Processing method and system for improving recognition processing efficiency based on key point automatic grouping

Abstract

The invention relates to a processing method and system for improving recognition processing efficiency based on key point automatic grouping, and belongs to the technical field of content recognition process.The method comprises the steps that a processing instruction proposed by a user is received, and all files contained in the processing instruction are analyzed to obtain an examination key point list and a corresponding grouping thinking chain; inputting all the review key points in the review key point list and the corresponding grouping thinking chain into a preset grouping agent, and classifying all the review key points through the grouping agent to form a plurality of first-level review groups; through an asynchronous calling framework, a review model is allocated to each first-level review group, a plurality of review threads are generated and started, and the review threads correspond to the first-level review groups one to one so as to be used for processing all review key points contained in the corresponding first-level review groups; and obtaining, merging and outputting review results of all the review threads, so that a user can obtain an output result. The method and the device have the effect of improving the multi-file processing efficiency.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of content recognition processing, in particular to a processing method and system for improving recognition processing efficiency based on automatic grouping of key points. BACKGROUND

[0002] With the development of informatization, the number of files that enterprises and individuals need to process and review increases exponentially, especially in the fields of law, finance, scientific research and government affairs, the demand for multi-file review is increasingly prominent.

[0003] The traditional file processing and review method mainly relies on manual reading and comparison of file content. When facing a large number of file materials and / or a large number of review points of file materials, the above-mentioned manual processing method is not only low in efficiency, but also easy to affect the accuracy of review processing, so it needs to be improved. SUMMARY

[0004] In order to improve the efficiency and accuracy of review processing, the present application provides a processing method and system for improving recognition processing efficiency based on automatic grouping of key points.

[0005] In a first aspect, the present application provides a processing method for improving recognition processing efficiency based on automatic grouping of key points, which adopts the following technical solution: Receiving a processing instruction proposed by a user, analyzing all files contained in the processing instruction to obtain a review key point list and a corresponding grouping thought chain; Inputting all review key points in the review key point list and the corresponding grouping thought chain into a preset grouping intelligent agent, classifying all review key points through the grouping intelligent agent to form a plurality of first review groups; Assigning a review model to each first review group through an asynchronous calling framework, generating and starting a plurality of review threads, and the review threads correspond one-to-one to the first review groups, for processing all review key points contained in the corresponding first review group; Obtaining and merging the review results of all review threads for output, so that the user can know the output results.

[0006] By adopting the above technical solution, all files are first analyzed, and all review key points required for processing the aforementioned all files (i.e. review key points contained in the review key point list) are determined. When there are multiple review key points, dynamic hierarchical grouping of review tasks can be realized in combination with the grouping thought chain. Finally, by asynchronously calling adaptive models through multiple threads, efficient parallel review is realized, which replaces manual work and improves review efficiency.

[0007] Optionally, the review key points at least include entities to be reviewed. The all review points in the review point list and the corresponding grouping thought chain are input into a preset grouping agent, all the review points are classified into a plurality of first review groups by the grouping agent, and the first review groups are formed. The all review points in the review point list are subjected to entity alignment processing, the review points subjected to the entity alignment processing and the corresponding grouping thought chain are input into a preset grouping agent, all the review points are classified into a plurality of first review groups by the grouping agent, and the first review groups are formed, so that different entities correspond to different first review groups, and a first mapping relationship between the entities and the first review groups is generated. Based on a preset review type, the review points in each first review group are classified to generate second review groups, and the review type at least includes a pure text review type and a visual image review type. The review model is allocated to each first review group by using an asynchronous calling framework, a plurality of review threads are generated and started, and the review threads correspond to the first review groups one by one. The review model is matched with each second review group, a plurality of review threads are generated and started, and the review threads correspond to the second review groups one by one.

[0008] By using the above technical solution, the preliminary grouping of the review points (i.e., the first review groups are formed) is realized based on the entity, and then the first review groups are further classified according to the different review types, so that the special review model is matched, and the review efficiency is improved.

[0009] Optionally, the all review points in the review point list and the corresponding grouping thought chain are input into a preset grouping agent, all the review points are classified into a plurality of first review groups by the grouping agent, and the first review groups are formed, and the method further includes the following steps. The review points included in each second review group are analyzed, a file list corresponding to each second review group is determined, the file list includes files called to execute the review points included in the corresponding second review group, and a second mapping relationship between the review points in the second review group and the files in the corresponding file list is established. According to the second mapping relationship, all the review points corresponding to the same file in each second review group are taken as associated review points, and third review groups are formed by merging. An association relationship is established between the third review groups corresponding to different second review groups and the same file. The review model is matched with each second review group, a plurality of review threads are generated and started, and the review threads correspond to the second review groups one by one. generating an examination thread for each of the secondary examination groups, respectively, splitting the examination thread corresponding to each of the secondary examination groups into a plurality of sub-threads corresponding to each of the tertiary examination groups according to the tertiary examination groups included in each of the secondary examination groups, and establishing an association relationship between the corresponding sub-threads according to the association relationship between the tertiary examination groups; allocating an examination model and an examination engine for each of the sub-threads through the asynchronous calling framework, controlling the corresponding sub-threads to be executed by the corresponding examination model in a preset examination order through the examination engine until the corresponding examination thread is completed, and triggering the execution of the sub-threads having an association relationship with the target sub-thread when the target sub-thread is executed, wherein the target sub-thread is any sub-thread.

[0010] According to the above technical solution, after the grouping of the secondary examination groups is completed, the secondary examination groups are further classified into tertiary examination groups according to the types of files to be called when the examination points included in each of the secondary examination groups are executed, that is, the examination points included in the tertiary examination groups are for the examination of the same file. Accordingly, the examination thread corresponding to each of the secondary examination groups is split into a plurality of sub-threads, the sub-threads correspond to the files one by one, and all the sub-threads corresponding to the same file are bound (i.e., an association relationship is established). When any sub-thread is executed and the corresponding file is loaded, all the sub-threads having a corresponding relationship with the file are executed in parallel. In this way, the sub-threads across groups are bound and executed in synchronization through file association, thereby maximizing the utilization rate of single file loading, achieving the effect of single file loading-multiple examination model parallel examination, dynamically balancing file I / O and computing resources, and avoiding resource contention caused by too high parallelism.

[0011] Optionally, the matching examination model for each of the secondary examination groups, generating and starting a plurality of examination threads, and the examination threads correspond to the secondary examination groups one by one, further comprises: grouping the sub-threads having an association relationship to form a sub-thread list, analyzing the examination points included in each of the sub-threads in the same sub-thread list, determining the dependency relationship between all the sub-threads in each of the sub-thread lists based on the analysis result, determining the execution mode of all the sub-threads in each of the sub-thread lists according to the dependency relationship, and determining the execution order for the sub-threads with sequential execution mode; wherein the execution mode at least includes parallel execution or sequential execution; generating a scheduling scheme for each of the sub-thread lists according to the execution mode and the execution order; the triggering the execution of the sub-threads having an association relationship with the target sub-thread when the target sub-thread is executed, comprises: When the target sub-thread is executed, a sub-thread having an association relationship with the target sub-thread is triggered to be executed according to a scheduling scheme corresponding to the target sub-thread.

[0012] By adopting the technical scheme, in the process of realizing the collaborative review of the sub-threads across groups, the execution order dependency relationship of all the sub-threads in the same sub-thread list is dynamically judged, and a scheduling scheme is generated based on the dependency relationship, so that the execution order of the sub-threads is coordinated, the efficiency and correctness are taken into account, and conflicts are avoided.

[0013] Optionally, the method further comprises: analyzing specific review contents of review points contained in all the sub-threads in the same sub-thread list, and judging the similarity of specific review contents of review points contained in different sub-thread lists; establishing a reuse relationship between review points in different sub-thread lists and satisfying a preset requirement, and determining the review points as reused review points, and merging the reused review points having the reuse relationship to form a reused review point set; analyzing the different review contents and the similar contents between the reused review points having the reuse relationship in each reused review point set and belonging to different sub-thread lists, wherein the different review contents at least include an entity to be reviewed, and the similar contents are the same review contents in the specific review contents corresponding to the reused review points having the reuse relationship; generating a reuse template for each reused review point set, and the reuse template includes the similar contents and the different review contents, and a reuse model for executing the reuse template is constructed; The method further comprises: If the target sub-thread contains a reused review point, a corresponding reuse model is called, the reused review point is executed by the reuse model according to the different review contents of the target sub-thread, and all the review points of the non-reused review points contained in the target sub-thread are executed by the review model corresponding to the target sub-thread.

[0014] By adopting the technical scheme, the review points having the similar review contents (i.e., the reused review points) are extracted from the sub-threads and the review points contained in other sub-threads, a review template is generated for the reused review points, a reuse model is allocated to the reused review points, the common logic (i.e., the similar contents) in the reuse template is separated from the different review contents, the reused review points are executed by the reuse model, and during the execution, only the different review contents need to be switched to realize the review of all the review points having the reuse relationship.

[0015] Optionally, the method further comprises: The recording multiplexing model outputs a natural language explanation report according to a review path of a review point of multiplexing, wherein the review path at least includes multiplexing template content, corresponding differentiated review content, and location information of the differentiated review content in a corresponding file.

[0016] By adopting the technical solution, the complete path of recording multiplexing review is recorded, and an explanation report is output for a user to know, thereby solving the problems of lack of interpretability of multiplexing review results and difficulty in manual review.

[0017] Optionally, the method further includes: receiving a correction instruction proposed by a user, and verifying the identity of the user; wherein the correction instruction at least includes a multiplexing template to be corrected and correction content for correcting corresponding differentiated review content; when the identity verification of the user is passed, correcting the multiplexing template included in the correction instruction based on the correction instruction.

[0018] By adopting the technical solution, a manual intervention review channel is provided to help correct the differentiated review content in the multiplexing template, optimize the multiplexing template, and ensure the execution accuracy of the multiplexing model.

[0019] In a second aspect, the present application provides a processing system for improving identification processing efficiency based on automatic grouping of points, including, a file identification and analysis module configured to receive a processing instruction proposed by a user, and analyze all files included in the processing instruction to obtain a review point list and corresponding grouping thought chain; a review point classification module configured to input all review points in the review point list and the corresponding grouping thought chain into a preset grouping intelligent agent, and classify all review points through the grouping intelligent agent to form a plurality of first-level review groups; a multi-thread synchronization enabling module configured to allocate a review model to each of the first-level review groups through an asynchronous calling framework, generate and start a plurality of review threads, and the review threads correspond to the first-level review groups one by one, so as to process all review points included in the corresponding first-level review groups; a processing result output module configured to obtain and merge the review results of all review threads, and output the obtained results for a user to know.

[0020] In a third aspect, the present application provides a processing device for improving identification processing efficiency based on automatic grouping of points, including a memory and a processor, and the memory stores a computer program capable of being loaded and executed by the processor to execute the method of any one of the first aspect.

[0021] In a fourth aspect, the present application provides a computer readable storage medium storing a computer program capable of being loaded and executed by a processor to perform the method according to any one of the first aspect.

[0022] In summary, the present application includes at least one of the following beneficial technical effects: 1. The present application aims to solve the problem of low efficiency in the prior art, specifically by aligning the entity expressions of the examination points through entity alignment technology, dynamically layering and grouping the examination tasks by combining grouping thought chains, and finally achieving efficient parallel examination through asynchronous multi-threaded calling of adaptive models. 2. Further, the present application classifies the examination points based on entities to achieve primary grouping of the examination points (i.e., to form a first examination group), and further classifies the first examination group according to different examination types to match special examination models and improve examination efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0024] Figure 1 is a flowchart of a processing method for improving identification processing efficiency based on automatic grouping of points disclosed by the embodiments of the present application.

[0025] Figure 2 is a structural block diagram of a processing system for improving identification processing efficiency based on automatic grouping of points disclosed by the embodiments of the present application.

[0026] Legend of reference numerals: 201, file identification and analysis module; 202, examination point classification module; 203, multi-thread synchronization enabling module; 204, processing result output module. DETAILED DESCRIPTION

[0027] The following will be described in detail in combination with the drawings Figures 1-2 The present application will be further described in detail.

[0028] The embodiments of the present application disclose a processing method for improving identification processing efficiency based on automatic grouping of points (hereinafter referred to as processing method), which is applicable to the automatic examination processing scene of multiple files. The execution subject of the processing method is a processing system for improving identification processing efficiency based on automatic grouping of points (hereinafter referred to as processing system), which will be described in combination with the drawings Figure 1 The specific flow steps of the processing system executing the processing method will be described in detail.

[0029] S101, receiving a processing instruction proposed by a user, and parsing all files contained in the processing instruction to obtain an examination point list and a corresponding grouping thinking chain.

[0030] In implementation, a user can access a processing system in the form of a webpage, and upload files through a preset upload button on the display interface of the processing system, the files can be multiple items and multiple types, such as pure text type, image type, etc., so as to trigger a processing instruction. The processing system will parse the files after receiving all the files contained in the processing instruction, and the parsing operation specifically includes: using an open source library (such as Apache PDFBox, pdfplumber / docx library of Python) to extract text and tables in PDF and Word formats, identifying files in images and scans through OCR technology (such as Tesseract, Aliyun OCR API), and extracting entities (such as names, addresses, dates, etc.) from the foregoing extracted content through a pre-trained model (such as SpaCy, Aliyun NLP).

[0031] Then, the processing system is used to determine the specific processing requirement of the user triggering the processing instruction according to the parsing result, and correspondingly, the processing requirement can be directly input by the user when triggering the processing instruction, or can be automatically parsed by the system based on the pre-stored processing requirements and the corresponding relationship between the files to obtain the processing requirement, such as the processing requirement of opening a store, which corresponds to the required files including “business license”, “rental contract”, “health permit” and the like.

[0032] In addition, the processing system also pre-stores a plurality of processing requirements and an examination point list corresponding to each processing requirement, and the examination point list includes a plurality of examination points, which can be considered as specific examination contents of specific entities in the corresponding files, such as checking whether entity B in file A meets the requirements; as can be seen, the examination points correspond to the entities, therefore, the examination points at least include the entities to be examined. The grouping thinking chain is a pre-set grouping basis for classifying all examination points in the examination point list to form a plurality of first examination groups, such as COT thinking chain.

[0033] Specifically, the processing method further includes the following steps: S102, inputting all examination points in the examination point list and the corresponding grouping thinking chain into a pre-set grouping intelligent agent, and classifying all examination points through the grouping intelligent agent to form a plurality of first examination groups; S103, respectively assigning an examination model to each first examination group through an asynchronous calling framework, generating and starting a plurality of examination threads, and the examination threads correspond one-to-one to the first examination groups, so as to process all examination points contained in the corresponding first examination groups.

[0034] The S102 and S103 specifically include the following sub-steps: All the review points in the review point list are subjected to entity alignment processing, and the review points subjected to entity alignment processing and the corresponding grouping thinking chain are input into a preset grouping intelligent agent, all the review points are classified by the grouping intelligent agent to form a plurality of first-level review groups, so that different entities correspond to different first-level review groups, and a first mapping relationship between the entities and the first-level review groups is generated; Based on the preset review type, the review points in each first-level review group are classified to generate second-level review groups; wherein the review type at least includes a pure text review type, a visual image review type; The review points included in each second-level review group are analyzed to determine the file list corresponding to each second-level review group, and the file list includes the files called to execute the review points included in the corresponding second-level review group; a second mapping relationship between the review points in the second-level review group and the files in the corresponding file list is established; According to the second mapping relationship, all the review points corresponding to the same file in each second-level review group are taken as associated review points, and are merged to form third-level review groups; An association relationship is established between the third-level review groups belonging to different second-level review groups and corresponding to the same file; An review thread is generated for each second-level review group, and according to the third-level review groups included in each second-level review group, the review thread corresponding to each second-level review group is split into sub-threads corresponding to each third-level review group one by one, and according to the association relationship between the third-level review groups, an association relationship is established between the corresponding sub-threads; The sub-threads with the association relationship are classified to form a sub-thread list, the review points included in each sub-thread in the same sub-thread list are analyzed, based on the analysis result, the dependency relationship between all the sub-threads in each sub-thread list is determined, the execution mode of all the sub-threads in each sub-thread list is determined according to the dependency relationship, and the execution order of the sub-threads with the sequential execution mode is determined; wherein the execution mode at least includes parallel execution or sequential execution, A scheduling scheme is generated for each sub-thread list according to the execution mode and the execution order; Through an asynchronous calling framework, a review model and a review engine are respectively allocated to each sub-thread, the corresponding sub-thread is executed by the review engine according to the preset review order, until the corresponding review thread is completed; and when the target sub-thread is executed, the sub-threads having the association relationship with the target sub-thread are triggered to be executed according to the scheduling scheme corresponding to the target sub-thread.

[0035] In implementation, based on existing entity alignment technology, all review points in the review point list are subjected to entity alignment processing, and then all review points in the review point list are classified to form a plurality of first review groups based on the grouping agent based on the grouping thought chain. In the embodiment of the application, the grouping logic of the grouping thought chain is: taking entity as the grouping basis, the review points containing the same entity to be reviewed are attributed to the same first review group to realize one-to-one correspondence between entity and first review group. The first mapping relationship and the first mapping relationship are used to store the entity name and the corresponding first review group ID.

[0036] Further, the processing system is configured to determine the review type of each review point according to the specific review content of each review point, the file type corresponding to each review point, and the entity name. The review type is a pure text review type and a visual image review type. For example, if the file type corresponding to the review point is a pure text format, the corresponding review type is considered to be a pure text review type. If the file type corresponding to the review point is an image form or the corresponding entity is an image in the file, the corresponding review type is a visual image review type. The processing system is configured to further classify the second review groups according to the classification logic of “review points of the same review type are attributed to the same group” to form second review groups.

[0037] Further, the processing system is configured to analyze all files called by all review points contained in each second review group, and form a file list. Further, according to the number of files in the file list corresponding to each second review group, all review points in each second review group are classified according to the classification logic of “review points reviewing the same file are attributed to the same group” to form a plurality of third review groups. The second mapping relationship is used to store each third review group ID and the corresponding file name. Then, the processing system is configured to associate all third review groups corresponding to the same file name according to the second mapping relationship. The review points contained in the third review group are the associated review points mentioned above.

[0038] Then, the processing system generates a review thread for each second review group. The review thread refers to the execution flow of all review points contained in the second review group. The processing system further divides the review thread into sub-threads corresponding to the third review groups contained in each second review group. The sub-thread can be considered as an execution flow of all review points in the corresponding third review group. For third review groups with an association relationship, the corresponding sub-threads are also associated.

[0039] Then, through a preset asynchronous calling framework (such as Celery, Ray), a pre-built review model is called for each sub-thread, and according to the different review types corresponding to the review points, the review model is also divided into a text review model NLP model (such as BERT, FastText) and a visual review model CV model (such as YOLO, ResNet) to execute all review points contained in the corresponding sub-thread.

[0040] And since each secondary review group corresponds to a review thread including several sub-threads, in order to control the execution sequence of all sub-threads, the processing system also configures a review engine for each secondary review group corresponding to the review thread, which is used to control the review model to sequentially execute all sub-threads contained in the corresponding review thread according to the preset review order until all sub-threads are executed. The preset review order here is a random order defined by the system in advance, which can also be customized by humans.

[0041] And it needs to be explained here that the processing system will control the synchronous start of the review engines corresponding to all secondary review groups, so that the review threads of all secondary review groups are executed in parallel, and since all sub-threads corresponding to each secondary review group correspond to independent review models, all review models will be executed independently under the trigger control of the review engine. And in this process, if any executed sub-thread (hereinafter referred to as target sub-thread) is a sub-thread having an association relationship with other sub-threads, at this time the processing system will trigger the review model of all sub-threads (hereinafter referred to as associated sub-threads) having an association relationship with the target sub-thread to start executing the associated sub-threads, that is, at present, there are multiple sub-threads being executed in the review thread to which the associated sub-threads belong (i.e. the associated sub-threads, and the sub-threads currently triggered for execution by the review engine corresponding to the review thread to which the associated sub-threads belong). The processing system is used to update the completion execution progress of each review thread in real time, so as to make the corresponding review engine trigger the execution operation of the remaining unprocessed sub-threads in real time according to the preset review order.

[0042] Further, the processing system also merges the sub-threads with the association relationship into the same list (i.e., the sub-thread list), and as known from the above, the sub-threads in the same sub-thread list will be triggered for execution at the same time. At this time, considering that there may be a dependency relationship between the sub-threads in terms of processing logic, such as sub-thread M needs the output result of sub-thread N (for example, sub-thread N is for extracting text, and sub-thread M is for performing semantic analysis on the text extracted by sub-thread N), it is necessary to limit the execution mode between the sub-threads with the dependency relationship to sequential execution, and further determine the execution order of the sub-threads with the sequential execution mode. The execution mode of the sub-threads without the dependency relationship is defined as a parallel mode, so as to finally determine the scheduling scheme of all the sub-threads in each sub-thread list. The scheduling scheme includes all the sub-threads arranged in accordance with the execution order. Finally, when the sub-thread to be executed belongs to any sub-thread list, the corresponding scheduling scheme is used to trigger the execution of all the sub-threads in the sub-thread list to which the sub-thread belongs.

[0043] Optionally, the processing method further comprises: analyzing the specific review contents of the review points contained in all the sub-threads in the same sub-thread list, and judging the similarity of the specific review contents of the review points contained in different sub-thread lists; establishing a reuse relationship between the review points in different sub-thread lists that meet the preset requirement in terms of similarity, and determining the review points as reused review points, and merging the reused review points with the reuse relationship to form a reused review point set; analyzing the different review contents and similar contents between the reused review points with the reuse relationship in each reused review point set and belonging to different sub-thread lists, wherein the different review contents at least include the entity to be reviewed, and the similar contents are the same review contents in the specific review contents corresponding to the reused review points with the reuse relationship; generating a reuse template for each reused review point set, and the reuse template includes the similar contents and the different review contents, and constructing a reuse model for executing the reuse template; when executing a target sub-thread, if the target sub-thread contains a reused review point, a corresponding reuse model is called to execute the reused review point according to the different review contents corresponding to the target sub-thread through the reuse model; and all the review points of the non-reused review points contained in the target sub-thread are executed through the review model corresponding to the target sub-thread.

[0044] In implementation, exemplarily, if the sub-thread list XI corresponding to the file X includes multiple review points (a1, a2, a3); the sub-thread list Y1 corresponding to the file Y includes multiple review points (b1, b2, b3), and (a1, a2, a3) are similar to (b1, b2, b3) in specific review content of the file X, such as a1 and b1 are both for extracting text, a2 and b2 are both for identifying text content, and a3 and b3 are both for verifying content clarity; the difference is only in the entity of review, at this time, a1 and b1, a2 and b2, a3 and b3 can be considered as review points with reuse relationship, and the processing system is used to generate a reuse review point set for each review point with reuse relationship, such as (a1, b1), (a2, b2), (a3, b3), in other embodiments, the reuse review set corresponding to the same sub-thread list can be further merged, such as (a1, b1), (a2, b2), (a3, b3) correspond to the sub-thread list XI and Y1, so the union of the foregoing reuse review point set can obtain (a1, b1, a2, b2, a3, b3); then a reuse template is generated according to each reuse review point set, the reuse template includes common operation logic (i.e. the specific review content shared by the review points with reuse relationship, i.e. the similar content mentioned above, such as the operations of extracting, identifying, and verifying clarity) and difference review content, the difference review content can be specifically the entity to be reviewed (such as the specific text content in the extracted text described by a1, and the specific text content in the extracted text described by b1), the processing system is used to determine the similar content and the difference review content between the review points with reuse relationship, generate a reuse template, establish the corresponding relationship between the difference review content and the reuse review point, and construct a reuse model according to the common operation logic included in the reuse template, and input the difference review content as an input parameter into the reuse model, and use the reuse model to execute the specific review content corresponding to the reuse review point.

[0045] When any sub-thread to be executed (i.e. the target sub-thread described above) includes a reuse review point, all review points included in the target sub-thread are further divided into reuse review points and non-reuse review points, and the review model is used to execute the specific review content corresponding to the non-reuse review point, and the reuse model is used to review the specific review content corresponding to the reuse review point, and in the process of executing the reuse review point, only the difference review content corresponding to the reuse review point is input into the reuse model, so that the reuse model reviews the corresponding difference review content according to the common operation logic.

[0046] Optionally, the processing method further includes the following steps: The recording multiplexing model is used to record the review path of the multiplexing review points, and a natural language explanation report is output, wherein the review path at least includes multiplexing template content, corresponding differentiated review content, and location information of the differentiated review content in the corresponding file; The correction instruction proposed by the user is received, and the identity of the user is verified; wherein the correction instruction at least includes a multiplexing template to be corrected, and a correction content for correcting the corresponding differentiated review content; When the user identity verification is passed, the multiplexing template contained in the correction instruction is corrected based on the correction instruction.

[0047] In implementation, the processing system is used to record the review path of the multiplexing review points of the recording multiplexing model, generate a visual evidence chain (such as marking the location of the differentiated review content in the corresponding file in a highlighted form), and then use LLM (such as GPT-4) to convert the multiplexing logic into a natural language report output. An exemplary output can be "the review result of the clause is multiplexed in sub-file A, because the matching degree is 92%, only the company name field is adjusted".

[0048] In addition, the user can propose a correction instruction and enter identity information and access keys in the correction instruction. The processing system is used to determine whether the correction permission is included in the user permission according to the user identity information and the access keys. If it is included, the user identity verification is passed, and the differentiated review content corresponding to the multiplexing review points is replaced with the correction content.

[0049] In S104, the review results of all review threads are obtained and merged to output the output results for the user to know the output results.

[0050] In implementation, the review results of all review threads are merged after the final execution to form the output results, which are displayed to the preset display interface for the user to know. The review results at least include the judgment content of whether the review is passed or not, and the judgment content of whether the review operation execution is successful or not.

[0051] Referring to Figure 2 The processing system for improving identification processing efficiency based on point automatic grouping disclosed by the embodiments of the present application comprises: The file identification and analysis module 201 is used to receive the processing instruction proposed by the user, analyze all files contained in the processing instruction to obtain a review point list and a corresponding grouping thought chain; The review point classification module 202 is used to input all review points in the review point list and the corresponding grouping thought chain into a preset grouping intelligent agent, and classify all review points through the grouping intelligent agent to form a plurality of first-level review groups; The multi-thread synchronization enabling module 203 is configured to assign an examination model to each of the first examination groups by using an asynchronous calling framework, generate and start a plurality of examination threads, and the examination threads correspond to the first examination groups one by one, so as to process all examination points included in the corresponding first examination group. The processing result output module 204 is configured to obtain and merge the examination results of all the examination threads, and output the merged results to the user.

[0052] Optionally, the examination point classification module 202 is further configured to perform entity alignment processing on all the examination points in the examination point list, input the examination points after the entity alignment processing and the corresponding grouping thought chains into a preset grouping agent, classify all the examination points by using the grouping agent to form a plurality of first examination groups, so that different entities correspond to different first examination groups, and generate a first mapping relationship between the entities and the first examination groups; and further configured to generate second examination groups for the examination points in each of the first examination groups based on a preset examination type, wherein the examination type at least includes a pure text examination type and a visual image examination type. The multi-thread synchronization enabling module 203 is further configured to match an examination model to each of the second examination groups, generate and start a plurality of examination threads, and the examination threads correspond to the second examination groups one by one.

[0053] Optionally, the examination point classification module 202 is further configured to analyze the examination points included in each of the second examination groups, determine a file list corresponding to each of the second examination groups, the file list includes files required to be called to execute the examination points included in the corresponding second examination group, establish a second mapping relationship between the examination points in the second examination group and the files in the corresponding file list, and according to the second mapping relationship, merge all the examination points corresponding to the same file in each of the second examination groups as associated examination points to form third examination groups, and establish an association relationship between the third examination groups corresponding to different second examination groups and the same file. The multi-thread synchronization enabling module 203 is further configured to generate an examination thread for each of the second examination groups, according to the third examination groups included in each of the second examination groups, split the examination thread corresponding to each of the second examination groups into sub-threads corresponding to each of the third examination groups one by one, and according to the association relationship between the third examination groups, establish an association relationship between the corresponding sub-threads; assign an examination model and an examination engine to each of the sub-threads by using an asynchronous calling framework, control the corresponding examination model to execute the corresponding sub-thread according to a preset examination order by using the examination engine, until the corresponding examination thread is completed; and when a target sub-thread is executed, trigger the execution of the sub-threads having an association relationship with the target sub-thread, wherein the target sub-thread is an arbitrary sub-thread.

[0054] Optionally, the multi-thread synchronization enabling module 203 is also used to classify sub-threads with associated relationships to form a sub-thread list, analyze the review points contained in each sub-thread in the same sub-thread list, and based on the analysis results, determine the dependency relationships between all sub-threads in each sub-thread list, determine the execution mode of all sub-threads in each sub-thread list according to the dependency relationships, and determine the execution order for sub-threads whose execution mode is sequential execution; wherein the execution mode includes at least parallel execution or sequential execution; generate a scheduling plan for each sub-thread list according to the execution mode and execution order; and trigger the execution of the sub-thread with an associated relationship with the target sub-thread according to the scheduling plan corresponding to the target sub-thread when executing the target sub-thread.

[0055] Optionally, it also includes a reuse review module, which is used to analyze the specific review contents of the review points contained in all sub-threads within the same sub-thread list, and determine the similarity of the specific review contents of the review points contained in different sub-thread lists; establish a reuse relationship between the review points in different sub-thread lists whose similarity meets the preset requirements, and determine them as reused review points, and merge the reused review points with a reuse relationship to form a reused review point set; it is also used to analyze the different review contents and similar contents between the reused review points in each reused review point set that belong to different sub-thread lists and have a reuse relationship, wherein the different review contents include at least the entity to be reviewed, and the similar contents are the same review contents in the specific review contents corresponding to the reused review points with a reuse relationship; generate a reuse template for each reused review point set, and the reuse template includes similar contents and different review contents, and construct a reuse model for executing the reuse template; The multi-thread synchronization enabling module 203 is also used to call the corresponding reuse model if the target sub-thread contains reused review points, and execute the reused review points through the reuse model according to the differentiated review content corresponding to the target sub-thread; and execute all non-reuse review points contained in the target sub-thread through the review model corresponding to the target sub-thread.

[0056] Optionally, it also includes an audit path recording module for recording the audit path of the reuse model for the reuse review points and outputting a natural language explanation report, wherein the audit path includes at least the reuse template content, the corresponding difference review content, and the location information of the difference review content in the corresponding file.

[0057] Optionally, a multiplex template correction module is further included, configured to receive a correction instruction proposed by a user, and verify the identity of the user, wherein the correction instruction at least includes a multiplex template to be corrected and correction content for correcting corresponding distinguished examination content; when the identity of the user is verified, the multiplex template included in the correction instruction is corrected based on the correction instruction.

[0058] The embodiment of the present application further discloses a processing device for improving recognition processing efficiency based on automatic grouping of points, which comprises a memory and a processor, and the memory stores a computer program capable of being loaded and executed by the processor to perform the efficient processing method for batch files.

[0059] The embodiment of the present application further discloses a computer readable storage medium which stores a computer program capable of being loaded and executed by the processor to perform the processing method for improving recognition processing efficiency based on automatic grouping of points, and the computer readable storage medium comprises various storage media capable of storing program codes, for example, a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0060] It should be noted that, in this document, relational terms such as first and second and the like can only be used to distinguish one entity or action from another entity or action, and do not necessarily require or imply that there is any such actual relationship or order between these entities or actions.

[0061] The above embodiments are only used to illustrate the technical solutions of the present application, but not limit the protection scope of the application. Obviously, the described embodiments are only part of the embodiments of the present application, but not all embodiments. Based on these embodiments, all other embodiments obtained by those of ordinary skill in the art without creative work belong to the scope of protection of the present application.

Claims

1. A processing method for improving recognition processing efficiency based on automatic grouping of key points, characterized in that: include: Receive processing instructions from users, parse all files included in the processing instructions to obtain a list of review points and corresponding group thinking chains; Input all the key review points in the key review point list and the corresponding grouping thinking chains into a preset grouping agent, and use the grouping agent to classify all the key review points into several first-level review groups; Through the asynchronous call framework, a review model is assigned to each of the first-level review groups, and a number of review threads are generated and started. Each review thread corresponds to a first-level review group one by one, so as to process all review points included in the corresponding first-level review group; Obtain and merge the review results of all review threads to output them for users to know the output results.

2. The method for improving recognition processing efficiency based on automatic grouping of key points according to claim 1 is characterized in that: The examination points shall at least include the entity to be examined; All the review points in the review point list and the corresponding group thinking chains are input into a preset grouping agent, and all the review points are classified into several first-level review groups through the grouping agent, including: Performing entity alignment processing on all key review points in the key review point list, inputting the key review points after the entity alignment processing and the corresponding grouping thinking chains into a preset grouping agent, and classifying all key review points into a plurality of first-level review groups through the grouping agent, so that different entities correspond to different first-level review groups, thereby generating a first mapping relationship between entities and first-level review groups; Based on the preset review types, classify the review points in each first-level review group to generate a second-level review group; wherein the review types include at least a plain text review type and a visual image review type; The asynchronous call framework is used to assign an examination model to each of the first-level examination groups, generate and start a number of examination threads, and the examination threads correspond to the first-level examination groups one by one, including: An examination model is matched for each of the secondary examination groups, and a number of examination threads are generated and started, and the examination threads correspond to the secondary examination groups one by one.

3. The method for improving recognition processing efficiency based on automatic grouping of key points according to claim 2, characterized in that: The step of inputting all the key review points in the key review point list and the corresponding grouping thinking chains into a preset grouping agent, and classifying all the key review points into a plurality of first-level review groups through the grouping agent, further includes: Analyzing the key examination points included in each of the second-level examination groups, determining a file list corresponding to each of the second-level examination groups, wherein the file list includes files required to implement the key examination points included in the corresponding second-level examination group; and establishing a second mapping relationship between the key examination points in the second-level examination group and the files in the corresponding file list; According to the second mapping relationship, all examination points corresponding to the same document in each of the second-level examination groups are respectively taken as associated examination points, and are merged to form a third-level examination group; Establish associations between third-level review groups that belong to different second-level review groups and correspond to the same documents; The above-mentioned steps are to match the review model for each secondary review group, generate and start a number of review threads, and each review thread corresponds to a secondary review group one by one, including: Generate a review thread for each second-level review group respectively. Based on the third-level review groups contained in each second-level review group, split the review thread corresponding to each second-level review group into sub-threads corresponding to each third-level review group. And establish association relationships between the corresponding sub-threads based on the association relationships between the third-level review groups. Through the asynchronous call framework, an inspection model and an inspection engine are respectively assigned to each of the sub-threads, and the inspection engine controls the corresponding inspection model to execute the corresponding sub-thread according to the preset inspection order until the corresponding inspection thread is completed; and when executing the target sub-thread, the sub-thread associated with the target sub-thread is triggered to execute at the same time, wherein the target sub-thread is any sub-thread.

4. The method for improving recognition processing efficiency based on automatic grouping of key points according to claim 3 is characterized in that: The method further includes matching the review model for each secondary review group, generating and starting a plurality of review threads, wherein the review threads correspond one to one with the secondary review groups, and further includes: Classifying associated sub-threads into a sub-thread list, analyzing key review points contained in each sub-thread in the same sub-thread list, and determining dependencies between all sub-threads in each sub-thread list based on the analysis results. Determining execution modes for all sub-threads in each sub-thread list based on the dependencies, and determining an execution order for sub-threads whose execution mode is sequential execution; wherein the execution mode includes at least parallel execution or sequential execution; Generate a scheduling plan for each sub-thread list according to the execution mode and execution sequence; The method of simultaneously triggering execution of a sub-thread associated with the target sub-thread when executing the target sub-thread includes: When executing a target sub-thread, a sub-thread associated with the target sub-thread is triggered for execution according to a scheduling scheme corresponding to the target sub-thread.

5. The method for improving recognition processing efficiency based on automatic grouping of key points according to claim 4 is characterized in that: The method further comprises: Analyze the specific review contents of the key examination points contained in all sub-threads within the same sub-thread list, and determine the similarity of the specific review contents of the key examination points contained in different sub-thread lists; Establish a reuse relationship between examination points in different sub-thread lists whose similarity meets the preset requirements and determine them as reused examination points. Merge the reused examination points with the reuse relationship to form a reused examination point set. Analyze the differences and similarities between reused key examination points in each reused key examination point set, belonging to different sub-thread lists, and having a reuse relationship. The differences include at least the entity to be reviewed, and the similarities are the same review contents in the specific review contents corresponding to the reused key examination points. Generate a reuse template for each reuse review point set, wherein the reuse template includes similar content and different review content, and construct a reuse model for executing the reuse template; The method of simultaneously triggering execution of a sub-thread associated with the target sub-thread when executing the target sub-thread also includes: If the target sub-thread contains reused review points, the corresponding reuse model is called, and the reused review points are executed through the reuse model according to the differentiated review content corresponding to the target sub-thread; all non-reuse review points contained in the target sub-thread are executed through the review model corresponding to the target sub-thread.

6. The method for improving recognition processing efficiency based on automatic grouping of key points according to claim 1, characterized in that: The method further comprises: Record the review path of the reuse model for the reuse review points and output a natural language explanation report, wherein the review path includes at least the reuse template content, the corresponding difference review content, and the location information of the difference review content in the corresponding file.

7. The method for improving recognition processing efficiency based on automatic grouping of key points according to claim 1, characterized in that: The method further comprises: Receive a correction instruction from a user and verify the user's identity; wherein the correction instruction includes at least the reuse template to be corrected and the correction content to be corrected for the corresponding differentiated review content; When the user identity authentication is passed, the multiplexing template contained in the correction instruction is corrected based on the correction instruction.

8. A processing system for improving recognition processing efficiency based on automatic grouping of key points, characterized in that: include, A file identification and parsing module (201) is used to receive a processing instruction proposed by a user, parse all files included in the processing instruction to obtain a list of review points and corresponding group thinking chains; An examination point classification module (202) is used to input all the examination points in the examination point list and the corresponding grouping thinking chains into a preset grouping intelligent body, and classify all the examination points into a number of first-level examination groups through the grouping intelligent body; A multi-threaded synchronous activation module (203) is used to allocate an examination model to each of the first-level examination groups through an asynchronous call framework, generate and start a plurality of examination threads, and the examination threads correspond to the first-level examination groups one by one, so as to process all the examination points included in the corresponding first-level examination groups; The processing result output module (204) is used to obtain and merge the review results of all review threads to output the results to the user.

9. A processing device for improving recognition processing efficiency based on automatic grouping of key points, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that A computer program is stored which can be loaded by a processor and execute the method according to any one of claims 1 to 7.

Images