity include:

[0027] Selecting the preset standard texts whose recognition similarity is greater than the minimum similarity and sorting them in descending order to generate a sequence of preset standard texts to be selected;

[0028] Interacting with the management end to determine the preset standard text selected in the preset standard text sequence as the determined preset standard text, and obtaining a first recognition segmentation different from the preset standard segmentation of the determined preset standard text as the training recognition segmentation;

[0029] The training recognition segmentation is used as a new preset standard segmentation in the determined preset standard text, and a first identical coefficient of the preset standard segmentation is obtained based on the recognition similarity calculation.

[0030] Optionally, in a possible implementation manner of the first aspect, taking the training recognition segmentation as a new preset standard segmentation in the determined preset standard text, and calculating a first identical coefficient of the preset standard segmentation based on the recognition similarity, includes:

[0031] Obtaining the training quantity of the training recognition segmentation words, and calculating the quantity coefficient of each training recognition segmentation word based on the training quantity;

[0032] Calculating the difference between the recognition similarity and the maximum similarity to obtain the difference similarity, and performing weighted calculation based on the difference similarity to obtain the similarity coefficient;

[0033] The quantity coefficient and similarity coefficient of each training recognition segmentation are calculated comprehensively to obtain the first identical coefficient of the corresponding preset standard segmentation. The first identical coefficient is calculated by the following formula:

[0034]

[0035] in, For the The first identical coefficient of the training recognition segmentation, is the lowest identical coefficient, For the The number of training words for training recognition segmentation, is the normalized value of quantity, is the maximum similarity, To include The recognition similarity of the first recognition text of the training recognition segmentation, is the normalized similarity value, have the same weight value.

[0036] Optionally, in a possible implementation manner of the first aspect, determining a first identification feature corresponding to a first document element set, identifying the first legal document based on the first identification feature, and obtaining first element content corresponding to the first identification feature includes:

[0037] Determining a first identification feature corresponding to a first document element set, each first document element set having a preset first identification feature;

[0038] Locating the text in the first legal document based on the first recognition feature, obtaining a first text word corresponding to each first recognition feature, and determining a coherent text corresponding to the first text word;

[0039] The continuous text is extracted until a cutoff character appears, and the extracted continuous text is used as the first element content corresponding to the first recognition feature.

[0040] Optionally, in a possible implementation manner of the first aspect, calling a standard element table corresponding to the first attribute, filling the first element content into the standard element table, determining a first identification feature of the first element content that is not obtained as a second identification feature, and determining a first identification feature of the first element content that does not meet requirements as a third identification feature includes:

[0041] Retrieving a standard element table corresponding to the first attribute, each first attribute having a preset standard element table, and filling the first element content into a corresponding slot in the standard element table;

[0042] Traverse all slots in the standard element table in sequence, and use the first identification feature corresponding to the slot that is not filled with the first element content as the second identification feature;

[0043] Traverse all the slots in the standard element table in turn to determine the number of texts corresponding to each slot. If it is determined that the number of texts is less than the minimum number of the corresponding slot, the first identification feature of the first element content that does not meet the requirements will be used as the third identification feature, and the slot corresponding to each identification feature has the minimum number.

[0044] Optionally, in a possible implementation manner of the first aspect, the processing the standard element table based on the second identification feature and the third identification feature to obtain a feedback element table and feeding it back to the target end, receiving feedback information of the target end based on the feedback element table and filling it into the standard element table, obtaining a total standard element table and sending it to the management end, includes:

[0045] The slot corresponding to the first identification feature in the standard feature table is deleted, and the remaining second identification features and third identification features are sorted according to the order in the standard feature table to obtain a feedback feature table;

[0046] Adding an information missing label to the slot corresponding to the second identification feature in the feedback element table, and adding an information missing label to the slot corresponding to the third identification feature in the feedback element table;

[0047] The element missing model performs a comprehensive calculation based on the information missing label and the slot corresponding to the information missing label to obtain the element information missing coefficient, adds the element information missing coefficient to the preset slot of the feedback element table and then feeds it back to the target end. If the element information missing coefficient is greater than the threshold missing coefficient, it is synchronously fed back to the management end;

[0048] After determining that the target end has filled the feedback element table, the feedback information in the slot is extracted and filled into the standard element table to obtain the total standard element table. Based on the element missing model, the element information missing coefficient in the total standard element table is updated and sent to the management end.

[0049] Optionally, in a possible implementation manner of the first aspect, the element missing model performs a comprehensive calculation based on the information missing label and the slot corresponding to the information missing label to obtain an element information missing coefficient, and adds the element information missing coefficient to a preset slot in the feedback element table and then feeds it back to the target end, including:

[0050] The difference between the number of texts in the slot corresponding to the missing label and the minimum number is calculated to obtain the difference text number;

[0051] The factor information missing coefficient is obtained by comprehensive calculation based on the number of difference texts, the number of first labels with missing information labels, and the number of second labels with missing information labels. The factor information missing coefficient is calculated by the following formula:

[0052]

[0053] in, is the factor information missing coefficient, is the number of first tags, is the first label weight, is the number of second tags, is the second label weight, For the The number of difference texts in the slots corresponding to the missing labels of the information, The upper limit of the number of slots corresponding to the missing information label. The normalized value of the difference text.

[0054] Technical Effect 1: Efficient and Accurate Attribute Identification and Factor Selection

[0055] The present invention receives and identifies the attribute keywords of the first legal document through a server, determines the case type and selects the corresponding document element set in the database. In an embodiment, the row is first located by identifying the attribute keywords such as the cause of action, and the text and punctuation related to the keywords are removed, and the remaining text is retained. The document attributes are accurately determined through word segmentation processing and similarity calculation with the preset standard text. This process greatly improves the accuracy and efficiency of document attribute recognition.

[0056] Technical Effect 2: Comprehensive and Detailed Extraction of Element Content

[0057] When extracting the element content, the present invention ensures the integrity of the element content by extracting the identification feature text coherently. In an embodiment, the legal document text is located based on the first identification feature, and the coherent text is extracted until the cutoff character is encountered, and the extracted content is filled into the standard element table. For elements that cannot be identified or do not meet the requirements, the system reprocesses them through the second and third identification features to ensure the integrity and accuracy of the element content. This technology not only reduces human omissions, but also makes the element content more comprehensive.

[0058] Technical Effect 3: Dynamic Update and Self-Optimization Capabilities

[0059] The present invention realizes the dynamic update and self-optimization of the standard element table through the feedback mechanism and the element missing model. In the embodiment, by generating a feedback element table and interacting with the target end and the management end, the system updates the element content and similarity coefficient according to the feedback information. The element missing model calculates the element information missing coefficient through the information missing label and the information missing label to ensure the accuracy of the element content. Through this self-optimization capability, the system can gradually improve the processing accuracy and efficiency of legal documents to meet the dynamic needs in judicial practice. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 It is a flow chart of a method for intelligent identification of legal document elements based on big data provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0061] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0062] See also Figure 1, is a flow chart of a method for intelligently identifying legal document elements based on big data provided by an embodiment of the present invention, the method comprising:

[0063] S1, the server determines the attribute of the first legal document received to obtain a first attribute, and selects a first document element set corresponding to the first attribute in the database.

[0064] Element-based judgment documents are a concise and clearly structured form of judicial documents, which are designed to improve judicial efficiency and transparency. This form of document usually contains the following key elements: Basic case information: including case number, party information, case cause and other basic contents. Focus of dispute: briefly list the main points of dispute and demands of both parties. Fact finding: identify and summarize the basic facts of the case, usually list the key facts concisely. Legal basis: cite relevant legal provisions and regulations to explain the basis for the judgment. Judgment result: clearly announce the court's ruling, including the main content of the judgment and the method of execution. Reasons for judgment: briefly explain the reasons and basis for the court's ruling, explain the application of law and the process of fact finding. Other matters: may include the execution arrangement of the judgment, reminder of the right of appeal and other contents. Element-based judgment documents can reduce unnecessary lengthy descriptions by simplifying and standardizing the structure of documents, facilitate the understanding of parties and the public, and improve judicial transparency and credibility.

[0065] This solution requires the identification and processing of multiple related documents to obtain the required elements in order to generate element-based judgment documents. The first legal document in this disclosure may be uploaded by the parties, such as an indictment, a statement of defense, etc. Among them, the first attribute refers to the type of the first legal document, such as motor vehicle traffic accident damage compensation disputes, labor contract disputes, sales contract disputes, etc. It can be understood that the document elements corresponding to different types are also different. For example, the document elements in a sales contract dispute may include the place where the contract is signed, the contract amount, etc., and the first document element set includes multiple of the above-mentioned document elements.

[0066] In some embodiments, the server determines the attribute of the received first legal document to obtain a first attribute, and selects a first document element set corresponding to the first attribute in the database, including:

[0067] S11, the server identifies attribute keywords of the received first legal document, wherein the attribute keywords include the cause of action, and locates the row corresponding to the first legal document to obtain the first row location.

[0068] After receiving the first legal document, the server first identifies the cause of action in the document, that is, the attribute keywords mentioned above, and locates the text line where these keywords are located, which is called first line positioning. It can be understood that the first legal document will include the cause of action, and the case type can be determined based on the cause of action. For example, the content corresponding to the line where the cause of action is (cause of action: labor contract dispute), then the content of this line can be used to determine that it is a labor contract dispute. This step will first locate the line where the cause of action is located as the first line.

[0069] S12, removing the text and punctuation text corresponding to the attribute keyword in the first line location, and recognizing the remaining text in the first line location to obtain a first recognized text.

[0070] From the first line of positioning, remove the text related to the attribute keyword (such as the cause of action) and the punctuation text (such as:). Recognize the remaining text in the first line of positioning to obtain the first recognized text. The first recognized text is, for example, a dispute over a labor contract.

[0071] S13: If the remaining text in the first row location is empty, determine the corresponding second row location based on the first row location, and recognize the remaining text in the second row location to obtain a first recognized text.

[0072] In some cases, the specific text of the case may be written below the case, not on the same line as the case, and this solution will make a judgment and process. If there is no remaining text in the first line positioning, the server will determine the corresponding second line positioning based on the first line positioning. It can be understood that the second line positioning is the next line of the first line positioning. The remaining text in the second line positioning is recognized, and the first recognition text is also obtained.

[0073] S14, determining the corresponding preset standard text and the attribute corresponding to the preset standard text by performing word segmentation processing on the first recognized text, obtaining a first attribute, selecting a first document element set corresponding to the first attribute in the database, and each first attribute has a preset first document element set.

[0074] The first recognized text is segmented to determine the preset standard text to which it corresponds. According to the attribute (first attribute) recognized by the preset standard text, a first document element set corresponding to the attribute is selected from the database. Each first attribute has a preset first document element set, and these element sets may contain specific elements.

[0075] The word segmentation processing of the first recognized text determines the corresponding preset standard text, and the attribute corresponding to the preset standard text obtains a first attribute, and a first document element set corresponding to the first attribute in the database is selected, and each first attribute has a preset first document element set, including:

[0076] S141, recognize and remove preset elimination words in the first recognized text, and segment the remaining first recognized text to obtain a plurality of first recognized segmented words.

[0077] Identify and remove preset exclusion words (such as dispute) from the first recognition text to clean up the text and prepare for more accurate word segmentation. Perform word segmentation on the remaining first recognition text to obtain multiple first recognition word segments in preparation for the next step of text matching. For motor vehicle traffic accident damage compensation disputes, the first recognition word segmentation can be, for example, motor vehicle, traffic accident, damage, and compensation.

[0078] S142, comparing the first recognition segmentation with the preset standard segmentation of each preset standard text, if there is a preset standard segmentation of the preset standard text that is exactly the same as the first recognition segmentation, then taking the corresponding preset standard text as the determined preset standard text.

[0079] Among them, the preset standard text includes multiple preset standard segmentation words. This solution will compare the recognized first recognition segmentation word with the standard segmentation word of the preset standard text in the database. If there is a preset standard segmentation word that is exactly the same as the first recognition segmentation word, the standard text is determined to be the preset standard text. For example, if the first recognition segmentation word is 5, then the preset standard segmentation word is also 5, and at this time, 5 is exactly the same as 5 preset standard segmentations.

[0080] S143: If it is determined that there is no situation where the preset standard word segmentation of the preset standard text is completely identical to the first recognition word segmentation, the recognition similarity between the preset standard text and the first recognition text is calculated to obtain the determined preset standard text.

[0081] It is understandable that if it is determined that there is no preset standard text in which the preset standard word segmentation is exactly the same as the first recognition word segmentation, it indicates that there may be an error. At this time, the present solution needs to calculate the recognition similarity to obtain the determined preset standard text.

[0082] If it is determined that there is no situation where the preset standard segmentation of the preset standard text is completely identical to the first recognition segmentation, calculating the recognition similarity between the preset standard text and the first recognition text to obtain the determined preset standard text includes:

[0083] Calculate the number of identical participles between the preset standard participles of the preset standard text and the first recognition participles, and determine the first identical coefficient corresponding to each identical first recognition participle, and each first recognition participle has a preset first identical coefficient. When calculating the similarity, this scheme needs to calculate the number of identical participles between the preset standard participles of the preset standard text and the first recognition participles. It can be understood that the more identical participles there are, the more similar they are. At the same time, it is also necessary to calculate the first identical coefficient corresponding to each identical first recognition participle, and different first recognition participles have preset first identical coefficients. For example, "contract" has a first identical coefficient, and "labor" has a first identical coefficient. Since "contract" exists in many cases, and "labor" only exists in a small number of cases, therefore, when the coefficient is preset, the first identical coefficient of "labor" needs to be greater than the first identical coefficient of "contract".

[0084] Based on the comprehensive calculation of the same number, the different number, and the first same coefficient between the preset standard segmentation and the first recognition segmentation, the recognition similarity between the preset standard text and the corresponding first recognition text is obtained, and the recognition similarity is calculated by the following formula:

[0085]

[0086] in, To identify similarity, The same number of preset standard segmentation words and first recognition segmentation words, is the difference between the preset standard segmentation and the first recognition segmentation, is the quantity weight, For the The first identical coefficient of the first identified participle, is the upper limit value of the number of identical standard segmentations and first recognition segmentations, is the coefficient constant value, is the coefficient weight value.

[0087] In the above formula, The coefficient representing the quantitative dimension, The larger it is, the larger the corresponding coefficient is. Represents the value of the first identical coefficient dimension. Finally, the data of the above two dimensions are integrated by combining the quantity weight and the coefficient weight value to obtain the recognition similarity. Among them, the quantity weight and the coefficient weight value can be pre-set by the staff, and the quantity weight can be set greater than the coefficient weight value.

[0088] The determined preset standard text is obtained based on the recognition similarity and interaction with the management terminal, and the preset standard participle in the preset standard text is updated based on the first recognition participle, and the first identical coefficient of the preset standard participle is obtained based on the recognition similarity.

[0089] Through the above steps, the system can not only automatically process the text and select the most suitable preset standard text, but also continuously optimize itself to improve the accuracy of processing through interaction with the management end and subsequent update process.

[0090] In the above embodiment, the determined preset standard text is obtained based on the recognition similarity and the interaction with the management terminal, and the preset standard segmentation in the preset standard text is updated based on the first recognition segmentation, and the first identical coefficient of the preset standard segmentation is obtained based on the recognition similarity, including:

[0091] Select the preset standard texts with recognition similarity greater than the minimum similarity and sort them in descending order to generate a preset standard text sequence to be selected. The system first selects the preset standard texts with recognition similarity greater than the minimum similarity standard. It can be understood that these selected texts are texts with higher familiarity. Then, these texts are sorted in descending order according to recognition similarity to form a preset standard text sequence to be selected. It can be understood that the higher the order, the greater the similarity.

[0092] Interact with the management end to determine the preset standard text selected in the preset standard text sequence as the determined preset standard text, and obtain the first recognition segmentation that is different from the preset standard segmentation of the determined preset standard text as the training recognition segmentation. In this solution, the management end can review the preset standard text sequence to be selected through an interactive interface. The management end selects a preset standard text as the determined preset standard text based on its own judgment or based on certain standards. At the same time, recognition segmentations that are different from its preset standard segmentations are extracted from the determined preset standard text, and these segmentations are called "training recognition segmentations."

[0093] The training recognition segmentation is used as a new preset standard segmentation in the determined preset standard text, and the first identical coefficient of the preset standard segmentation is calculated based on the recognition similarity. This solution integrates the training recognition segmentation into the determined preset standard text, replacing or adding it to the existing preset standard segmentation. This step is crucial to maintaining the practical applicability and high accuracy of the preset standard text. At the same time, based on the recognition similarity, the "first identical coefficient" of the updated preset standard segmentation is calculated.

[0094] The step of using the training recognition segmentation as a new preset standard segmentation in the determined preset standard text and calculating a first identical coefficient of the preset standard segmentation based on recognition similarity includes:

[0095] The training quantity of the training recognition segmentation is obtained, and the quantity coefficient of each training recognition segmentation is calculated based on the training quantity. This solution needs to record the number of times each training recognition segmentation appears in a series of training processes, and this number is the training quantity. Given that the number of times different training recognition segmentations may appear varies greatly, in order to facilitate subsequent processing, the system will use a quantified value (quantity normalization value) ) to calculate the quantity coefficient of each training recognition word. This quantity coefficient is obtained by dividing the training quantity by the quantity normalization value It is used to balance the influence of each participle in the comprehensive evaluation.

[0096] The difference between the recognition similarity and the maximum similarity is calculated to obtain the difference similarity, and the similarity coefficient is obtained by weighted calculation based on the difference similarity. This solution calculates the difference between the maximum similarity and the actual recognition similarity to obtain the difference similarity. The greater the difference similarity, the more the corresponding segmentation can represent the case. When the difference similarity is greater, it is considered to be corresponding, and the corresponding segmentation can be regarded as having greater importance, so the corresponding segmentation can represent the case more. The difference similarity will be normalized by a similarity value Weighting is performed to obtain the similarity coefficient.

[0097] The quantity coefficient and similarity coefficient of each training recognition segmentation are calculated comprehensively to obtain the first identical coefficient of the corresponding preset standard segmentation. The first identical coefficient is calculated by the following formula:

[0098]

[0099] in, For the The first identical coefficient of the training recognition segmentation, is the lowest identical coefficient, For the The number of training words for training recognition segmentation, is the normalized value of quantity, is the maximum similarity, To include The recognition similarity of the first recognition text of the training recognition segmentation, is the normalized similarity value, have the same weight value.

[0100] In the above formula, For example, it is 0.3, the preset minimum value. The coefficient representing the quantity dimension. The larger the number of trainings, the smaller the corresponding coefficient. Represents the similarity coefficient. The larger the difference, the larger the corresponding similarity coefficient. Among them, the quantity normalization value, the similarity normalization value and the same weight value can be preset by the user.

[0101] Through this composite calculation method, the system is able to assign an accurate first identical coefficient to each training recognition segmentation, which comprehensively considers the similarity difference between the training quantity and the recognition text. These coefficients are then used to update the preset standard text of legal documents to improve the accuracy of future text recognition and matching.

[0102] S2, determining a first identification feature corresponding to a first document element set, identifying the first legal document based on the first identification feature, and obtaining a first element content corresponding to the first identification feature.

[0103] In some embodiments, determining the first identification feature corresponding to the first document element set, identifying the first legal document based on the first identification feature, and obtaining the first element content corresponding to the first identification feature includes:

[0104] S21, determining a first identification feature corresponding to a first document element set, each first document element set having a preset first identification feature.

[0105] First, this solution needs to determine the first identification feature corresponding to the first document element set (such as the subject matter, the subject amount, etc.). For example, the preset first identification feature corresponding to the subject matter may be the subject matter, the subject amount.

[0106] S22, locating the text in the first legal document based on the first recognition feature, obtaining the first text word corresponding to each first recognition feature, and determining the coherent text corresponding to the first text word.

[0107] This solution will locate the text in the legal document in combination with the preset first identification feature. This usually involves natural language processing technology, such as pattern matching, keyword search, etc., to find text keywords corresponding to the first identification feature in the document. After finding these keywords, the location of the keywords will be determined, and coherent text will be extracted from this location. Coherent text refers to the text fragments after the keywords, which usually contain the complete elements of the legal document. For example, coherent text can be "the subject amount is 10,000 yuan."

[0108] S23, extracting the continuous text until a cutoff character appears, and using the extracted continuous text as the first element content corresponding to the first recognition feature.

[0109] This solution will continue to read and extract coherent text along the location of the keyword until it encounters a preset cutoff character. The cutoff character can be a period, comma, semicolon, paragraph mark, etc., indicating the end of a complete element content. After extracting the coherent text, the system uses this text as the first element content corresponding to the first recognition feature. Using the above steps, key elements can be accurately extracted from legal documents, making document analysis and processing faster and more accurate. This method can significantly reduce the time and labor costs of manually processing documents, while reducing errors caused by manual operations.

[0110] S3, call the standard element table corresponding to the first attribute, fill the first element content into the standard element table, determine the first identification feature of the first element content that is not obtained as the second identification feature, and determine the first identification feature of the first element content that does not meet the requirements as the third identification feature.

[0111] In some embodiments, the retrieving the standard element table corresponding to the first attribute, filling the first element content into the standard element table, determining the first identification feature of the first element content that is not obtained as the second identification feature, and determining the first identification feature of the first element content that does not meet the requirements as the third identification feature, includes:

[0112] S31, calling a standard element table corresponding to the first attribute, each first attribute having a preset standard element table, and filling the first element content into a corresponding slot in the standard element table.

[0113] First, it is necessary to retrieve the preset standard element table according to the first attribute. The standard element table is pre-set and contains all possible elements in the element-based judgment document. This solution will fill the extracted first element content into the corresponding slot in the standard element table.

[0114] S32, traverse all the slots in the standard element table in sequence, and use the first identification feature corresponding to the slot that is not filled with the first element content as the second identification feature.

[0115] This solution will traverse all the slots in the standard feature table to check whether there are any slots that are not filled with content. For those slots that are not filled with content, the first identification features corresponding to these slots are marked as second identification features. These second identification features are for elements that could not be extracted in the first round of processing.

[0116] S33, traverse all the slots in the standard element table in turn to determine the number of texts corresponding to each slot. If it is determined that the number of texts is less than the minimum number of the corresponding slot, the first identification feature of the first element content that does not meet the requirements is used as the third identification feature, and the slot corresponding to each identification feature has the minimum number.

[0117] The system will continue to traverse all slots in the standard feature table to determine the number of texts corresponding to each slot. If it is determined that the number of texts corresponding to a slot is less than the minimum number specified for the slot, the system will mark the first identification feature of the slot as the third identification feature. This means that the extracted feature content does not meet the requirements.

[0118] These secondary and tertiary identification features will be used in subsequent steps, which may include re-identification, extraction, or verification of feature content that was not successfully extracted or did not meet the requirements.

[0119] S4, based on the second identification feature and the third identification feature, the standard element table is processed to obtain a feedback element table which is fed back to the target end, based on the feedback element table, feedback information from the target end is received and filled into the standard element table, and the total standard element table is obtained and sent to the management end.

[0120] In some embodiments, the processing of the standard element table based on the second identification feature and the third identification feature to obtain a feedback element table and feeding it back to the target end, receiving feedback information from the target end based on the feedback element table and filling it into the standard element table, obtaining a total standard element table and sending it to the management end, includes:

[0121] S41, deleting the slot corresponding to the first identification feature in the standard feature table, and sorting the remaining second identification features and third identification features according to the order in the standard feature table to obtain a feedback feature table.

[0122] Delete those slots that have been successfully filled in the first round of processing from the standard feature table, and only keep those slots with no content or insufficient content. Sort the remaining second recognition features and third recognition features according to the order in the standard feature table to generate a feedback feature table.

[0123] S42, adding an information missing label to the slot corresponding to the second identification feature in the feedback element table, and adding an information missing label to the slot corresponding to the third identification feature in the feedback element table.

[0124] In the feedback element table, a "missing information" label is added to the slot corresponding to the second identification feature, and a "missing information" label is added to the slot corresponding to the third identification feature.

[0125] S43, the element missing model performs comprehensive calculation based on the information missing label and the slot corresponding to the information missing label to obtain the element information missing coefficient, adds the element information missing coefficient to the preset slot of the feedback element table and then feeds it back to the target end, and if the element information missing coefficient is greater than the threshold missing coefficient, it is synchronously fed back to the management end;

[0126] Using the missing element model, a comprehensive calculation is performed based on the slots corresponding to the "missing information" and "missing information" labels to obtain the missing element information coefficient. The calculated missing element information coefficient is added to the preset slot of the feedback element table. The feedback element table with the missing element information coefficient is sent back to the target end so that the target end can view and supplement the missing information. If the missing element information coefficient is greater than a preset threshold coefficient, the information is synchronously fed back to the management end for further processing.

[0127] The missing element model performs a comprehensive calculation based on the missing information label and the slot corresponding to the missing information label to obtain the missing element information coefficient, adds the missing element information coefficient to the preset slot of the feedback element table, and then feeds back to the target end, including:

[0128] The difference between the number of texts in the slot corresponding to the missing label and the minimum number is calculated to obtain the difference text number. It is understandable that the larger the difference text number, the more serious the problem.

[0129] The factor information missing coefficient is obtained by comprehensive calculation based on the number of difference texts, the number of first labels with missing information labels, and the number of second labels with missing information labels. The factor information missing coefficient is calculated by the following formula:

[0130]

[0131] in, is the factor information missing coefficient, is the number of first tags, is the first label weight, is the number of second tags, is the second label weight, For the The number of difference texts in the slots corresponding to the missing labels of the information, The upper limit of the number of slots corresponding to the missing information label. The normalized value of the difference text.

[0132] In the above formula, Represents the coefficient of the first label quantity dimension. The larger the number of first labels, the larger the corresponding coefficient. The coefficient representing the dimension of the number of second labels. The larger the number of second labels, the larger the corresponding coefficient. The coefficient representing the dimension of the number of difference texts. The greater the difference, the greater the coefficient. Since missing is more serious than missing, the weight of the first label can be greater than the weight of the second label. The normalized value of the difference text can be preset by the user.

[0133] The missing element coefficient reflects the overall degree of missing element information. The higher this coefficient is, the more serious the missing elements of the text are, and the more supplementation or correction work is needed. By adding this coefficient to the feedback element table, the target end can have a clear idea of ​​the information that needs to be supplemented or modified, and perform corresponding operations accordingly.

[0134] S44, after determining that the target end has filled the feedback element table, the feedback information in the slot is extracted and filled into the standard element table to obtain a total standard element table, and the element information missing coefficient in the total standard element table is updated based on the element missing model and sent to the management end.

[0135] After the target end fills the feedback element table, the system extracts the feedback information in the slot. The feedback information provided by the target end is filled into the original standard element table to obtain the updated total standard element table. Based on the element missing model, the element information missing coefficient in the total standard element table is updated. The updated total standard element table is sent to the management end for final management and review. This process helps to ensure that the elements of the legal document are complete and accurate, complete or correct information through feedback from the target end, and ensure the quality of the legal document processing process through the final review of the management end.

[0136] The present invention also provides a storage medium, in which a computer program is stored. When the computer program is executed by a processor, it is used to implement the methods provided by the various embodiments described above.

[0137] Among them, the storage medium can be a computer storage medium or a communication medium. The communication medium includes any medium that facilitates the transmission of a computer program from one place to another. The computer storage medium can be any available medium that can be accessed by a general-purpose or special-purpose computer. For example, the storage medium is coupled to the processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can be located in an application-specific integrated circuit (Application Specific Integrated Circuits, referred to as: ASIC). In addition, the ASIC can be located in a user device. Of course, the processor and the storage medium can also exist in a communication device as discrete components. The storage medium can be a read-only memory (ROM), a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, and an optical data storage device, etc.

[0138] The present invention also provides a program product, which includes an execution instruction, which is stored in a storage medium. At least one processor of a device can read the execution instruction from the storage medium, and at least one processor executes the execution instruction so that the device implements the methods provided in the above various embodiments.

[0139] In the above-mentioned terminal or server embodiments, it should be understood that the processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. A general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in the present invention may be directly implemented as being executed by a hardware processor, or may be executed by a combination of hardware and software modules in the processor.

[0140] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent identification method for legal document elements based on big data, characterized in that: include: The server determines the attribute of the first legal document received to obtain a first attribute, and selects a first document element set corresponding to the first attribute in the database; Determine a first identification feature corresponding to the first document element set, identify the first legal document based on the first identification feature, and obtain a first element content corresponding to the first identification feature; Retrieving a standard element table corresponding to the first attribute, filling the first element content into the standard element table, determining the first identification feature of the first element content that is not obtained as the second identification feature, and determining the first identification feature of the first element content that does not meet the requirement as the third identification feature; Based on the second identification feature and the third identification feature, the standard element table is processed to obtain a feedback element table which is fed back to the target end; based on the feedback element table, feedback information from the target end is received and filled into the standard element table, and a total standard element table is obtained and sent to the management end; The retrieving the standard element table corresponding to the first attribute, filling the first element content into the standard element table, determining the first identification feature that does not obtain the first element content as the second identification feature, and determining the first identification feature of the first element content that does not meet the requirements as the third identification feature, includes: Retrieving a standard element table corresponding to the first attribute, each first attribute having a preset standard element table, and filling the first element content into a corresponding slot in the standard element table; Traverse all slots in the standard element table in sequence, and use the first identification feature corresponding to the slot that is not filled with the first element content as the second identification feature; Traverse all the slots in the standard element table in turn to determine the number of texts corresponding to each slot. If it is determined that the number of texts is less than the minimum number of the corresponding slot, the first identification feature of the first element content that does not meet the requirements will be used as the third identification feature, and the slot corresponding to each identification feature has the minimum number.

2. The intelligent identification method of legal document elements based on big data according to claim 1 is characterized in that: The server determines the attribute of the received first legal document to obtain a first attribute, and selects a first document element set corresponding to the first attribute in the database, including: The server identifies attribute keywords of the received first legal document, wherein the attribute keywords include the cause of action, and locates the line corresponding to the first legal document to obtain the first line location; Eliminating the text and punctuation marks corresponding to the attribute keyword in the first line location, and recognizing the remaining text in the first line location to obtain a first recognized text; If the remaining text in the first row position is empty, determining the corresponding second row position based on the first row position, and recognizing the remaining text in the second row position to obtain a first recognized text; The first recognized text is segmented to determine the corresponding preset standard text, and the attribute corresponding to the preset standard text is obtained to obtain a first attribute, and a first document element set corresponding to the first attribute in the database is selected, and each first attribute has a preset first document element set.

3. The intelligent identification method of legal document elements based on big data according to claim 2 is characterized in that: The word segmentation processing of the first recognized text determines the corresponding preset standard text, and the attribute corresponding to the preset standard text obtains a first attribute, and selects a first document element set corresponding to the first attribute in the database, each first attribute having a preset first document element set, including: Recognize and remove the preset removal words in the first recognition text, and perform word segmentation on the remaining first recognition text to obtain a plurality of first recognition word segments; Compare the first recognition segmented word with the preset standard segmented word of each preset standard text, and if there is a preset standard segmented word of the preset standard text that is exactly the same as the first recognition segmented word, use the corresponding preset standard text as the determined preset standard text; If it is determined that there is no situation where the preset standard word segment of the preset standard text is completely identical to the first recognition word segment, the recognition similarity between the preset standard text and the first recognition text is calculated to obtain the determined preset standard text.

4. The intelligent identification method of legal document elements based on big data according to claim 3 is characterized in that: If it is determined that there is no situation where the preset standard segmentation of the preset standard text is completely identical to the first recognition segmentation, calculating the recognition similarity between the preset standard text and the first recognition text to obtain the determined preset standard text includes: Calculating the number of identical segmented words between the preset standard segmented words and the first recognized segmented words of the preset standard text, and determining a first identical coefficient corresponding to each identical first recognized segmented word, each first recognized segmented word having a preset first identical coefficient; Based on the comprehensive calculation of the same number, the different number, and the first same coefficient between the preset standard segmentation and the first recognition segmentation, the recognition similarity between the preset standard text and the corresponding first recognition text is obtained, and the recognition similarity is calculated by the following formula: in, To identify similarity, The same number of preset standard segmentation words and first recognition segmentation words, is the difference between the preset standard segmentation and the first recognition segmentation, is the quantity weight, For the The first identical coefficient of the first identified word, is the upper limit value of the number of identical standard segmentations and first recognition segmentations, is the coefficient constant value, is the coefficient weight value; The determined preset standard text is obtained based on the recognition similarity and interaction with the management terminal, and the preset standard participle in the preset standard text is updated based on the first recognition participle, and the first identical coefficient of the preset standard participle is obtained based on the recognition similarity.

5. The method for intelligent identification of legal document elements based on big data according to claim 4 is characterized in that: The method of obtaining the determined preset standard text based on the recognition similarity and interacting with the management terminal, updating the preset standard segmentation words in the preset standard text based on the first recognition segmentation words, and obtaining the first identical coefficient of the preset standard segmentation words based on the recognition similarity includes: Selecting the preset standard texts whose recognition similarity is greater than the minimum similarity and sorting them in descending order to generate a sequence of preset standard texts to be selected; Interacting with the management end to determine the preset standard text selected in the preset standard text sequence as the determined preset standard text, and obtaining a first recognition segmentation different from the preset standard segmentation of the determined preset standard text as the training recognition segmentation; The training recognition segmentation is used as a new preset standard segmentation in the determined preset standard text, and a first identical coefficient of the preset standard segmentation is obtained based on the recognition similarity calculation.

6. The intelligent identification method of legal document elements based on big data according to claim 5 is characterized in that: The step of using the training recognition segmentation as a new preset standard segmentation in the determined preset standard text and calculating a first identical coefficient of the preset standard segmentation based on the recognition similarity includes: Obtaining the training quantity of the training recognition segmentation words, and calculating the quantity coefficient of each training recognition segmentation word based on the training quantity; Calculating the difference between the recognition similarity and the maximum similarity to obtain the difference similarity, and performing weighted calculation based on the difference similarity to obtain the similarity coefficient; The quantity coefficient and similarity coefficient of each training recognition segmentation are calculated comprehensively to obtain the first identical coefficient of the corresponding preset standard segmentation. The first identical coefficient is calculated by the following formula: in, For the The first identical coefficient of the training recognition segmentation, is the lowest identical coefficient, For the The number of training words for training recognition segmentation, is the normalized value of quantity, is the maximum similarity, To include The recognition similarity of the first recognition text of the training recognition segmentation, is the normalized similarity value, have the same weight value.

7. The intelligent identification method of legal document elements based on big data according to claim 1 is characterized in that: The determining of the first identification feature corresponding to the first document element set, identifying the first legal document based on the first identification feature, and obtaining the first element content corresponding to the first identification feature includes: Determining a first identification feature corresponding to a first document element set, each first document element set having a preset first identification feature; Locating the text in the first legal document based on the first recognition feature, obtaining a first text word corresponding to each first recognition feature, and determining a coherent text corresponding to the first text word; The continuous text is extracted until a cutoff character appears, and the extracted continuous text is used as the first element content corresponding to the first recognition feature.

8. The intelligent identification method of legal document elements based on big data according to claim 1 is characterized in that: The step of processing the standard element table based on the second identification feature and the third identification feature to obtain a feedback element table and feeding it back to the target end, receiving feedback information from the target end based on the feedback element table and filling it into the standard element table, obtaining a total standard element table and sending it to the management end includes: The slot corresponding to the first identification feature in the standard feature table is deleted, and the remaining second identification features and third identification features are sorted according to the order in the standard feature table to obtain a feedback feature table; Adding an information missing label to the slot corresponding to the second identification feature in the feedback element table, and adding an information missing label to the slot corresponding to the third identification feature in the feedback element table; The element missing model performs a comprehensive calculation based on the information missing label and the slot corresponding to the information missing label to obtain the element information missing coefficient, adds the element information missing coefficient to the preset slot of the feedback element table and then feeds it back to the target end. If the element information missing coefficient is greater than the threshold missing coefficient, it is synchronously fed back to the management end; After determining that the target end has filled the feedback element table, the feedback information in the slot is extracted and filled into the standard element table to obtain the total standard element table. Based on the element missing model, the element information missing coefficient in the total standard element table is updated and sent to the management end.

9. The method for intelligent identification of legal document elements based on big data according to claim 8 is characterized in that: The element missing model performs a comprehensive calculation based on the information missing label and the slot corresponding to the information missing label to obtain the element information missing coefficient, adds the element information missing coefficient to the preset slot of the feedback element table, and then feeds back to the target end, including: The difference between the number of texts in the slot corresponding to the missing label and the minimum number is calculated to obtain the difference text number; The factor information missing coefficient is obtained by comprehensive calculation based on the number of difference texts, the number of first labels with missing information labels, and the number of second labels with missing information labels. The factor information missing coefficient is calculated by the following formula: in, is the factor information missing coefficient, is the number of first tags, is the first label weight, is the number of second tags, is the second label weight, For the The number of difference texts in the slots corresponding to the missing labels of the information, The upper limit of the number of slots corresponding to the missing information label. The normalized value of the difference text.

Images