Adaptive knowledge graph construction and deep analysis method supporting multiple types of files
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SUZHOU ZHONGGE SOFTWARE
- Filing Date
- 2025-05-06
- Publication Date
- 2026-06-05
Smart Images
Figure CN120409649B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of document processing technology, and in particular to an adaptive knowledge graph construction and deep analysis method that supports multiple types of documents. Background Technology
[0002] With the development of information technology, the number of documents and materials that enterprises and individuals need to process and review is growing exponentially, especially in the legal, financial, scientific research, and government sectors, where the demand for multi-document review is becoming increasingly prominent. Traditional document processing and review methods mainly rely on manual review and comparison of document content. When faced with a large number of documents and numerous review points to be reviewed and compared, this manual approach is not only inefficient but also prone to errors or omissions due to human factors, thus requiring improvement. Summary of the Invention
[0003] To achieve efficient and accurate processing of multiple types of documents, this application provides an adaptive knowledge graph construction and deep analysis method that supports multiple document types.
[0004] Firstly, this application provides an adaptive knowledge graph construction and deep analysis method that supports multiple file types, employing the following technical solution:
[0005] Receive a processing instruction from a user, parse all files contained in the processing instruction, and the parsing operation includes at least extracting file entities;
[0006] Based on the parsing results, a processing item and a corresponding processing template are matched for the processing instruction; wherein, the processing template contains processing content for processing entities in the file;
[0007] Based on a preset entity alignment technique, the association relationship is determined and established among all files included in the processing instructions;
[0008] Based on the determined processing template and file relationships, all processing is executed through a preset processing engine, and the processing results are fed back to the user.
[0009] By adopting the above technical solution, for multiple files input by the user, this application uses the above technical solution to automatically parse and extract key content (such as entities) from each file, and matches specific processing matters based on the key content of the file. In this way, even without knowing the specific processing purpose triggered by the user's processing instruction, the user's processing needs can be adaptively known by analyzing the file content. Then, based on entity alignment technology and the file entities extracted above, cross-file associations are established, and the file entities are used as the link between the association and the processing content to realize the correspondence between the association and the processing content. This allows the processing engine to quickly locate the file to be processed and the specific entities in the file based on the correspondence when executing the processing content. This can efficiently summarize the relationships between files from a complex number of files, help improve processing efficiency, and replace manual processing with this intelligent processing method, which can also solve the problem of processing accuracy affected by human error.
[0010] Optionally, the method further includes:
[0011] Once a processing item is matched, it is determined whether there is a derivative requirement that meets the preset derivative conditions. If so, the derivative item corresponding to the derivative requirement is determined, and a processing guide for the derivative item is generated and fed back to the user based on the processing template corresponding to the derivative item.
[0012] By adopting the above technical solutions and achieving processing through direct matching, we can independently discover derived needs, upgrade from "passive response" to "proactive service", deeply analyze and explore user needs, plan and guide users on handling matters from a long-term perspective, optimize the level of service intelligence and automation, and improve user experience.
[0013] Optionally, determining whether there is a derivative requirement that satisfies the preset derivative conditions, and if so, determining the derivative matter corresponding to the derivative requirement, includes:
[0014] By using a pre-constructed item relationship graph, it is determined whether there exists a relationship chain containing the processing item, the relationship chain also contains other items besides the processing item, and there are association relationships between adjacent items in the relationship chain, and there are corresponding association requirements for describing the association relationships; wherein, the item relationship graph is used to characterize the association relationships between different items;
[0015] Based on all the files contained in the processing instruction, the association strength between each item in the association chain and the processing item is calculated. Items whose association strength meets the preset derivation conditions are designated as derivation items, and the corresponding association requirements are designated as derivation requirements. The association strength includes at least the processing completion rate of the item to which the processing instruction is used to process the item.
[0016] By adopting the above technical solution, this application pre-defines a matter relationship graph to describe the correlation between different matters. The matters in the matter relationship graph are bridged by the correlation requirements. Based on this relationship graph, all matters related to the currently processed matter are captured, thereby achieving a leap from single-point matching to global intelligent matter mining, helping users to more comprehensively explore their long-term needs. Furthermore, this application calculates the correlation strength between each matter and the processed matter to achieve accurate capture and intelligent filtering of derivative matters. When calculating the correlation, this application fully considers the processing completion rate of existing files used to execute derivative matters, thereby improving the filtering accuracy of derivative matters and making the filtered derivative matters more in line with actual objective conditions.
[0017] Optionally, the parsing operation further includes semantic recognition of the file and extraction of keywords;
[0018] The process of matching processing items and corresponding processing templates for the processing instructions based on the parsing results includes:
[0019] Based on the dataset corresponding to each item stored in the pre-set structured database, the keywords in the parsing results are compared with the dataset to determine all candidate items and the similarity of each candidate item.
[0020] The candidate items with the highest similarity are selected as the processing items and their corresponding processing templates are determined. All candidate items other than the processing items are selected as derivative items.
[0021] By adopting the above technical solution, this solution specifically discloses a matching logic for matching processing items. When faced with multiple items (i.e., candidate items) that are similar to the document content (i.e., keywords), this application proposes to select the item with the highest similarity as the processing item, thus limiting the processing item to be unique. At the same time, in order to ensure that the user's actual desired task can be completed more efficiently, this application treats all other candidate items that were not selected as processing items as derivative items, thereby connecting with the subsequent processing scheme of derivative items mentioned above.
[0022] Optionally, the method further includes:
[0023] Determine whether there are any conflicts between the processing item and all its corresponding derivative items. If so, determine the conflict resolution solution corresponding to the conflict type based on the conflict issue, and add the conflict issue and its corresponding conflict resolution solution to the processing guidelines for the derivative items.
[0024] The conflict issue refers to a situation where the entities shared by the processing matter and the derived matters are processed according to the corresponding processing content, resulting in a contradiction.
[0025] The conflict types include at least the following: when processing matters and derived matters perform corresponding processing on a common entity, they contradict each other because the specific content of the common entity is inconsistent.
[0026] By adopting the above technical solution, this application further analyzes the processing items and derivative items to predict whether conflicts will occur during the execution of the processing items and derivative items. The essence of the conflict is the contradiction between the specific processing content contained in the processing templates corresponding to each item. For example, the files uploaded by the user include file A: business license (containing address a) and file B: new lease contract (containing address b). The final selected processing item is "annual inspection of business license", which includes the derivative item "address change filing". There is a common entity: "address". However, when executing the processing item, it is required to lock the original address (i.e., address a), while when processing the above derivative item, it is required to overwrite the original address (i.e., address a) with address b. Therefore, conflicting operations occur when processing the processing item and the derivative item. In this regard, this application proposes to predict the conflict in the execution process of the processing item and all its corresponding derivative items, and proposes a corresponding conflict resolution solution to resolve the conflict.
[0027] Optionally, the method further includes:
[0028] Periodically, based on historically identified conflict issues, identify relevant matters as target matters.
[0029] The processing template corresponding to the target item is decomposed and recombined to satisfy the following condition: there are conflicting contents in the recombined processing contents, and the conflicting contents are processing contents that only contain the corresponding conflicting issues. The conflicting contents, conflicting issues, and conflicting relationships between the corresponding items are established and stored.
[0030] The conflict resolution scheme includes at least prioritizing the handling of content other than the conflicting content.
[0031] By adopting the above technical solution, the specific processing content of each matter can be broken down and reorganized in detail according to the conflict between matters. In order to prioritize the processing content of multiple matters other than the conflicting content when a conflict occurs, on the one hand, the granularity of conflict detection can be improved, and on the other hand, the processing content of multiple matters that do not have conflict issues can be processed in parallel, thereby improving the processing progress of each matter.
[0032] Optionally, the method further includes:
[0033] Whenever a processing item is matched, the matching relationship between the processing item and its corresponding key file is determined and stored. The key file refers to the file that plays a decisive role in the matching process of determining the processing item.
[0034] Based on the matching relationships stored in historical periods, the sensitivity of each key document is analyzed periodically, and key documents with sensitivity higher than a preset sensitivity threshold are marked as high-sensitivity documents. A set of items is determined for each high-sensitivity document, and the set of items contains items that have a matching relationship with the corresponding high-sensitivity document.
[0035] When matching processing items for a processing instruction, if a highly sensitive file exists in all the files contained in the processing instruction, the items in the item set corresponding to the highly sensitive file will be retrieved first to perform the matching operation.
[0036] By adopting the above technical solution, the relationship between documents and processing items is analyzed to determine document sensitivity. Document sensitivity can be used to characterize the scarcity of documents, such as documents that are only used when handling certain special matters (i.e., high-sensitivity documents). By marking high-sensitivity documents and using them in subsequent matching operations for processing items, and by retrieving the items corresponding to the high-sensitivity documents to perform matching operations, the matching efficiency of processing items can be improved.
[0037] Secondly, this application provides an adaptive knowledge graph construction and deep analysis system that supports multiple file types, including:
[0038] The file parsing module is used to receive processing instructions from users and parse all files contained in the processing instructions. The parsing operation includes at least extracting file entities.
[0039] The item matching module is used to match the processing instruction with processing items and corresponding processing templates based on the parsing results; wherein, the processing template contains processing content for processing entities in the file;
[0040] The cross-file association module is used to determine and establish association relationships among all files included in the processing instruction based on a preset entity alignment technology.
[0041] The task processing module is used to execute all processing content through a preset processing engine based on the determined processing template and the relationship between files, and to provide feedback on the processing results to the user.
[0042] Thirdly, this application provides an adaptive knowledge graph construction and deep analysis apparatus that supports multiple types of files, including a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and executed as described in any of the first aspects.
[0043] Fourthly, this application provides a computer-readable storage medium storing a computer program that can be loaded by a processor and executed as described in any of the first aspects.
[0044] In summary, this application includes at least one of the following beneficial technical effects:
[0045] 1. In this application, multiple files are acquired and each file is automatically parsed. Based on the parsing results, specific processing items are matched. In this way, the user's processing needs can be adaptively known by analyzing the file content without knowing the specific processing purpose of the user's triggered processing instructions.
[0046] 2. Furthermore, this application establishes cross-document relationships based on knowledge graph data and entity alignment technology. Then, based on these relationships and corresponding processing templates, it ultimately achieves efficient processing of the matters to be processed. In summary, by efficiently extracting and summarizing the relationships between numerous documents, processing efficiency is improved. Moreover, this intelligent processing method, replacing manual processing, also solves the problem of accuracy issues caused by human error. Attached Figure Description
[0047] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0048] Figure 1 This is a flowchart illustrating the adaptive knowledge graph construction and deep analysis method supporting multiple file types disclosed in the embodiments of this application.
[0049] Figure 2 This is a structural block diagram of the adaptive knowledge graph construction and deep analysis system supporting multiple file types disclosed in the embodiments of this application.
[0050] Explanation of reference numerals in the attached diagram: 201, File parsing module; 202, Item matching module; 203, Cross-file association module; 204, Item processing module. Detailed Implementation
[0051] The following is in conjunction with the appendix Figure 1-2 This application will be described in further detail.
[0052] This application discloses an adaptive knowledge graph construction and deep analysis method (hereinafter referred to as the analysis method) supporting multiple file types. The method aims to efficiently parse multiple source files to derive user requirements, and then respond to and process these requirements. The execution entity of the analysis method is an adaptive knowledge graph construction and deep analysis system supporting multiple file types (hereinafter referred to as the analysis system). The following will combine... Figure 1 The specific steps of the analysis system in executing the analysis method are described in detail.
[0053] S101: Receive the processing instructions from the user, parse all the files contained in the processing instructions, and the parsing operation includes at least extracting the file entities.
[0054] In practice, for example, users can access the analysis system via a webpage and upload multiple source files to trigger processing instructions. Multiple source files refer to several files with different formats (e.g., PDF, Word, images, etc.) and contents. After receiving all the files included in the processing instructions, the analysis system will parse them. The parsing process specifically includes: extracting text and tables from PDF and Word documents using open-source libraries (e.g., Apache PDFBox, Python's pdfplumber / docx libraries); recognizing documents in images and scanned documents using OCR technology (e.g., Tesseract, Alibaba Cloud OCR API); and then extracting entities (e.g., names, addresses, dates, etc.) from the extracted content using pre-trained models (e.g., SpaCy, Alibaba Cloud NLP).
[0055] S102, based on the parsing result, match the processing item and the processing template corresponding to the processing item for the processing instruction; wherein, the processing template contains processing content for processing entities in the file.
[0056] S102 specifically includes the following sub-steps:
[0057] Based on the dataset corresponding to each item stored in the pre-set structured database, the keywords in the parsing results are compared with the dataset to determine all candidate items and the similarity of each candidate item; the candidate item with the highest similarity is selected as the processing item, and the corresponding processing template is determined.
[0058] In implementation, the "processing item" represents the task a user needs to handle. The analysis system analyzes all the files entered by the user when triggering the processing instruction, and based on the parsing results of the file content, it determines the processing item the user wants to handle—that is, matching the processing item. Correspondingly, the analysis system pre-builds a structured knowledge base, which can be a graph database (Neo4j) or a relational database, to store the specific file content required for each processing item, realizing the correspondence between "processing item - required file content". For example, when the processing item is "opening a shop", it requires documents such as a "business license", "lease agreement", and "hygiene permit".
[0059] The analysis system compares the content of the specific files corresponding to all processing items stored in the knowledge base with the parsing results and calculates the similarity, such as using an NLP model (e.g., BERT) to calculate the similarity. For example, if the parsing results contain the keywords "lease" or "shop," then the corresponding matching content is "open a shop."
[0060] If multiple items are involved in the above comparison process, such as the appearance of "fire inspection checklist" in the analysis results, the corresponding matching content is "hygiene permit" as the processing item. In this regard, this application further proposes to combine TF-IDF keyword weights and BERT sentence embedding similarity to calculate the matching score, and select the item with the highest matching score as the processing item, that is, select the item with the highest matching degree as the only processing item.
[0061] Furthermore, each processing item also corresponds to a processing template. The processing template contains the specific processing content for handling the corresponding processing item. The processing content can be represented in the form of "entity + processing operation," that is, it specifically describes the specific processing operation performed on the entity. For example, when the processing item is "opening a store," its corresponding processing template must contain at least one processing item: "Verify the consistency between the business license address and the lease agreement address." The entity corresponding to this content is "address," and both "business license address" and "lease agreement address" are specific descriptions of the "address" entity, and they come from different documents. Therefore, this application proposes to use entity alignment technology to establish the association between documents to facilitate the processing of the above-mentioned processing content. The corresponding process steps are as follows:
[0062] S103, based on the preset entity alignment technology, determines and establishes the association relationship among all files included in the processing instruction;
[0063] S104: Based on the determined processing template and file relationships, execute all processing content through the preset processing engine and provide feedback on the processing results to the user.
[0064] In implementation, the analysis system pre-constructs a knowledge graph for each entity. This knowledge graph contains all descriptive content for the same entity and the files to which each description belongs. Therefore, the analysis system can first establish associations between different descriptive contents of the same entity using entity alignment technology and the aforementioned knowledge graph. Then, it establishes associations between files based on the files to which the descriptive content belongs. Finally, it executes the processing content in the processing template based on these associations until all processing content in the template is processed. Afterward, it provides feedback to the user with the processing result, which includes processing completion and processing failure. Furthermore, if the inability to establish file associations due to missing files in the processing instructions results in unprocessable content, the corresponding feedback result is "processing failure." In this case, the analysis system provides the reason for the failure, i.e., outputs the specific processing content that failed.
[0065] Optionally, the analysis method may also include the following steps:
[0066] S105: Whenever a processing item is matched, all alternative items other than the processing item are treated as derivative items; based on the processing item, it is determined whether there is a derivative requirement that meets the preset derivative conditions. If so, the derivative item corresponding to the derivative requirement is determined, and based on the processing template corresponding to the derivative item, a processing guide for the derivative item is generated and fed back to the user.
[0067] Specifically, S105, "determining whether there are derivative requirements that meet the preset derivative conditions, and if so, determining the derivative matters corresponding to the derivative requirements," includes:
[0068] By using a pre-built item relationship graph, it is determined whether there are relationship chains containing processing items, which also contain other items besides processing items, and whether there are association relationships between adjacent items in the relationship chain, and corresponding association requirements for describing the association relationships; wherein, the item relationship graph is used to represent the association relationships between different items;
[0069] Based on all the files contained in the processing instruction, calculate the association strength between each item in the association chain and the processing item. Items whose association strength meets the preset derivation conditions are designated as derivation items, and the corresponding association requirements are designated as derivation requirements. The association strength includes at least the processing completion rate of the item to which the processing instruction is used to process the processing of all the files contained in the processing instruction.
[0070] In practice, as can be seen from the preceding text, the processing item matched by this application is the only processing item. Therefore, for other alternative items besides the processing item, this application defines them as derivative items.
[0071] Furthermore, this application proposes to further screen other derivative matters related to the processing matters based on the processing matters and a preset matter relationship graph. Then, the union of all derivative matters obtained from the two screening processes is used to obtain the final derivative matter set. Finally, based on the processing template corresponding to each derivative matter, a corresponding derivative matter processing guide is generated and provided to the user. This guide describes the documents required to execute the corresponding derivative matter and the specific processing content. The specific method for further screening other derivative matters related to the processing matters based on the processing matters and the preset matter relationship graph is as follows:
[0072] The relationship graph of items can be represented as a topological graph, with items as nodes. Items with relationships are connected by arrows, and the association requirements are annotated at the line segments (e.g., item A is a prerequisite for item B). The association requirements are used to describe their relationship. When the item being processed is "Restaurant Opening Registration", other items related to this item include "Food Business License", "Fire Inspection", and "Hygiene Permit Application". When the item being processed is "Company Registration", it is associated with the item "Tax Registration". Since "Tax Registration" itself is a prerequisite for another item, "Invoice Application", a relationship chain of "Company Registration → Tax Registration → Invoice Application" is generated.
[0073] After identifying all items and relationship chains related to the processing task, the analysis system analyzes the correlation strength between each item and the processing task. The rules for calculating the correlation strength can be predefined by the user. The following is an example of a method for calculating correlation strength:
[0074] The analysis system predefines several association types and corresponding association strength values for each type. The association types include: legally mandated association types (i.e., government regulations require certain procedures to be combined, such as "restaurant opening registration" and "food business license," "fire inspection," and "hygiene license" being legally related matters); business logic dependency types (i.e., matters that are not legally required but have preconditions or dependencies between them, such as "business loan application" and "mortgage registration" and "credit inquiry" having business dependencies); and user potential demand types (through historical data mining, after handling the matters, most users will handle additional matters, such as after handling "individual business license," users will further handle "tax registration" and "social security account opening."
[0075] The analysis system will analyze the relationship between each item and the processing item, determine the type of association, and identify the strength of the association. Then, it will compare the required documents for each item with all documents included in the current processing instruction to obtain a similarity score, which will be used to represent the processing completion rate. Finally, based on preset weights, the association strength score and the similarity score will be weighted and summed to obtain a total score. If the total score is higher than the preset score, it is considered to meet preset derivation conditions, and the corresponding item will be treated as a derivative item of the processing item.
[0076] Optionally, the analysis method may also include the following steps:
[0077] Determine whether there are any conflicts between the matter to be handled and all its corresponding derivative matters. If so, determine the conflict resolution solution based on the conflict type and add the conflict and its corresponding conflict resolution solution to the derivative matter handling guidelines.
[0078] Among them, the conflict problem refers to the situation where the entities shared by the subject matter and the derivative matters are processed according to the corresponding processing content and contradictions occur.
[0079] Conflict types include at least the following: when processing matters and derived matters perform corresponding processing on a common entity, they contradict each other because the specific content of the common entity is inconsistent.
[0080] In implementation, conflict types include data-level conflicts, where the processing item and the derivative item contradict each other when performing corresponding processing on a shared entity due to inconsistencies in the specific content of the shared entity. Specifically, the specific values of the same entity field may differ in different files. For example, a user submits: File A, "Business License" (including address: No. 1, XX Road), and File B, "New Lease Contract" (including address: No. 2, XX Road). Based on the aforementioned solution, the processing item "Annual Inspection of Business License" has been matched for the current case. However, since the alternative implementation "Address Change Filing" matched from File B becomes a derivative item, a conflict arises where the specific values of the same entity conflict in different files. Furthermore, the two files correspond to different items, leading to a conflict between the processing item and the derivative item. When executing the processing item (i.e., processing the annual inspection of the business license), the original address (No. 1, XX Road) needs to be locked. However, when processing the derivative item (i.e., processing the address change filing), the new address (No. 2, XX Road) needs to be used to overwrite the original address.
[0081] Conflict types can also include rule-level conflicts, where business rules prohibit certain combinations of implementations (such as "deregistration" and "social security"), thus creating a conflict between the processing item "company deregistration" and the derivative item "social security payment".
[0082] Conflict types can also include state-level conflicts, where both Item A and Item B require resource X (such as document X), but their required states for resource X differ. For example, a user submits document C, "Loan Application Form," which corresponds to the processing item "Business Loan," and document D, "Contract Filing Application," which corresponds to the derivative item "Contract Filing." However, applying for a "Business Loan" requires mortgaging a property certificate (i.e., resource X), but applying for "Contract Filing" also requires the property certificate. That is, applying for "Contract Filing" requires the property certificate to be in an "idle" state, but applying for a "Business Loan" requires the property certificate to be in an "occupied" state. Therefore, a conflict exists between the aforementioned "Business Loan" and "Contract Filing."
[0083] The analysis system can predefine conflict types and corresponding conflict cases for each type. Conflict cases contain specific conflicting matters, facilitating the system's analysis of potential conflicts after each derived matter is generated. Furthermore, for each conflict type, the system pre-stores corresponding conflict resolution solutions. For example, a conflict resolution solution might include: a priority coverage strategy: legally mandated matters take precedence over user-initiated matters, which in turn take precedence over system-recommended derived matters. For instance, address changes (a legally mandated matter) are processed before business license annual inspections (a user-initiated matter). It could also include: user interaction, allowing users to choose their preferred resolution method, and resolving the conflict based on the user's feedback.
[0084] Optionally, the analysis methods also include:
[0085] Regularly identify conflict issues based on historical findings and designate related matters as target matters.
[0086] Decompose and reorganize all processing content contained in the processing template corresponding to the target item, and satisfy the following conditions: there are conflicting contents in the reorganized processing content, and the conflicting contents are processing contents that only contain the corresponding conflicting issues. Establish and store the conflicting contents, conflicting issues, and conflicting relationships between the corresponding items.
[0087] Conflict resolution solutions should at least include prioritizing the handling of issues other than the content of the conflict.
[0088] During implementation, conflicts arising in historical periods and their frequency of occurrence are periodically analyzed. When the frequency exceeds a preset range, the corresponding issue is designated as a target issue. The processing content of the target issue is then decomposed and reorganized. This means that each processing content is ultimately broken down into indivisible processing steps. Each processing step is defined as a minimum business processing rule that is independent of other processing steps and can be executed independently. Conflicts are identified from the decomposed processing content, and the non-conflict processing content can be reorganized. Subsequently, whenever a conflict related to the conflicting content is encountered, the non-conflict content of the issue can be processed first. This decomposition method categorizes the processing content, making it easier to identify processing content that can be processed in parallel, thus helping to advance the processing progress of the corresponding issue.
[0089] Optionally, the analysis method may also include the following steps:
[0090] Whenever a task is matched, the matching relationship between the task and its corresponding key documents is determined and stored. Key documents are those that play a decisive role in the matching process of determining the task.
[0091] Periodically analyze the sensitivity of each key document based on the matching relationships stored in historical periods, and mark key documents with sensitivity higher than the preset sensitivity threshold as high-sensitivity documents. For each high-sensitivity document, determine a set of items, which contains items that have a matching relationship with the corresponding high-sensitivity document.
[0092] When matching processing items for a processing instruction, if a highly sensitive file exists in all the files included in the processing instruction, the items in the item set corresponding to the highly sensitive file will be retrieved first to perform the matching operation.
[0093] In practice, based on the specific matching process of the above-mentioned matching processing items, it can be seen that this application is used to compare the similarity between the file parsing results and the specific file content stored in the knowledge base corresponding to the item. In this process, the analysis system will establish a matching relationship of "item-key file". It can be considered that the parsing results extracted from the key file are consistent with the specific file content stored in the corresponding item. That is to say, all the key files corresponding to the item are matched together to obtain the corresponding item.
[0094] After determining the above matching relationships, the analysis system will determine the frequency of each key file's occurrence in all matching relationships as the sensitivity of the corresponding key file, and consider that the fewer the occurrences, the higher the sensitivity. When the sensitivity is higher than the preset sensitivity threshold, the corresponding key file is marked as a high-sensitivity file. Accordingly, each time a processing instruction needs to be matched with a processing item, it will first determine whether the processing instruction contains a high-sensitivity file. If it does, it will first retrieve the items that have a matching relationship with the high-sensitivity file to perform the matching operation, thereby limiting the order of matching with all items.
[0095] This application also discloses an adaptive knowledge graph construction and deep analysis system that supports multiple file types. (See also...) Figure 2 ,include:
[0096] The file parsing module 201 is used to receive processing instructions from the user and parse all the files contained in the processing instructions. The parsing operation includes at least extracting file entities.
[0097] The item matching module 202 is used to match processing items and corresponding processing templates for processing instructions based on the parsing results; wherein, the processing template contains processing content for processing entities in the file;
[0098] The cross-file association module 203 is used to determine and establish association relationships among all files included in the processing instruction based on a preset entity alignment technology.
[0099] The task processing module 204 is used to execute all processing content through a preset processing engine based on the determined processing template and the relationship between files, and to provide feedback on the processing results to the user.
[0100] Optionally, it also includes a derivative demand mining module, which is used to determine whether there are derivative demands that meet preset derivative conditions after each matching and processing item is obtained. If there are, the module determines the derivative items corresponding to the derivative demands and generates and feeds back the derivative item processing instructions to the user based on the processing template corresponding to the derivative items.
[0101] Optionally, the derivative requirement mining module is also used to determine, through a pre-built item relationship graph, whether there exists a relationship chain containing processing items, where the relationship chain also contains other items besides the processing items, and there are associations between adjacent items in the relationship chain, and corresponding association requirements to describe the associations; wherein, the item relationship graph is used to characterize the associations between different items; and is also used to calculate the association strength between each item in the association chain and the processing item based on all the files contained in the processing instruction, and to designate items whose association strength meets preset derivative conditions as derivative items, and the corresponding association requirements as derivative requirements; wherein, the association strength includes at least: the processing completion rate of processing the item using all the files contained in the processing instruction.
[0102] Optionally, the item matching module 202 is also used to compare the keywords in the parsing results with the dataset stored in the preset structured database for each item, determine all candidate items and the similarity corresponding to each candidate item; take the candidate item with the highest similarity as the processing item and determine the corresponding processing template, and take all candidate items other than the processing item as derivative items.
[0103] Optionally, a conflict resolution module is also included to determine whether there are any conflicts between the processed item and all its corresponding derivative items. If so, a conflict resolution solution is determined based on the conflict type, and the conflict and its corresponding conflict resolution solution are added to the derivative item processing guidelines. Here, a conflict refers to a situation where the entities shared by the processed item and the derivative items are processed according to the corresponding processing content and there is a contradiction. The conflict type includes at least the following: when the processed item and the derivative items perform the corresponding processing content on the shared entities, there is a contradiction due to the inconsistency of the specific content of the shared entities.
[0104] Optionally, it also includes a content classification module, which is used to periodically identify conflict issues based on historically discovered conflict issues, and to take matters related to the conflict issues as target matters; to decompose and reorganize all the processing content contained in the processing template corresponding to the target matters, and to satisfy the following: the reorganized processing content contains conflict content, the conflict content is processing content that only contains the corresponding conflict issues, and to establish and store the conflict relationship between conflict content, conflict issues, and corresponding matters; the conflict resolution scheme includes at least prioritizing the processing content other than conflict content.
[0105] Optionally, a sensitivity matching module is also included, which is used to determine and store the matching relationship between the processing item and its corresponding key file whenever a processing item is matched. The key file refers to the file that plays a decisive role in the matching process of determining the processing item. The module periodically analyzes the sensitivity of each key file based on the matching relationship stored in the historical period, and marks key files with sensitivity higher than a preset sensitivity threshold as high-sensitivity files. A set of items is determined for each high-sensitivity file. The set of items contains items that have a matching relationship with the corresponding high-sensitivity file. When matching processing items for a processing instruction, if there is a high-sensitivity file in all the files contained in the processing instruction, the items in the set of items corresponding to the high-sensitivity file are retrieved first to perform the matching operation.
[0106] This application also discloses an adaptive knowledge graph construction and deep analysis apparatus that supports multiple file types. The apparatus includes a memory and a processor. The memory stores a computer program that can be loaded by the processor and executed as described above for the adaptive knowledge graph construction and deep analysis method that supports multiple file types.
[0107] This application also discloses a computer-readable storage medium that stores a computer program that can be loaded by a processor and executed as described above for the adaptive knowledge graph construction and deep analysis method supporting multiple file types. The computer-readable storage medium includes, for example, various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.
[0108] It should be noted that in this paper, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations.
[0109] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit the scope of protection of the application. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on these embodiments, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
Claims
1. A method for adaptive knowledge graph construction and deep analysis supporting multiple file types, characterized in that, include: Receive a processing instruction from a user, parse all files contained in the processing instruction, and the parsing operation includes at least extracting file entities; Based on the parsing results, a processing item and a corresponding processing template are matched for the processing instruction; wherein, the processing template contains processing content for processing entities in the file; Based on a preset entity alignment technique, the association relationship is determined and established among all files included in the processing instructions; Based on the determined processing template and file relationships, all processing content is executed through the preset processing engine, and the processing results are fed back to the user. The method further includes: Each time a processing item is matched, it is determined whether there is a derivative requirement that meets the preset derivative conditions. If so, the derivative item corresponding to the derivative requirement is determined, and a processing guide for the derivative item is generated and fed back to the user based on the processing template corresponding to the derivative item. The step of determining whether there is a derivative requirement that meets the preset derivative conditions, and if so, determining the derivative item corresponding to the derivative requirement, includes: By using a pre-constructed item relationship graph, it is determined whether there exists a relationship chain containing the processing item, the relationship chain also contains other items besides the processing item, and there are association relationships between adjacent items in the relationship chain, and there are corresponding association requirements for describing the association relationships; wherein, the item relationship graph is used to characterize the association relationships between different items; Based on all the files contained in the processing instruction, the association strength between each item in the relationship chain and the processing item is calculated. Items whose association strength meets the preset derivation conditions are designated as derivation items, and the corresponding association requirements are designated as derivation requirements. The association strength includes at least the processing completion rate of the item to which the processing instruction is used to process the item.
2. The adaptive knowledge graph construction and deep analysis method supporting multiple file types as described in claim 1, characterized in that, The parsing operation also includes semantic recognition of the file and extraction of keywords; The process of matching processing items and corresponding processing templates for the processing instructions based on the parsing results includes: Based on the dataset corresponding to each item stored in the pre-set structured database, the keywords in the parsing results are compared with the dataset to determine all candidate items and the similarity of each candidate item. The candidate items with the highest similarity are selected as the processing items and their corresponding processing templates are determined. All candidate items other than the processing items are selected as derivative items.
3. The adaptive knowledge graph construction and deep analysis method supporting multiple file types as described in claim 2, characterized in that, The method further includes: Determine whether there are any conflicts between the processing item and all its corresponding derivative items. If so, determine the conflict resolution solution corresponding to the conflict type based on the conflict issue, and add the conflict issue and its corresponding conflict resolution solution to the processing guidelines for the derivative items. The conflict issue refers to a situation where the entities shared by the processing matter and the derived matters are processed according to the corresponding processing content, resulting in a contradiction. The conflict types include at least the following: when processing matters and derived matters perform corresponding processing on a common entity, they contradict each other because the specific content of the common entity is inconsistent.
4. The adaptive knowledge graph construction and deep analysis method supporting multiple file types as described in claim 2, characterized in that, The method further includes: Periodically, based on historically identified conflict issues, identify relevant matters as target matters. The processing template corresponding to the target item is decomposed and recombined to satisfy the following condition: there are conflicting contents in the recombined processing contents, and the conflicting contents are processing contents that only contain the corresponding conflicting issues. The conflicting contents, conflicting issues, and conflicting relationships between the corresponding items are established and stored. The conflict resolution scheme includes at least prioritizing the handling of content other than the conflicting content.
5. The adaptive knowledge graph construction and deep analysis method supporting multiple file types according to claim 1, characterized in that, The method further includes: Whenever a processing item is matched, the matching relationship between the processing item and its corresponding key file is determined and stored. The key file refers to the file that plays a decisive role in the matching process of determining the processing item. Based on the matching relationships stored in historical periods, the sensitivity of each key document is analyzed periodically, and key documents with sensitivity higher than a preset sensitivity threshold are marked as high-sensitivity documents. A set of items is determined for each high-sensitivity document, and the set of items contains items that have a matching relationship with the corresponding high-sensitivity document. When matching processing items for a processing instruction, if a highly sensitive file exists in all the files contained in the processing instruction, the items in the item set corresponding to the highly sensitive file will be retrieved first to perform the matching operation.
6. An adaptive knowledge graph construction and deep analysis system supporting multiple file types, applied to the adaptive knowledge graph construction and deep analysis method supporting multiple file types as described in claim 1, characterized in that, The system includes: The file parsing module (201) is used to receive processing instructions from the user and parse all files contained in the processing instructions. The parsing operation includes at least extracting file entities. The item matching module (202) is used to match the processing item and the corresponding processing template for the processing instruction based on the parsing result; wherein the processing template contains processing content for processing entities in the file; The cross-file association module (203) is used to determine and establish association relationships among all files included in the processing instruction based on a preset entity alignment technology; The task processing module (204) is used to execute all processing content through a preset processing engine based on the determined processing template and the relationship between files, and to provide feedback on the processing results to the user.
7. An adaptive knowledge graph construction and deep analysis device supporting multiple file types, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and executed as described in any one of claims 1 to 5.
8. A computer-readable storage medium, characterized in that, The computer program is stored that can be loaded by a processor and executed as described in any one of claims 1 to 5.