Intelligent notarization case matching method, electronic equipment and storage medium

Abstract

The invention provides an intelligent notarization case pushing method, electronic equipment and a storage medium, and the method comprises the steps: obtaining a matching feature of a to-be-checked notarization case, the matching feature being at least from a pre-collected case material of the to-be-checked notarization case and first user input data; based on the matching features, calculating a class case notarization case most matched with the notarization case to be checked in a notarization case database; and sending the class case notarization case to a request end. According to the method, the matching features used for class case matching are obtained from the case materials and the first user input data respectively, class case screening is performed in the notarization case database by utilizing the matching features, and therefore class case matching is performed through double guidance based on requirements and the extracted matching features according to the requirements, and class case matching efficiency is improved. Therefore, class case notarization cases closer to user requirements can be obtained.

Description

Technical Field

[0001] The present application mainly relates to the field of notarization information technology; specifically, the present application relates to an intelligent matching method for notarization cases, an electronic device and a storage medium. Background Art

[0002] Notarization case, also known as typical notarization case, refers to representative notarization cases that can reflect the legal principles, rules and procedures of notarization. In notarization practice, there are many application scenarios for notarization case inquiries, such as: 1. Establish an internal inquiry system for notarization affairs, collect and review relevant notarization evidence through association, so as to improve the efficiency and quality of notarization work, and ensure the consistency and accuracy of notarization affairs, such as will notarization. 2. Notaries refer to previous successful practices and solutions when handling matching cases to ensure the consistency and correctness of legal application. 3. Notaries improve their professional level and service capabilities through systematic case inquiries, self-learning and reference to case experience. 4. For complex matters such as will notarization, property division, intellectual property confirmation, etc., the parties or judges can provide auxiliary evidence for subsequent possible legal procedures (such as litigation, inheritance, transfer, etc.) through case inquiries. Based on this, the applicant found that with different needs for case inquiries, even if the notarization cases to be checked are the same, the target case inquiries may be different.

[0003] However, the traditional method of similar case search usually matches related similar cases by keyword matching. This method of search considers relatively single factors, and whether the most valuable similar cases can be found depends entirely on the experience and knowledge of the searcher. From the application scenarios listed above, it can be seen that in addition to notaries, searchers may also be ordinary applicants, parties or judges, etc. For users who are not notary professionals, it is more difficult to search for the most valuable similar cases. Summary of the invention

[0004] In view of this, the present application provides a notarization case intelligent matching method and an electronic device and storage medium for executing the method. The specific technical solution of the present application is as follows:

[0005] According to one aspect of an embodiment of the present application, a method for intelligent matching of similar notarial cases is provided, including: obtaining matching features of a notarial case to be checked, the matching features being derived from at least pre-collected case materials of the notarial case to be checked and first user input data; based on the matching features, calculating a similar notarial case in a notarial case database that best matches the notarial case to be checked; and sending the similar notarial case to a requesting end.

[0006] According to another aspect of an embodiment of the present application, there is provided an electronic device, comprising: at least one processor, at least one memory, and at least one communication bus, wherein a computer program is stored in the memory, and the processor reads the computer program in the memory through the communication bus; when the computer program is executed by the processor, any one of the above-mentioned intelligent matching methods based on notarization-related cases is implemented.

[0007] According to another aspect of the embodiments of the present application, a computer-readable storage medium is provided, including: a computer program stored thereon, and when the computer program is executed by a computer processor, any of the above-mentioned intelligent matching methods based on notarization-related cases is implemented.

[0008] Beneficial effects:

[0009] The embodiments of the present application obtain matching features for similar case matching from the case materials and the first user input data respectively, and use the matching features to screen similar cases in the notarization case database, thereby matching similar cases based on the needs and the extracted matching features through dual guidance, and obtaining similar notarization cases that are closer to user needs.

[0010] The above summary is for illustrative purposes only and is not intended to be limiting in any way. In addition to the illustrative aspects, embodiments and features described above, further aspects, embodiments and features of the present invention will be readily apparent by reference to the accompanying drawings and the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in describing the embodiments of the present application.

[0012] Figure 1 It is a schematic diagram of an operating environment of a notarization case intelligent matching method shown in an exemplary embodiment of the present application;

[0013] Figure 2 It is a schematic diagram of a system framework of a similar case matching module of a notarization similar case intelligent matching method shown in an exemplary embodiment of the present application;

[0014] Figure 3 It is a flowchart of a notarization case intelligent matching method shown in an exemplary embodiment of the present application;

[0015] Figure 4 It is a flowchart of a matching feature generation step of a notarization case intelligent matching method shown in an exemplary embodiment of the present application;

[0016] Figure 5It is a flowchart of a matching degree calculation method in a notarization case intelligent matching method shown in an exemplary embodiment of the present application;

[0017] Figure 6 It is a flowchart of a method for selecting the most matching notarization case in a notarization case intelligent matching method shown in an exemplary embodiment of the present application;

[0018] Figure 7 It is a schematic diagram of an electronic device shown in an exemplary embodiment of the present application. DETAILED DESCRIPTION

[0019] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be interpreted as limiting the present application.

[0020] It will be understood by those skilled in the art that, unless expressly stated, the singular forms "one", "said", and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present application refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we refer to an element as being "connected" or "coupled" to another element, it may be directly connected or coupled to the other element, or there may be intermediate elements. In addition, the "connection" or "coupling" used herein may include wireless connection or wireless coupling. The term "and / or" used herein includes all or any unit and all combinations of one or more associated listed items.

[0021] As mentioned above, the demand for notarization case inquiries comes from a variety of application scenarios, and the identity roles of the inquiring personnel are different. Even if the notarization case to be checked is the same, the target query results are different due to different application scenarios. For example: the notarization case to be checked is a will notarization; for the applicant, the purpose of the applicant's need to conduct a similar case query may be to understand the handling process of the will notarization and the materials that need to be prepared. Therefore, the possible matching features are the type of will, procedure, identity eligibility, etc.; and if there is a focus of dispute in the will notarization; the inquiring personnel may be a judge, a lawyer, or a party, and the possible matching features are the focus of dispute, legal basis, and other notarizations related to the party; and if it is a pending will notarization, for a novice notary, the desired matching features may be legal basis, handling process, notarial certificate, etc. The three situations listed above may result in different matching results. Therefore, in order to cope with different application scenarios and query requirements raised by different users, providing a personalized case matching solution has become a problem that needs to be solved at present. To this end, the embodiment of the present application provides a notarization case intelligent matching method. First, the operating environment of the technical solution involved in one or more embodiments of this specification is explained.

[0022] An exemplary embodiment of the present application provides an intelligent matching method for notarial cases, which can be run in an electronic device that deploys AI engines such as intent recognition models and matching algorithm models. The electronic device can be a computer device configured with a von Neumann architecture, for example, the computer device can be a desktop computer, a laptop computer, a server, a virtual machine, etc. At the same time, the program corresponding to the intelligent matching method for notarial cases can also be stored in a computer-readable storage medium, which can be a hard disk, a CD-ROM, a cloud storage, etc.

[0023] In order to make the purpose, technical solutions and advantages of the present application clearer, the following will take the server cluster of the present method running on the notary cloud platform as an example, and further describe the embodiments of the present application in detail in combination with other drawings.

[0024] Figure 1 A schematic diagram of the operating environment of a notarization case-like intelligent matching method provided for an exemplary embodiment of the present application. The technical solution of the exemplary embodiment of the present application can be deployed on a notarization cloud platform 100; and when a user successfully logs in to the notarization cloud platform 100 through a request terminal 200, a similar case query service is provided for the user. The notarization cloud platform 100 includes at least a similar case query module 10. Please refer to Figure 2, the similar case query module 10 at least includes a user input unit 11, a case material uploading unit 12, a similar case display unit 13 and a feedback unit 14. The user input unit 11 is used to provide a dialog box for user input and receive the first user input data, and send the first user input data to other functional components of the notarization cloud platform 100. The case material uploading unit 12 is used to provide the user with an upload entry for the case materials of the notarization case to be checked. Of course, not every similar case query requires the user to actively upload the case materials of the notarization case to be checked; that is, if the user knows that the case materials of the notarization case to be checked have been uploaded to the notarization cloud platform 100, there is no need to upload the case materials, and only needs to directly retrieve them from the notarization cloud platform 100 with a unique code. The similar case display unit 13 is used for the system to display the hit similar case notarization case on the operation module interface after obtaining the similar case matching result, so that the user can read it. The feedback unit 14 is used to collect user feedback information on similar case matching results. The feedback unit 14 can collect user feedback information in the form of feedback options or text information, thereby providing a basis for improving the quality of similar case query services.

[0025] Please refer to Figure 3 An exemplary embodiment of the present application provides a notarization case intelligent matching method, which runs on a notarization cloud platform and specifically includes the following steps:

[0026] SP1. Obtain matching features of the notarization case to be checked, where the matching features are at least derived from pre-collected case materials of the notarization case to be checked and first user input data.

[0027] After the user successfully logs in to the notarial cloud platform, he will find the user interface of the similar case query module on the notarial cloud platform. The user can upload the case materials of the notarial case to be checked to the similar case query module, and enter the query information according to the purpose of his own similar case query. Of course, if the user knows that the notarial case to be checked has been stored on the notarial cloud platform, he can also input the unique code assigned to the notarial case by the notarial cloud platform into the case material upload unit of the similar case query module, so that the notarial cloud platform can directly retrieve the notarial case to be checked to the similar case query module based on the unique code, so as to carry out the next step of matching feature generation task.

[0028] After receiving the first user input data and the case materials of the notarized case to be checked, the similar case query module extracts information from the case materials of the notarized case to be checked and the first user input data to generate matching features. Figure 4 , one possible implementation is:

[0029] SP11. First, extract information from the case materials based on multiple preset feature dimensions to generate structured data.

[0030] After analyzing typical notarization cases, the applicant found that at least 7 valuable features can be extracted from notarization case materials; therefore, a possible implementation method is to pre-set the feature dimensions of notarization cases, including but not limited to: basic information, notarization matters, case types, legal basis, factual features, legal relations and evidence features; and define the scope of information extraction for each feature dimension. For example, taking the above feature dimensions as an example, the following definition can be used:

[0031] Basic information: case number, case name, applicant, person to be notarized, notarization matter, notarization date, notarization agency, case progress, etc.;

[0032] Notarization matters: entrustment, declaration, evidence preservation, inheritance, will and other notarization matters;

[0033] Case types: inheritance, property distribution, contracts, debts, ownership, identity proof, etc.;

[0034] Legal basis: relevant laws and regulations, judicial interpretations, precedents, etc.;

[0035] Factual characteristics: the facts, persons involved, time and place involved in the case, etc.;

[0036] Legal relations: rights and obligations, contractual relations, inheritance relations, etc. between the parties;

[0037] Evidence characteristics: type, content, probative value, etc. of relevant evidence;

[0038] Of course, other definitions may be used depending on the particularity of the case, and are not limited here.

[0039] According to the above-mentioned characteristic dimensions, information is extracted from the case materials of the notarization cases to be checked, and labels of the characteristic dimensions are added to generate structured data based on the notarization cases to be checked.

[0040] SP12. Extract features from structured data to obtain feature vectors of case materials;

[0041] Specifically, first, extract feature vectors related to the case from structured data. For example: notarization matters, case types, legal basis, factual features, legal relations and evidence features. Therefore, it is necessary to extract information features of notarization matters, case types, legal basis, factual features, legal relations and evidence features. The feature extraction method can be through keyword extraction and extraction using a semantic understanding model. Specifically, on the one hand, if a method such as keyword extraction is adopted, the word frequency-inverse document frequency method can be used to perform text statistics, find out the key information in the text, and then vectorize the key information. The vectorization method, for example, can be vectorized using the Word2Vec model; on the other hand, a method for information extraction using a semantic understanding model can also be used. This method is mainly used when the above-mentioned feature dimension contains a large amount of text content. In this case, a semantic understanding model can be used to obtain a summary of the text content in the feature dimension. The semantic understanding model can be a BERT model or a Bi-LSTM model. When a BERT model or a Bi-LSTM model is used, the normalization layer can be removed, and the result of the embedding layer can be directly saved, so that the feature vector of the summary can be directly extracted.

[0042] Then the basic information is feature extracted and vectorized. Some data can be directly extracted, such as ID card number, and some data can be extracted by coding, such as gender, which can be set to 0 for male and 1 for female.

[0043] One possible implementation method is: the notarization matters and / or case types determine the notarization type of the notarization case to be checked, and based on the notarization type, similar case matching is performed under the notarization type case branch corresponding to the notarization case database. A subtable can be established in advance for the notarization case database according to the notarization type. When similar case matching is required, if the notarization type of the notarization case to be checked has been determined by the notarization matters and / or case types, that is, the matching features include the notarization type feature dimension, there is no need to perform global matching, and only needs to match in the subtable of the determined notarization type.

[0044] The feature vectors related to the above case and the feature vectors of basic information are grouped together to obtain the feature vectors of the case materials.

[0045] SP13. Extract features from the first user input data based on a natural language processing method to obtain a feature vector of the first user input data;

[0046] As mentioned above, there are many application scenarios for similar case retrieval. Therefore, for different application scenarios, even if the case materials to be checked are the same, the target similar cases (i.e. target notarization cases) to be obtained are also different. In order to improve the quality of similar case matching tasks, it is necessary to fully consider the characteristics of the first user's input data in the matching features of similar case matching to correctly understand the similar case matching requirements.

[0047] Typically, the first user input data is a piece of text data. Therefore, the natural language processing method can be used to extract features from the first user input data. Natural language processing methods (i.e., NLP tasks) have been developed in the field of artificial intelligence to date, and there are various feature extraction methods for semantic understanding. One possible implementation method is to use the Transformer model for semantic understanding. The specific approach is: use the transformer model to extract a summary of the first user input data, so as to determine the key meaning to be expressed by the first user input data. The obtained summary is typed and vectorized to obtain a feature vector of the first user input data. Of course, there are other applicable methods of semantic understanding, which are not limited here.

[0048] SP14. Collect the feature vectors of the case materials and the feature vectors of the first user input data to obtain matching features of the notarization case to be checked.

[0049] The feature vector of the case material and the feature vector of the first user input data are aggregated to obtain matching features of the notarization case to be checked;

[0050] In order to deal with the situation where the first user inputs inaccurate data and to query related similar cases, another possible implementation is that, based on the above steps, the method for generating matching features further includes the following steps:

[0051] SP15. Extract entities and relationships from the first user input data to obtain entities and relationships from the first user input data.

[0052] SP16. Based on the entity, search for a corresponding or related entity set in the pre-generated notarization knowledge graph. The notarization knowledge graph is a visualization of the notarization field entity relationship graph generated by extracting entity relationships from historical notarization archives and based on Neo4j in the storage database. Entity queries can be performed using the database's query language cypher to find entities that correspond to or are related to the entity input by the first user. The so-called corresponding refers to another entity that has a corresponding relationship with the entity to be checked, for example, the entity to be checked is the husband and the target query entity is the wife; the so-called related refers to another entity that has a similar title to the entity to be checked.

[0053] SP17. Based on the relationship, search for the relationship set corresponding to or related to the relationship on the same entity in the notarized knowledge graph. In the entity set, based on each entity, query the relationship corresponding to or related to the relationship to be queried, and integrate it into the relationship set.

[0054] SP18. Based on the entity set and the relationship set, the second entity relationship combination closest to the entity relationship of the first user input data is selected through similarity calculation and sorting, and the entities and relationships in the first user input data are replaced to generate the second user input data.

[0055] SP19. Perform feature extraction on the second user input data based on a natural language processing method to obtain a feature vector of the second user input data.

[0056] The natural language processing method here may adopt the same or similar processing method as the first user input data, which will not be described in detail here.

[0057] SP20. Collect the feature vectors of the case materials, the feature vectors of the data input by the first user, and the feature vectors of the data input by the second user to obtain matching features of the notarization case to be checked.

[0058] That is, by extracting entities and relationships from the first user input data, triple data of the first user input data is obtained; the triple data is queried in the pre-generated notarization knowledge graph using Cypher query statements; thereby expanding the scope of interpretation of the first user input data. That is, through the entities in the triples, corresponding or related entities in the notarization knowledge graph are searched to form an entity set; based on the relationship, for the same entity in the entity set, corresponding or related relationships in the notarization knowledge graph are searched to form a relationship set. And the entities in the entity set are combined with the relationships in the relationship set to form a new entity relationship group. By calculating and sorting the similarity of the new entity relationship group and the triples of the original entity relationship, a new entity relationship closest to the entity relationship of the first user input data is obtained, and the new entity relationship replaces the entity relationship of the first user input data in the first user input data, thereby generating the second user input data, and the second user input data is used as a supplement to the first user input data, and feature extraction is performed, and the data is aggregated into the original matching features to participate in the matching calculation, so as to cope with the scenarios where the meaning of the first user input data is unclear, cannot be checked, or needs to be queried for related cases.

[0059] SP2. Based on the matching features, calculate the similar notarization cases in the notarization case database that best match the notarization case to be checked.

[0060] Specifically, based on the matching features, the notarization case database is calculated to find the similar notarization case that best matches the notarization case to be checked. Figure 5 , a possible implementation method includes the following steps:

[0061] SP21. Based on each matching feature, calculate the matching degree with each case in the notarization case database.

[0062] Specifically, first, the matching degree algorithm model is used to calculate the matching degree between the feature vector of each matching feature and the feature vector of the case in the notarization case database.

[0063] SP22. The cases that have reached the preset matching threshold are grouped into a candidate case set; the BERT model is used as the matching algorithm model.

[0064] Specifically, a matching degree threshold is preset, and the cases whose matching degree calculated in step SP21 reaches the threshold are grouped into a candidate case set. Here, each matching feature is used as a matching condition, and at least one notarization case whose matching feature reaches the matching degree threshold is screened out from the notarization case database, and the cases that reach the preset matching degree threshold are grouped into a candidate case set. This means that after the first round of screening, notarization cases that do not have any features that meet the matching conditions are deleted, thereby greatly reducing the amount of calculation in the subsequent comprehensive matching degree matching.

[0065] Among them, the BERT model is selected as the matching algorithm model; the BERT model is a Transformer-based neural network model that can be effectively used to calculate text matching.

[0066] SP23. From the candidate case set, select the similar notarization case that best matches the notarization case to be checked.

[0067] Please refer to Figure 6 , wherein a possible implementation method includes the following steps:

[0068] SP231. Based on the source of the matching features, determine the weight corresponding to each matching feature.

[0069] Since the contribution of matching features to the judgment of similar case matching is different, it is necessary to configure the weight of the importance of each matching feature; one weight configuration method may be: presetting the weight of each feature dimension; querying the weight of the feature dimension corresponding to the matching feature, and using it as the weight of the corresponding matching feature. Generally speaking, feature dimensions related to the case are more important, such as factual features and legal basis. Therefore, it is necessary to configure higher weights for these feature dimensions. In addition, in order not to deviate from the user's similar case query intention, it is necessary to set the weight of the matching feature derived from the first user input data to be higher than the weight of the matching feature derived from the case material, thereby guiding the similar case matching results in the direction of the user's real needs.

[0070] For the notarization heat factor, the weight of the notarization heat factor can be pre-set and shared among all the notarization heat factors. That is, when the notarization heat factor participates in the second comprehensive matching calculation, no matter which case the matching task is, the weight value of the notarization heat factor is taken from the weight pre-configured for the notarization heat factor.

[0071] SP232. Perform weighted calculation on the similarity of each matching feature and the corresponding weight to obtain the standard deviation, and obtain the first comprehensive matching degree between the notarization case to be checked and the candidate notarization cases; sort based on the first comprehensive matching degree, and select the first K candidate notarization cases with the highest first comprehensive matching degree to generate a candidate case set, 5≤K≤20;

[0072] In order to comprehensively consider multiple matching features and obtain the notarization case to be checked with the notarization case with the highest first comprehensive matching degree in the case database, a possible calculation method is to use weighted summation to calculate the standard deviation to obtain the first comprehensive matching degree. For example, if the final matching features are 7 (including the features extracted from the first user input data), the reference formula is as follows:

[0073] S=(w1×s1+w2×s2+w3×s3+w4×s4+w5×s5+w6×s6+w7×s7) / 7

[0074] Among them, S represents the first comprehensive matching degree; s1 to s7 represent matching features; w1 to w7 represent the weights of the corresponding matching features.

[0075] Of course, depending on the case and the need for matching similar cases, the matching features may be less than 7.

[0076] Furthermore, when determining the weight corresponding to each matching feature, the first user input data needs to be considered, and the user's real similar case retrieval intention is taken into consideration. Therefore, another possible calculation method is: using a pre-trained intention recognition model to analyze the first user input data, predict the user's intention to query similar cases, and obtain the predicted user intention; based on the user intention, adaptively adjust the weight for each matching feature. Among them, the intention recognition model can be obtained by pre-training the ChatGLM-6B model. By inputting the first user input data into the intention recognition model, the user's intention is understood and classified; thereby the category of user intention is obtained; according to the category of user intention, each matching feature is weighted according to the preset rules, and the weight is adjusted according to the preset rules; the preset rules can be set according to different user intentions. The weight range corresponding to the feature dimension of each matching feature is set, and when the comparison result exceeds the weight range, the weight of the matching feature is adjusted to fall within the weight range according to the weight range.

[0077] The behavior preference is obtained from the user's historical similar case search behavior, and the behavior preference is mapped into a notarization heat factor, and is associated and stored with the case association in the notarization case database; in order to also include the user's similar case search behavior preference in the comprehensive matching degree consideration, the notarization heat factor of each candidate notarization case in the candidate case set can be extracted from the notarization case database and added to the calculation of the comprehensive matching degree. Therefore, another possible implementation method is to further include the following steps:

[0078] SP233. Query the notarization heat factor of each candidate notarization case in the candidate case set, perform weighted operation on the matching features of the candidate notarization case and the corresponding notarization heat factor to obtain the standard deviation, obtain the second comprehensive matching degree, and re-sort based on the second comprehensive matching degree; select the candidate notarization case with the highest second comprehensive matching degree as the similar notarization case of the notarization case to be checked.

[0079] Specifically, after sorting the candidate notarial cases from large to small based on the first comprehensive matching degree and screening out the set of candidate notarial cases, the notarial heat factor of each candidate notarial case is queried and read, and the matching features of the candidate notarial case are weighted together with the notarial heat factor, and the standard deviation of the result is taken to obtain the second comprehensive matching degree of each candidate notarial case, and the case is sorted based on the second comprehensive matching degree, and the candidate notarial case with the highest second comprehensive matching degree is selected as the hit similar notarial case.

[0080] SP3. Send similar notarization cases to the requesting end.

[0081] SP4. After sending similar notarization cases to the requesting end, it also includes: obtaining user feedback information on matching cases; inputting the feedback information into the pre-trained intent recognition model for analysis; judging whether it is necessary to readjust the weight of each matching feature based on the analysis results; if so, after the adjustment, recalculating the comprehensive matching degree based on the adjustment results.

[0082] Specifically, in order to better serve the user's similar case query needs, after the first round of similar case matching, there may be unsatisfactory similar case matching. Therefore, user feedback on similar case matching results can be collected in the user interface, and improvement operations can be performed based on the feedback information.

[0083] One possible implementation is to set options in the user interface for users to select their satisfaction levels, thereby knowing whether the matched notarization cases meet the user's real similar case query needs; if complex situations arise, a feedback information input window can also be provided for users to collect feedback information for supplementing or correcting the user's input; the feedback information input window here can be reused with the user's input window to simplify the user operation interface.

[0084] After that, user feedback is collected and analyzed. If the user provides feedback by inputting text, the feedback needs to be sent to the intent recognition model for intent recognition, so as to determine whether a new round of comprehensive matching degree calculation is needed, and then a new round of similar case matching is achieved. If so, the similar case matching task is re-executed.

[0085] In summary, the embodiment of the present application obtains matching features for similar case matching from the case materials and the first user input data respectively, and uses the matching features to screen similar cases in the notarization case database, thereby matching similar cases based on the needs and the extracted matching features through dual guidance, and obtaining similar notarization cases that are closer to user needs.

[0086] In addition, feature dimensions are preset, and case materials are feature extracted and vectorized according to the preset feature dimensions; in addition, feature extraction is performed on the first user input data through natural language processing methods; and the two are aggregated to obtain matching features; compared with the prior art method of matching similar cases only by entering keywords, multi-dimensional matching conditions are provided, so that notarization cases with more reference value can be obtained.

[0087] In addition, based on each matching feature, the matching degree of each case in the notarization case database is calculated; and the cases that reach the preset matching degree threshold are grouped into the candidate case set. In the candidate case set, the most matching notarization case is screened out by calculating the comprehensive matching degree between the notarization case to be checked and the candidate case set. On the one hand, the accuracy of similar case screening can be improved by using the comprehensive matching calculation; on the other hand, by first performing a one-dimensional matching calculation, completely irrelevant cases are screened out, and then performing a comprehensive matching calculation, the amount of comprehensive matching calculation can be significantly reduced, thereby reducing the loss of computing resources and improving computing efficiency.

[0088] In addition, by presetting the weight of each feature dimension; sharing the weight between matching features belonging to the same feature dimension; and setting the weight of the matching feature derived from the first user input data to be higher than the weight of the matching feature derived from the case material, this allows the present solution to not only consider multi-dimensional features, but also differentiate the importance of each matching feature, while keeping the importance of matching features of the same feature dimension consistent, thereby more reasonably configuring the weight for each matching feature, highlighting the search focus of user needs while also taking into account the individual characteristics of the notarization case to be checked.

[0089] In addition, in order to take the user's behavioral preferences into consideration, the notarization heat factor can also be added to the calculation of the comprehensive matching degree, so as to better meet the user's search needs.

[0090] In addition, the pre-trained intent recognition model is used to analyze the first user input data, analyze the user's intention to query similar cases, and obtain the predicted user intent; based on the user intent, the weight is adaptively adjusted for each matching feature; thereby adaptively adjusting the matching calculation scheme, accurately providing reference notarization cases for user needs in different scenarios, so as to better serve users.

[0091] In addition, in order to prevent the first user input data entered by the user from being biased, the entity relationships in the first user input data are generalized and interpreted with the help of the notarized knowledge graph, that is, the corresponding or related entities and relationships in the notarized knowledge graph are searched through the entities and relationships of the first user input data, and new entity relationships that are most similar to the entity relationships of the first user input data are selected to replace the original entities and relationships in the first user input data to generate the second user input data, and the second user input data is used as a supplement to the first user input data, thereby expanding the interpretation scope of the first user input data and performing similar case matching more broadly and comprehensively.

[0092] In addition, by obtaining user feedback on matching cases, the feedback information is input into the pre-trained intent recognition model for analysis, and based on the analysis results, it is determined whether the weight of each matching feature needs to be readjusted; and similar case matching is re-performed based on the adjustment results, thereby reusing the intent recognition model to perform intelligent analysis of user feedback information, self-learn, and continuously improve the service quality of similar case matching.

[0093] See also Figure 7 In another exemplary embodiment of the present application, an electronic device 4000 is provided. The electronic device 4000 may be a computer, a server, a virtual machine, etc. Figure 7 In the embodiment, the electronic device 4000 includes at least one processor 4001, at least one communication bus 4002 and at least one memory 4003. The processor 4001 and the memory 4003 are connected, such as through the communication bus 4002. Optionally, the electronic device 4000 may also include a transceiver 4004, which can be used for data interaction between the electronic device and other electronic devices, such as data transmission and / or data reception. A computer program is stored in the memory 4003, and the processor 4001 reads the computer program in the memory 4003 through the communication bus; when the computer program is executed by the processor 4001, a notarization case intelligent matching method in the above-mentioned embodiment is implemented. It should be noted that in actual applications, the transceiver 4004 is not limited to one, and the structure of the electronic device 4000 does not constitute a limitation on the embodiments of the present application.

[0094] In addition, another exemplary embodiment of the present application provides a storage medium on which a computer program is stored. When the computer program is executed by a processor, a method for intelligent matching of notarization cases in the above embodiment is implemented.

[0095] It should be understood that, although the steps in the flowchart of the accompanying drawings are displayed in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0096] The above description is only a partial implementation method of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A notarization case intelligent matching method, It is characterized in that include: Acquire matching features of the notarization case to be checked, where the matching features are at least derived from pre-collected case materials of the notarization case to be checked and data input by the first user; Based on the matching features, calculating the similar notarization case in the notarization case database that best matches the notarization case to be checked; The notarization case of the same type is sent to the requesting end.

2. The notarization case intelligent matching method according to claim 1, It is characterized in that The method for generating the matching feature comprises: Extracting information from the case materials based on a plurality of preset feature dimensions to generate structured data; wherein the feature dimensions include: basic information, notarization matters, case type, legal basis, factual characteristics, legal relationship and evidence characteristics; Performing feature extraction on the structured data to obtain a feature vector of the case material; Performing feature extraction on the first user input data based on a natural language processing method to obtain a feature vector of the first user input data; The feature vector of the case material and the feature vector of the first user input data are aggregated to obtain matching features of the notarization case to be checked.

3. The notarization case intelligent matching method according to claim 2, It is characterized in that The method for generating the matching feature further includes: Extracting entities and relationships from the first user input data to obtain entities and relationships of the first user input data; Based on the entity, searching for a corresponding or related entity set in a pre-generated notarization knowledge graph; and, Based on the relationship, searching the notarized knowledge graph for a relationship set corresponding to or related to the relationship on the same entity in the entity set; Based on the entity set and relationship set, a second entity-relationship combination of entities and relationships closest to the first user input data is selected by sorting after similarity calculation, and the entities and relationships in the first user input data are replaced to generate second user input data; Performing feature extraction on the second user input data based on a natural language processing method to obtain a feature vector of the second user input data; The feature vectors of the case materials, the feature vectors of the first user input data, and the feature vectors of the second user input data are aggregated to obtain matching features of the notarization case to be checked.

4. The notarization case intelligent matching method according to claim 3, Features: The calculating, based on the matching features, of similar notarization cases in a notarization case database that best matches the notarization case to be checked specifically includes: Based on each of the matching features, the similarity with each case in the notarization case database is calculated; and the cases that reach the preset similarity threshold are grouped into a candidate case set; wherein the similarity algorithm model uses the BERT model; From the candidate case set, select a similar notarization case that best matches the notarization case to be checked.

5. The notarization case intelligent matching method according to claim 4, Features: The selecting of the similar notarization case that best matches the notarization case to be checked from the candidate case set specifically includes: Based on the source of the matching feature, determine the weight corresponding to each of the matching features; Performing a weighted operation on the similarity of each matching feature and the corresponding weight and taking the standard deviation, to obtain a first comprehensive matching degree between the notarization case to be checked and the candidate notarization cases; sorting based on the first comprehensive matching degree, selecting the first K candidate notarization cases with the highest first comprehensive matching degree to generate a candidate case set, 5≤K≤20; querying the notarization heat factor of each candidate notarization case in the candidate case set, performing a weighted operation on the matching features of the candidate notarization cases and the corresponding notarization heat factor to obtain a standard deviation, obtaining a second comprehensive matching degree, and re-ranking based on the second comprehensive matching degree; The candidate notarization case with the second highest comprehensive matching degree is selected as a similar notarization case of the notarization case to be checked.

6. The notarization case intelligent matching method according to claim 5, Features: Determining the weight corresponding to each matching feature specifically includes: Preset the weight of each feature dimension; query the weight of the feature dimension corresponding to the matching feature, and use it as the weight of the corresponding matching feature; and, preset the weight of the notarization heat factor, and share the weight of the notarization heat factor among all notarization heat factors, and, The weight of the matching feature derived from the first user input data or the second user input data is set higher than the weight of the matching feature derived from the case material.

7. The notarization case intelligent matching method according to claim 6, Features: The determining of the weight corresponding to each matching feature further includes: Analyze the first user input data using a pre-trained intent recognition model to obtain a predicted user intent; Based on the user intent, a weight is adaptively adjusted for each matching feature.

8. The notarization case intelligent matching method according to claim 1, Features: When the notarization case of the same type is sent to the requesting end, it also includes: Get feedback from users; The feedback information is input into the pre-trained intent recognition model for analysis; based on the analysis result, it is determined whether the weight of each matching feature needs to be readjusted; if so, after the adjustment, the comprehensive matching degree is recalculated based on the adjustment result.

9. An electronic device, It is characterized in that include: At least one processor, at least one memory, and at least one communication bus, wherein the memory stores a computer program, and the processor reads the computer program in the memory through the communication bus; when the computer program is executed by the processor, a notarization case intelligent matching method as described in any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium, It is characterized in that include: A computer program is stored thereon, and the computer program is executed by a computer processor as a method for intelligent matching of notarial cases as described in any one of claims 1 to 8.

Images