Data retrieval method, system and equipment

Abstract

The invention provides a data retrieval method, system and device, and relates to the technical field of big data processing, the method comprises the following steps: obtaining a search text of a user; performing data permission assembly according to the search text to obtain a user permission set of the user; according to the user permission set, generating a retrieval filtering condition of the search text; according to the retrieval filtering condition, performing multi-source data search in combination with the search text to obtain an initial retrieval result corresponding to the search text; and performing unauthorized rejection operation on the initial retrieval result to obtain a final retrieval result of the search text. According to the method, the range is defined through permission assembly, the condition focusing target is filtered, the efficiency is improved through pre-matching, and the precision is guaranteed through post-removing, so that the retrieval data volume is compressed to improve the efficiency, and the matching pertinence is enhanced to improve the precision.

Description

Technical Field

[0001] This invention relates to the field of big data processing technology, and more specifically, to a data retrieval method, system, and device. Background Technology

[0002] Currently, with the acceleration of enterprise digital transformation, data sources are becoming increasingly diversified, including databases, documents, and web pages. To meet enterprises' data retrieval needs, traditional retrieval technologies mainly extract, transform, and load structured data, and then use keyword matching to achieve basic information queries.

[0003] In related technologies, due to the current diversity of data, multi-source heterogeneous data varies significantly in format, transmission protocol, and update frequency, with the proportion of unstructured content rapidly increasing. Because of the massive volume of multi-source heterogeneous data, ETL tools designed for static models need to perform keyword matching sequentially across the entire range of all multi-source heterogeneous data. This results in low accuracy, long matching times, and difficulty in handling dynamic changes in multi-source heterogeneous data, leading to underutilization of data resources and hindering efficient retrieval based on multi-source heterogeneous data. Summary of the Invention

[0004] The problem addressed by this invention is how to improve the retrieval accuracy and efficiency of multi-source heterogeneous data.

[0005] To address the above problems, this invention provides a data retrieval method, system, and device.

[0006] In a first aspect, the present invention provides a data retrieval method comprising: Get the user's search text; Data permissions are assembled based on the search text to obtain the user's user permission set; Based on the user permission set, generate the search filtering conditions for the search text; Based on the search filtering conditions, a multi-source data search is performed in conjunction with the search text to obtain the initial search results corresponding to the search text. The initial search results are subjected to an unweighted removal operation to obtain the final search results for the search text.

[0007] Optionally, the step of assembling data permissions based on the search text to obtain the user's user permission set includes: Based on the user identity information associated with the search text, determine the permission index corresponding to the user; Extracting from the permission index yields multiple permission information matching the user; Based on all the aforementioned permission information, the user permission set is generated.

[0008] Optionally, generating the search filtering conditions for the search text based on the user permission set includes: By using the search service, multiple initial filtering rules that match the query syntax of the ES search engine are obtained by mapping the permission information in the user permission set. The permission metadata in the business data index of the ES retrieval engine is used to perform rule adaptation verification on each of the initial filtering rules to obtain multiple final filtering rules. The final filtering rules are aggregated using a pre-approval permission filter to obtain the search filtering conditions.

[0009] Optionally, the step of performing a multi-source data search based on the search filtering conditions and the search text to obtain the initial search results corresponding to the search text includes: Based on the search text, data stored in other databases is crawled to obtain non-ES retrieved data and unstored data not stored in the ES database; The target ES data is obtained by crawling the data according to the search filtering conditions through the ES search engine. Based on the non-ES retrieval data and the ES database, combined with the target ES data, a target retrieval dataset is obtained; Based on the search filtering conditions and the search text, semantic similarity matching is performed on the target search dataset to obtain the initial search results.

[0010] Optionally, the other databases include independent databases of other search engines and core databases that are not connected to the ES search engine; The step of fetching data from other databases based on the search text to obtain non-ES search data and unstored data not stored in the ES database includes: The independent database and the core database are matched and verified according to the user permission set to determine the data access scope corresponding to the independent database and the core database respectively; The API interface corresponding to the independent database is determined based on the data access scope of the independent database; By calling the API interface, the non-ES retrieved data is obtained by searching the independent database; Based on the user permission set and the data access scope of the core database, a search is performed to obtain document-type data and non-document-type data that are not connected to the ES database; The document-type data and the non-document-type data are parsed and permission tags are embedded to obtain the unstored data.

[0011] Optionally, obtaining the target retrieval dataset by combining the non-ES retrieval data and the ES database with the target ES data includes: Using the target ES data storage structure as a benchmark, the non-ES retrieved data and the unstored data are adapted and assembled in ES format to obtain a standardized set of permission metadata. The standardized permission metadata set is deduplicated and merged, and then combined with the target ES data to obtain the target retrieval dataset.

[0012] Optionally, the step of performing semantic similarity matching on the target retrieval dataset based on the retrieval filtering conditions and the search text to obtain the initial retrieval results includes: The search filtering conditions are converted into query statements for the ES search engine; The query statement is used to search the target retrieval dataset to obtain multiple retrieval data. The semantics of the search text and the retrieved data are extracted using an NLP model to obtain the semantic vectors of the search text and the retrieved data. A similarity matching algorithm is used to match the semantic vector of the search text with the semantic vector of the retrieved data to obtain a semantic similarity score between the search text and each retrieved data. The search data is filtered based on the semantic similarity score to obtain the initial search results.

[0013] Optionally, the step of performing an unweighted removal operation on the initial search results to obtain the final search results for the search text includes: The post-permission filter performs real-time verification on each piece of search data in the initial search results to determine whether the data source, organizational scope, and sensitive fields of the search data are all within the user permission set. If yes, the search data verification is deemed successful; otherwise, the search data verification is deemed unsuccessful. The search data that fails the verification is removed, and the remaining search data is anonymized to generate the final search results.

[0014] Secondly, a data retrieval system according to the present invention includes: The text acquisition module is used to acquire the user's search text; The permission assembly module is used to assemble data permissions based on the search text to obtain the user's user permission set. The filter condition generation module is used to generate retrieval filter conditions for the search text based on the user permission set. The multi-source search module is used to perform multi-source data search based on the search filtering conditions and the search text to obtain the initial search results corresponding to the search text. The unauthorized removal module is used to perform unauthorized removal operations on the initial search results to obtain the final search results for the search text.

[0015] Thirdly, the electronic device of the present invention includes a memory and a processor; The memory is used to store computer programs; The processor is configured to implement the data retrieval method described above when executing the computer program.

[0016] The data retrieval method, system, and device of this invention significantly improve the accuracy and efficiency of retrieval of multi-source heterogeneous data through pre-emptive permission control, precise scope retrieval, and compliant result purification. Specifically, data permissions are assembled based on the search text, deeply binding user identity and permission configuration to generate a user permission set containing rules such as data source access scope and institutional permission boundaries. By pre-defining the boundaries of heterogeneous data accessible to users (such as specific database tables, file types, or system data sources) through the user permission set, invalid matching of unauthorized data is avoided, laying the foundation for subsequent precise retrieval. After generating retrieval filtering conditions based on the permission set, the scope is locked, ensuring that the system only performs matching operations on heterogeneous data (whether structured database records or unstructured document content) within the permission range. This significantly reduces the total number of retrieval objects and the computational cost of data traversal and matching, achieving efficiency improvement from the retrieval entry point. Simultaneously, because the filtering conditions are generated based on explicit permission rules, the focus of the retrieval scope is significantly enhanced, avoiding matching deviations caused by mixed data sources and interference from irrelevant data. When conducting multi-source data searches, by integrating data from different databases and combining the dual constraints of search text and retrieval filtering conditions, feature extraction and matching can be specifically performed on heterogeneous data within the permission scope. By applying filtering conditions upfront, the matching algorithm only operates on valid datasets, reducing computational load, improving matching speed, and allowing the algorithm to concentrate resources on deeper semantic analysis of more relevant data. This achieves comprehensive retrieval of multi-source heterogeneous data, expanding the scope of the retrieved data and further improving the comprehensiveness and accuracy of the retrieval. Finally, a secondary purification process is performed to remove potentially missed unauthorized data or sensitive information from the initial results, ensuring that the final returned results comply with permission specifications while retaining highly relevant content to the greatest extent possible, avoiding accuracy loss due to missing permission checks. In summary, this invention, through permission assembly to define the scope, filtering conditions to focus on targets, upfront matching to improve efficiency, and post-removal to ensure accuracy, compresses the amount of retrieved data to improve efficiency and strengthens the targeting of matching to improve accuracy. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating a data retrieval method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the data processing flow of a data retrieval method in one embodiment of the present invention; Figure 3 This is one of the timing diagrams of the data retrieval process in one embodiment of the present invention; Figure 4 This is a second timing diagram of the data retrieval process in one embodiment of the present invention; Figure 5 This is a schematic diagram of the structure of a data retrieval system in another embodiment of the present invention. Detailed Implementation

[0018] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Although some embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the present invention. It should be understood that the accompanying drawings and embodiments of the present invention are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.

[0019] It should be understood that the various steps described in the method embodiments of the present invention may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present invention is not limited in this respect.

[0020] The term "comprising" and its variations as used herein are open-ended, meaning "including but not limited to"; the term "based on" means "at least partially based on"; the term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments"; and the term "optionally" means "optional embodiments". Definitions of other terms will be given in the following description. It should be noted that the concepts of "first," "second," etc., mentioned in this invention are used only to distinguish different devices, modules, or units, and are not intended to limit the order of functions performed by these devices, modules, or units or their interdependencies.

[0021] It should be noted that the terms "a" and "a plurality of" used in this invention are illustrative rather than restrictive. Those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0022] The names of the messages or information exchanged between the multiple devices in the embodiments of the present invention are for illustrative purposes only and are not intended to limit the scope of these messages or information.

[0023] Combination Figure 1 As shown in the figure, an embodiment of the present invention provides a data retrieval method, including: Get the user's search text.

[0024] Specifically, acquiring users' search text allows for the accurate capture of their information needs, providing the foundational input for subsequent permission adaptation, scope filtering, and semantic matching. In multi-source heterogeneous data scenarios, user search text can encompass various forms such as keywords and fuzzy queries. The needs it carries may point to structured database data or involve unstructured documents and web page content. Only by accurately acquiring the search text can we ensure that subsequent permission assembly and retrieval filtering revolve around the user's true needs, avoiding deviations in search results caused by biases in demand capture.

[0025] Data permissions are assembled based on the search text to obtain the user's user permission set.

[0026] Specifically, the permission index (e.g., permission_index) stores the relationships between users, roles, and permissions. This includes information such as user ID, assigned role, allowed data sources (e.g., KMS, BPM), organizational scope (e.g., dept_01), and a list of sensitive fields (e.g., customer name, customer contact information). Scheduled tasks synchronize changes to the enterprise's permission system to ensure real-time permission data. Data permissions are assembled based on the search text. Key information such as the user's corresponding role, allowed data sources, organizational scope, and sensitive field restrictions are extracted from the independently maintained permission index and ultimately integrated into a user permission set. By clearly defining user permission boundaries in advance, precise criteria are provided for limiting the scope of subsequent searches, avoiding invalid processing of unauthorized data and laying the foundation for data security management.

[0027] Based on the user permission set, the search filtering conditions for the search text are generated.

[0028] Specifically, search filtering conditions are generated based on the user permission set, and permission rules are transformed into executable search instructions, enabling pre-emptive intervention of permission control in the search process. In this embodiment, the user permission set includes rules such as data source scope, organization code, and sensitive fields. These rules are transformed into filtering conditions adapted to the search engine through specific mapping and aggregation logic, ensuring that the search engine clearly defines data access boundaries before executing the search. For example, by transforming the permission rule "Allow access to department A data source" into corresponding search filtering logic, the search scope can be directly limited to cover only multi-source data related to department A. This fundamentally reduces the amount of searched data, decreases invalid matches, and defines a safe and efficient scope for subsequent multi-source data searches.

[0029] Based on the search filtering conditions, a multi-source data search is performed in conjunction with the search text to obtain the initial search results corresponding to the search text.

[0030] Specifically, multi-source data searches are performed based on search filters and search text, balancing data access security with information relevance. Search filters pre-define the boundaries of multi-source data that users are authorized to access, ensuring searches are only performed within compliant limits. By integrating data from different databases, such as Elasticsearch and other databases, comprehensive retrieval of heterogeneous multi-source data is achieved, expanding the scope of the search. The search text serves as the core requirement, extracting and matching features from heterogeneous data within this scope (such as database records and document content). This approach avoids interference from unauthorized data while ensuring the search focuses on the user's actual needs. The final initial search results comply with access restrictions and possess preliminary relevance to the user's needs.

[0031] The initial search results are subjected to an unweighted removal operation to obtain the final search results for the search text.

[0032] Specifically, the unauthorized removal operation on the initial search results aims to achieve result-level purification to ensure data security. Although the pre-selective search filtering conditions have limited the search scope, in scenarios with dynamic updates of multi-source data and complex nested permission rules, a small amount of unauthorized data or sensitive information may be mixed into the initial search results. The unauthorized removal operation verifies each piece of data in the initial search results in real time, checking whether its data source, organizational scope, and sensitive fields are entirely within the user's permission set, removing non-compliant data, and anonymizing the remaining data. This compensates for potential oversights in the pre-selective filtering, ensuring that the final output search results not only meet the required relevance but also fully comply with permission control requirements, eliminating the risk of sensitive data leakage.

[0033] The data retrieval method of this invention significantly improves the accuracy and efficiency of retrieval of multi-source heterogeneous data through pre-emptive permission control, precise scope retrieval, and compliant result purification. Specifically, data permissions are assembled based on the search text, deeply binding user identity and permission configuration to generate a user permission set containing rules such as data source access scope and institutional permission boundaries. By pre-defining the boundaries of heterogeneous data accessible to users (such as specific database tables, file types, or system data sources) through the user permission set, invalid matching of unauthorized data is avoided, laying the foundation for subsequent accurate retrieval. After generating retrieval filtering conditions based on the permission set, the scope is locked, ensuring that the system only performs matching operations on heterogeneous data (whether structured database records or unstructured document content) within the permission range. This significantly reduces the total number of retrieval objects and the computational cost of data traversal and matching, achieving efficiency improvement from the retrieval entry point. At the same time, since the filtering conditions are generated based on explicit permission rules, the focus of the retrieval scope is significantly enhanced, avoiding matching deviations caused by mixed data sources and interference from irrelevant data. When conducting multi-source data searches, by integrating data from different databases and combining the dual constraints of search text and retrieval filtering conditions, feature extraction and matching can be specifically performed on heterogeneous data within the permission scope. By applying filtering conditions upfront, the matching algorithm only operates on valid datasets, reducing computational load, improving matching speed, and allowing the algorithm to concentrate resources on deeper semantic analysis of more relevant data. This achieves comprehensive retrieval of multi-source heterogeneous data, expanding the scope of the retrieved data and further improving the comprehensiveness and accuracy of the retrieval. Finally, a secondary purification process is performed to remove potentially missed unauthorized data or sensitive information from the initial results, ensuring that the final returned results comply with permission specifications while retaining highly relevant content to the greatest extent possible, avoiding accuracy loss due to missing permission checks. In summary, this invention, through permission assembly to define the scope, filtering conditions to focus on targets, upfront matching to improve efficiency, and post-removal to ensure accuracy, compresses the amount of retrieved data to improve efficiency and strengthens the targeting of matching to improve accuracy.

[0034] Optionally, the step of assembling data permissions based on the search text to obtain the user's user permission set includes: Based on the user identity information associated with the search text, determine the permission index corresponding to the user; Extracting from the permission index yields multiple permission information matching the user; Based on all the aforementioned permission information, the user permission set is generated.

[0035] Specifically, by assembling data permissions based on search text, a user's set of permissions is obtained, enabling fine-grained management of user search permissions. First, based on the user's identity information associated with the search text, a corresponding permission index is determined to ensure accurate user identification and access to the user's corresponding permission configuration. Then, multiple permission information matching the user is extracted based on the permission index. This permission information covers the user's access range, data type, and specific data attributes. Finally, this permission information is integrated to generate a user permission set, providing a foundation for subsequent search filtering conditions. This process not only ensures the security of data retrieval but also improves the efficiency and accuracy of retrieval through precise matching of permission indexes. In this way, user permissions can be effectively managed, ensuring that only authorized users can access specific data, thereby preventing the risks of data leakage and unauthorized access.

[0036] In a preferred embodiment of the invention, user A needs to retrieve data related to financial reports. The system first determines user A's corresponding permission index based on their login information. This index points to user A's permission configuration within the system. Based on the permission index, it is determined that user A has permission to access data generated by the finance department and can only view the financial reports for the current year. This permission information is integrated to generate a permission set for user A, which clearly indicates the range and conditions of data that user A can access. In subsequent retrieval processes, the search results are filtered based on this permission set to ensure that user A can only see financial report data that matches their permissions, thereby achieving accurate retrieval and secure access to multi-source heterogeneous data.

[0037] In this embodiment of the invention, by associating user identity information with the permission index, the specific permission configuration of a user can be quickly located, reducing the time overhead of permission query and improving the system's response speed. Secondly, the extraction and integration process of permission information ensures the accuracy and completeness of the user permission set, enabling retrieval filtering based on the user's actual permissions and avoiding invalid searches of unauthorized data, thereby significantly improving retrieval efficiency.

[0038] Optionally, generating the search filtering conditions for the search text based on the user permission set includes: By using the search service, multiple initial filtering rules that match the query syntax of the ES search engine are obtained by mapping the permission information in the user permission set. The permission metadata in the business data index of the ES retrieval engine is used to perform rule adaptation verification on each of the initial filtering rules to obtain multiple final filtering rules. The final filtering rules are aggregated using a pre-approval permission filter to obtain the search filtering conditions.

[0039] Specifically, firstly, the initial filtering rules are mapped through the search service and transformed into query syntax fragments recognizable by the Elasticsearch engine, such as DSL filtering statements. When the business system triggers a data sensitivity level adjustment operation (e.g., upgrading a type of data from ordinary to confidential), the sensitivity level of the data in the data_index is updated synchronously. When the business system triggers a user permission level adjustment operation for this data (e.g., upgrading a type of data from accessible to inaccessible), the system synchronously updates the user permission mapping for accessing the sensitive data in the permission_index. Users then query the constructed DSL and access the data through the latest DSL. This establishes a mapping relationship between the business permission language and the engine query language, ensuring that permission constraints can be effectively interpreted by the search system. Secondly, based on the permission metadata of the business data index (such as data_source associated data source, menu_code menu code, dep_code department code, post_code job code, etc.), adaptation and verification are performed. By verifying the matching of the initial filtering rules with this metadata, such as whether the rule "allow access to dep_03" is consistent with the dep_code (department code) in the data metadata, invalid filtering conditions caused by ambiguous rules or missing metadata can be eliminated, avoiding the problem of rules failing to apply to the data. Finally, the final filtering rules are aggregated through the pre-permission filter, realizing the systematic integration of permission constraints. The scattered single-dimensional rules (such as data source rules, organization rules) are aggregated into composite filtering conditions covering multiple dimensions, ensuring that the search engine can lock the scope of compliant data based on the aggregated conditions before executing the search. In a preferred embodiment of the present invention, user A needs to retrieve data related to financial reports. First, according to the permission information in user A's permission set, the initial filtering rules are mapped to ES query syntax through the search service. For example, user A has permission to access data generated by the finance department and can only view the financial report for this year. These permission information are converted into ES query syntax to generate initial filtering rules. Subsequently, these initial filtering rules are adapted and verified using the permission metadata in the business data index of the ES search engine. Finally, the verified final filtering rules are aggregated through the pre-permission filter to form complete search filtering conditions.

[0040] In this embodiment of the invention, by mapping user permission information to initial filtering rules in Elasticsearch query syntax, search conditions matching user permissions are quickly generated, reducing permission verification time during the search process and improving system response speed. Secondly, the initial filtering rules are adapted and validated using permission metadata to ensure the accuracy and consistency of search conditions, avoiding search errors caused by inconsistent permission information. Finally, the final filtering rules are aggregated through a pre-filter, further optimizing the search conditions and improving search efficiency.

[0041] Optionally, the step of performing a multi-source data search based on the search filtering conditions and the search text to obtain the initial search results corresponding to the search text includes: Based on the search text, data stored in other databases is crawled to obtain non-ES retrieved data and unstored data not stored in the ES database; The target ES data is obtained by crawling the data according to the search filtering conditions through the ES search engine. Based on the non-ES retrieval data and the ES database, combined with the target ES data, a target retrieval dataset is obtained; Based on the search filtering conditions and the search text, semantic similarity matching is performed on the target search dataset to obtain the initial search results.

[0042] Specifically, firstly, cross-database data crawling is conducted based on the search text, covering non-ES search data not managed by the ES search engine (such as API output data from independent business systems) and unstored data (such as raw documents and semi-structured forms from the core database), breaking through the search boundaries of a single ES search engine and achieving comprehensive coverage of multi-source data. Secondly, relying on the ES search engine and combining search filtering conditions, the target ES data is accurately located, and the efficient search capabilities of the ES search engine are used to quickly filter out basic data that meets the permission and content characteristics. Next, based on the data storage structure of the ES database, the non-ES search data and unstored data are format adapted and assembled. Through operations such as field mapping, data type conversion, and metadata supplementation, heterogeneous data is transformed into a standardized permission metadata set consistent with the ES database. This set is then deduplicated and merged with the target ES data to form a target search dataset with a unified structure and clear permissions, solving the problem of heterogeneous formats of multi-source data. Finally, semantic similarity matching is performed again based on search filtering conditions and search text, transforming the search text and the data in the dataset into high-dimensional semantic vectors. The degree of semantic association is quantified through similarity algorithms (such as cosine similarity), and highly semantically related data is filtered out to obtain the initial search results. Combination Figure 2As shown, after the user enters the search text, the data scraping process is initiated first, targeting the Knowledge Management System (KMS) and Business Process Management (BPM) databases not stored in the Elasticsearch (ES) database. KMS stores enterprise documents, knowledge bases, and other data, and provides data to the ES database via API interfaces; BPM stores enterprise process-related data (such as documents, users, and basic information), which is synchronized to the ES database via the Logstash pipeline. After the user enters the search text, on the one hand, for BPM and other search engine KMS data not stored in ES, data permission assembly is performed to form a user permission set; simultaneously, if the search involves documents, documents, reports, forms, personnel, etc., it first determines whether it is a document. If it is a document, it then determines whether text parsing is required. If text parsing is required, it proceeds to the text parsing stage; otherwise, if it is not a document or not required, it obtains HTML text, database attributes, links, and other data through the data scraping service, then performs ES data assembly, and stores the assembled data in the ES database (which includes OA systems, etc.). Next, the search service combines the user's permission set with the retrieval data provided by the ES database, and uses similarity matching algorithms and permission filters to perform retrieval processing based on the search text and query conditions. At the same time, the ES database will also provide retrieval data directly based on the search text, together completing the entire search process.

[0043] In a preferred embodiment of the invention, incremental synchronization to the ES database is performed periodically via an API interface. During the synchronization process, the Sync Service can call the enterprise permission system's interface to obtain the data ownership roles (such as "department administrator" or "project member") corresponding to the batch of data, and embed permission metadata when the data is written to the ES database to ensure the initial association between data and access permissions. For the BPM system: data extraction, transformation, and loading are performed using a Logstash pipeline. A new permission metadata injection step is added to the Logstash pipeline: when connecting to the BPM database via JDBC, the menu code (menu_code), restriction type, and permission point fields (postCode, depCode, etc.) to which the data belongs are synchronously queried and mapped to ES documents. For the OA system: the existing OA ES database is reused and associated with a newly created ES database. The Sync Service periodically synchronizes the user-organization relationship table from the OA system and adds it to the user_org_mapping index of the OA's ES retrieval engine, providing an organizational structure basis for subsequent permission verification.

[0044] The Sync Service, as the core coordination component, is responsible not only for synchronizing business data with the Elasticsearch database but also for the coordinated updating of permission data and business data, ensuring consistency between the two. Specifically, when a user's role or organizational permissions change in the permission index (e.g., a user is promoted to department manager), the Sync Service automatically triggers a permission reverse process. This involves querying business data that the user previously had no access to but now has access to (e.g., BPM documents from the original department), batch updating the tags of this data, and adding the user's ID, reducing the pressure of permission calculations during subsequent retrievals. If the sensitivity level of a certain type of data in the BPM system is upgraded, the Logstash pipeline triggers the Sync Service's permission verification interface during synchronization, automatically filtering out roles with access permissions from the permission_index (e.g., role_admin) and updating the data's roles to these roles, preventing a disconnect between permissions and data sensitivity. Before business data is written to the Elasticsearch database, the permission pre-screening module of the Logstash pipeline / API gateway filters data that exceeds the current data source's access permissions (e.g., sensitive documents uploaded by ordinary users are directly blocked), reducing Elasticsearch storage redundancy and the cost of permission filtering during subsequent retrievals. An intermediate index, `permission_business_link`, is constructed to record the relationship between permission changes and business data (e.g., "User A's role change affects access permissions for data X / Y / Z"). When a user queries, the application service can directly locate the data requiring permission re-verification through this index, avoiding a full scan and improving retrieval response speed in complex permission scenarios. Simultaneously, role inheritance relationships are introduced into the permission index (e.g., `role_dept_manager` inherits permissions from `role_dept_member`). When synchronizing to the Elasticsearch database, dynamic permission calculation is achieved through nested queries, solving the problem that traditional static permission tags cannot handle changes in role hierarchy.

[0045] By calling the KMS API to obtain relevant document-type data as non-ES search data, and after authorization verification, accessing the BPM core database to extract data not connected to the ES search engine as unstored data. Then, based on the ES database's document structure and metadata specifications, the two types of data were formatted and field adapted. After deduplication and merging, they were integrated with the R&D-related data already stored in the ES search engine to form the target search dataset. Next, pre-generated search filters were applied to the target dataset to filter out compliant data. Then, a Natural Language Processing (NLP) model was used to extract semantic vectors between the search text and the compliant data. A similarity algorithm was used to calculate matching scores, and finally, multiple data entries with scores higher than a threshold were selected as the initial search results.

[0046] It is worth mentioning that other databases are managed separately by the XXL-JOB crawler framework, integrated with business data as two parallel organization categories to achieve accurate acquisition of permission information. In another embodiment of the invention, the permission system API interface is triggered periodically (every 30 minutes) by XXL-JOB to incrementally synchronize data such as user role changes and organization permission adjustments, ensuring that the timeliness of permission information matches that of business data.

[0047] In this embodiment of the invention, the comprehensiveness, accuracy, and adaptability of search results are significantly improved through systematic multi-source data integration and intelligent matching design. Specifically, the multi-source data crawling mechanism breaks through the limitations of the ES database, incorporating data scattered across various external databases into the search scope, solving the problem of one-sided results caused by data silos in traditional retrieval, and improving the comprehensiveness of search results. The combination of search filtering conditions and semantic similarity matching achieves dual control over permission compliance and demand matching. The filtering conditions exclude unauthorized data in advance, reducing invalid matching operations, while semantic matching improves the relevance of results through deep understanding of textual connotation. Compared with traditional keyword retrieval, it greatly reduces matching deviations caused by semantic ambiguity, and significantly improves the accuracy and practicality of the initial search results.

[0048] Optionally, the other databases include independent databases of other search engines and core databases that are not connected to the ES search engine; The step of fetching data from other databases based on the search text to obtain non-ES search data and unstored data not stored in the ES database includes: The independent database and the core database are matched and verified according to the user permission set to determine the data access scope corresponding to the independent database and the core database respectively; The API interface corresponding to the independent database is determined based on the data access scope of the independent database; By calling the API interface, the non-ES retrieved data is obtained by searching the independent database; Based on the user permission set and the data access scope of the core database, a search is performed to obtain document-type data and non-document-type data that are not connected to the ES database; The document-type data and the non-document-type data are parsed and permission tags are embedded to obtain the unstored data.

[0049] Specifically, firstly, using the user permission set as the verification basis, matching operations are performed separately for two different types of resources: independent databases and core databases. By parsing the rules such as data source access permissions and organizational scope restrictions contained in the permission set, the data boundaries accessible to users in the two types of databases are accurately defined. This achieves pre-binding of permission control and data source access, preventing invalid crawling of unauthorized data from the source. In the data acquisition stage of independent databases, the corresponding API interface is located based on the determined data access scope. The standardized interaction characteristics of the API interface are used to achieve targeted data crawling. This not only adapts to the technical characteristics of independent databases (such as knowledge management systems and specific business platforms) that provide data externally through API interfaces, but also ensures the standardization and efficiency of data acquisition. The acquired data is non-ES search data. For the core database, direct search is conducted under the dual constraints of permission set and data access scope. The acquired data is divided into document and non-document categories. Text and metadata are extracted using document parsing tools, and the format of non-document data is converted. Permission tags containing information such as data source identifier and organization code are uniformly embedded, ultimately forming unstored data. This process not only solves the parsing problem of heterogeneous data in the core database, but also provides data-level identity identification for permission verification in subsequent retrieval stages, ensuring that the crawled data can be identified by the full-process permission control system.

[0050] Combination Figure 3As shown, utilizing the timed functionality of the crawler scheduling framework, the Sync Service periodically queries the incremental query interface of KMS / SEP to achieve timed data retrieval, ensuring data timeliness and dynamic updates. First, the Sync Service proactively initiates a query for incremental data from KMS / SEP. KMS / SEP receives the query request, processes it, and returns the corresponding data. Subsequently, the Sync Service writes the retrieved data to the Elasticsearch search engine, utilizing Elasticsearch's data storage capabilities. Elasticsearch returns the operation results after the data is written, ensuring the traceability of the data writing process. Basic data, user data, and document data in the BPM Database are written to the Elasticsearch database via Logstash, enabling the extraction, transformation, and loading of data from the BPM Database. This ensures that the data is stored in the Elasticsearch database in a suitable format, providing basic data support for subsequent retrieval functions. Related data in the BPM Database is processed by Logstash. After cleaning and transforming the data according to certain rules, Logstash writes the data to Elasticsearch, which returns the operation results, confirming successful data synchronization. The Sync Service queries the BPM Database for document data to be synchronized. When reading a single document, it uses document parsing technology (such as Tika) to parse the document into text format for storage and subsequent retrieval processing in the Elasticsearch database. Elastic Search returns the results, and this process is repeated until all document data to be synchronized has been processed. This cyclical processing method ensures that document data is completely and accurately synchronized to the Elasticsearch database, providing comprehensive document data support for intelligent retrieval.

[0051] For example, the user permission set includes the independent database allowed access as the KMS knowledge management system, the core database as the BPM business process management database, and the organizational scope limited to the marketing department. The system first matches and verifies the KMS and BPM based on this permission set, determining that the accessible data scope in KMS is the marketing department's knowledge base categories, and the accessible data scope in BPM is marketing activity-related forms and documents. For KMS, based on the data access scope corresponding to its knowledge base categories, the system calls its provided document query API interface, passing in user identity and scope parameters to retrieve 15 marketing department documents as non-ES retrieval data. For BPM, combining the permission set and data access scope, a search is performed, obtaining 8 marketing documents (document-type data) and 12 form data (non-document-type data) not integrated into ES. Subsequently, a document parsing tool is used to extract text from the planning documents, the form data is formatted and standardized, and all data is uniformly embedded with permission tags containing "data source: BPM" and "organization code: marketing department," ultimately forming 20 unstored data entries.

[0052] In this embodiment of the invention, the database type-based classification crawling mechanism adapts to the API access characteristics of independent databases and the direct retrieval needs of core databases, avoiding the limitations of a single crawling method in adapting to heterogeneous data sources, and significantly improving the coverage and efficiency of data crawling. The pre-emptive permission matching and verification ensures that data crawling is always performed within compliance limits, blocking the inflow of unauthorized data from the source of data acquisition and reducing the cost and risk of subsequent permission filtering. The parsing and permission tag embedding processing of data crawled from core databases achieves the structured transformation of non-standardized data and the association of permission attributes, giving heterogeneous data that was originally difficult for retrieval systems a unified parsing format and permission identifier. This solves the problem of fragmented formats in multi-source data and ensures that crawled data can be seamlessly integrated into subsequent permission filtering and semantic matching processes, improving the coherence and accuracy of data processing.

[0053] Optionally, obtaining the target retrieval dataset by combining the non-ES retrieval data and the ES database with the target ES data includes: Using the target ES data storage structure as a benchmark, the non-ES retrieved data and the unstored data are adapted and assembled in ES format to obtain a standardized set of permission metadata. The standardized permission metadata set is deduplicated and merged, and then combined with the target ES data to obtain the target retrieval dataset.

[0054] Specifically, the first step is to establish the data storage structure of the target Elasticsearch (ES) data as a unified adaptation standard. This benchmark covers core elements such as data field definitions, metadata specifications, and permission identifier formats. Targeted format adaptation and assembly are then carried out to address the heterogeneity of non-ES retrieved data (mostly from API outputs of independent databases from other search engines) and unstored data (mostly from document and non-document data in the core database). Through operations such as field mapping, data type conversion, and metadata supplementation, the structure and attributes of both types of data are adjusted to be consistent with the ES database. Simultaneously, it is ensured that the permission tags embedded in the data (such as data source identifiers and organization codes) conform to the ES permission metadata specifications, ultimately forming a standardized permission metadata set.

[0055] In a preferred embodiment of the present invention, the target ES data adopts a storage structure of "data identifier - content text - permission metadata (data source, organization code) - creation time". The non-ES retrieved data is 10 market analysis documents (including title, content, and department tag fields) returned by the KMS system API. The unstored data is 8 event planning forms (including form ID, field values, and affiliated organization) and 6 meeting minutes documents (including text, author, and department) parsed by the BPM system. First, the format is adapted based on the ES storage structure: the title and content of the KMS documents are merged into the content text, the department tags are mapped to the organization code, and the data identifier and creation time are added; the form ID of the BPM forms is used as the data identifier, the field values ​​are integrated into the content text, and the affiliated organization corresponds to the organization code; the text of the meeting minutes is used as the content text, and the department is converted into the organization code. All three types of data are uniformly embedded with the data source identifier (KMS or BPM), forming a standardized permission metadata set of 24 data.

[0056] In this embodiment of the invention, the format adaptation operation based on the ES storage structure solves the heterogeneity problem between non-ES retrieved data and unstored data, giving the originally fragmented data a unified structure and attribute characteristics, eliminating matching deviations caused by data format differences in subsequent retrieval processes, and improving the coherence of data processing; the standardized processing of permission metadata ensures that the permission attributes of multi-source data can be uniformly identified by the retrieval system, providing a consistent data foundation for full-process permission control.

[0057] Optionally, the step of performing semantic similarity matching on the target retrieval dataset based on the retrieval filtering conditions and the search text to obtain the initial retrieval results includes: The search filtering conditions are converted into query statements for the ES search engine; The query statement is used to search the target retrieval dataset to obtain multiple retrieval data. The semantics of the search text and the retrieved data are extracted using an NLP model to obtain the semantic vectors of the search text and the retrieved data. A similarity matching algorithm is used to match the semantic vector of the search text with the semantic vector of the retrieved data to obtain a semantic similarity score between the search text and each retrieved data. The search data is filtered based on the semantic similarity score to obtain the initial search results.

[0058] Specifically, the search filtering conditions are first transformed into query statements executable by the Elasticsearch (ES) search engine, achieving deep adaptation between permission constraints and engine search syntax. This requires precise mapping of permission metadata (such as data sources and institutional codes) to the ES query syntax, ensuring that pre-emptive permission control rules directly affect the initial screening of the target search dataset, excluding unauthorized data from the search entry point and defining the compliance range for subsequent semantic matching. After the initial data screening, an NLP model is introduced to extract semantics from the search text and the initially screened search data. Through deep analysis of text connotation, contextual relationships, and core semantics, unstructured text information is transformed into quantifiable, computationally calculable high-dimensional semantic vectors. This overcomes the limitations of traditional keyword retrieval that relies solely on literal character matching, achieving accurate capture of text semantic features. Based on the semantic vectors, similarity matching algorithms (such as cosine similarity and BERT semantic matching) are used to calculate the vector space distance, transforming the degree of semantic association into a specific similarity score, achieving quantitative evaluation of semantic matching. Finally, the search data is filtered based on the score threshold, ensuring that the initial search results focus on content highly semantically relevant to user needs. Figure 4As shown, in a preferred embodiment of the present invention, user query and login information acquisition: the user inputs keywords to query, and obtains user login information through the APPService to identify the user's identity. The APPService is the entry point for user operations, directly receiving the keyword query information input by the user. The APPService obtains relevant permission information from Elastic Search based on the user query. Multi-source data retrieval: the APPService queries Elastic Search based on keywords and the obtained permission information, and then queries OA Elastic Search based on keywords to obtain the returned results. OA Elastic Search uses Elastic Search as its underlying technology to index and build various types of data generated in OA (such as official documents, approval records, meeting minutes, schedules, etc.). When the user inputs search text, it will query different data sources according to the user's permission information. OA Elastic Search, as one of the data sources, will perform data retrieval based on the search text and permission rules. Data enters a loop for data filtering. Through a post-permission filter, data obtained from different Elastic Search data sources is filtered out, removing data that the user does not have permission to access, and then integrated and sorted to ensure that the final presented data meets the permission requirements, further strengthening data security and access control.

[0059] In a preferred embodiment of the present invention, the search results can be filtered based on the relationship between the semantic similarity score and a preset threshold, with search data whose semantic similarity score is greater than or equal to the preset threshold included in the initial search results. Specifically, the spatial distance between the semantic vector of each search data and the semantic vector of the search text is calculated using a cosine similarity algorithm, and a similarity score in the range of 0-1 is obtained, among which 15 data points have a score higher than the preset threshold of 0.8. These 15 data points are then used as the initial search results.

[0060] In this embodiment of the invention, the conversion of search filtering conditions into Elasticsearch query statements achieves a front-end integration of access control and data retrieval, avoiding invalid semantic calculations on unauthorized data. This reduces system computational load, improves retrieval efficiency, and fundamentally ensures the compliance of retrieved data, preventing interference from unauthorized data. Simultaneously, the application of NLP models and similarity matching algorithms breaks through the semantic barriers of traditional keyword retrieval, accurately identifying deep semantic relationships between search text and retrieved data. Even with differences in expression (such as synonyms or sentence variations), efficient matching can be achieved, significantly improving the recall and accuracy of search results.

[0061] Optionally, the step of performing an unweighted removal operation on the initial search results to obtain the final search results for the search text includes: The post-permission filter performs real-time verification on each piece of search data in the initial search results to determine whether the data source, organizational scope, and sensitive fields of the search data are all within the user permission set. If yes, the search data verification is deemed successful; otherwise, the search data verification is deemed unsuccessful. The search data that fails the verification is removed, and the remaining search data is anonymized to generate the final search results.

[0062] Specifically, firstly, a multi-dimensional permission check is conducted on each piece of data in the initial search results. The verification dimensions focus on the data source, organizational scope, and sensitive fields. By comparing the actual attributes of the searched data with the permitted scope in the permission set one by one, granular verification of data access compliance is achieved. Compared to the scope-based filtering based on batch rules in the pre-filtering stage, this approach can accurately capture unauthorized data caused by dynamic data updates (such as temporary adjustments to data source permissions), nested permission rules (such as blurred permission boundaries due to role inheritance), or simplified pre-filter rules (such as special fields not covered during batch filtering). In determining the verification results, the verification is considered passed only when the data source, organizational scope, and sensitive fields are all within the permission set, ensuring that there are no single-dimensional permission flaws. Finally, through a combination of removing the failed data and the remaining de-identified data, unauthorized content is completely eliminated, and sensitive information that may exist in the compliant data (such as privacy fields that are not completely masked) is further processed, resulting in a final result that meets both permission and security standards.

[0063] In this embodiment of the invention, the security, compliance, and reliability of the search results are improved through the intervention and full-process processing of a post-processing permission filter, effectively compensating for the limitations of pre-processing. The real-time single-record verification mechanism can identify and eliminate unauthorized data missed in the pre-processing stage, solving the problem of blind spots in control caused by dynamic changes in multi-source data or complex permission rules, and significantly reducing the risk of sensitive data leakage.

[0064] Combination Figure 5 As shown, an embodiment of the present invention provides a data retrieval system, comprising: The text acquisition module is used to acquire the user's search text; The permission assembly module is used to assemble data permissions based on the search text to obtain the user's user permission set. The filter condition generation module is used to generate retrieval filter conditions for the search text based on the user permission set. The multi-source search module is used to perform multi-source data search based on the search filtering conditions and the search text to obtain the initial search results corresponding to the search text. The unauthorized removal module is used to perform unauthorized removal operations on the initial search results to obtain the final search results for the search text.

[0065] The data retrieval system of the present invention has the same advantages over the prior art as the data retrieval method described above, and will not be repeated here.

[0066] An electronic device provided by an embodiment of the present invention includes a memory and a processor; The memory is used to store computer programs; The processor is configured to implement the agent registration and service invocation method described above when executing the computer program.

[0067] The electronic device of the present invention has the same advantages over the prior art as the data retrieval method described above, and will not be repeated here.

[0068] While the present invention has been disclosed above, its scope of protection is not limited thereto. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the present invention, and all such changes and modifications will fall within the scope of protection of the present invention.

Claims

1. A data retrieval method, characterized by, The method comprises the following steps: acquiring a search text of a user; performing data permission assembly according to the search text to obtain a user permission set of the user; generating a retrieval filtering condition of the search text according to the user permission set; performing multi-source data search according to the retrieval filtering condition and the search text to obtain an initial retrieval result corresponding to the search text; performing a permission-free elimination operation on the initial retrieval result to obtain a final retrieval result of the search text.

2. The data retrieval method of claim 1, wherein, The step of performing data permission assembly according to the search text to obtain the user permission set of the user comprises the following steps: determining a permission index corresponding to the user according to user identity information associated with the search text; extracting a plurality of permission information matched with the user according to the permission index; generating the user permission set according to all the permission information.

3. The data retrieval method of claim 1, wherein, The step of generating the retrieval filtering condition of the search text according to the user permission set comprises the following steps: mapping the permission information in the user permission set to obtain a plurality of initial filtering rules matched with the query syntax of an ES retrieval engine through a search service; performing rule adaptation verification on each of the initial filtering rules through the permission metadata in the business data index of the ES retrieval engine to obtain a plurality of final filtering rules; aggregating the final filtering rules through a pre- permission filter to obtain the retrieval filtering condition.

4. The data retrieval method of claim 3, wherein, The step of performing multi-source data search according to the retrieval filtering condition and the search text to obtain the initial retrieval result corresponding to the search text comprises the following steps: performing data crawling on the data stored in other databases according to the search text to obtain non-ES retrieval data and non-stored data not stored in the ES database; performing crawling according to the retrieval filtering condition through the ES retrieval engine to obtain target ES data; combining the target ES data according to the non-ES retrieval data and the ES database to obtain a target retrieval data set; performing semantic similarity matching on the target retrieval data set according to the retrieval filtering condition and the search text to obtain the initial retrieval result.

5. The data retrieval method of claim 4, wherein, The other databases comprise independent databases of other search engines and core databases not connected to the ES retrieval engine. The step of performing data crawling on the data stored in other databases according to the search text to obtain non-ES retrieval data and non-stored data not stored in the ES database comprises the following steps: performing matching verification on the independent databases and the core databases respectively according to the user permission set to determine the data access ranges of the independent databases and the core databases respectively; determining the API interface corresponding to the independent databases according to the data access range of the independent databases; performing search from the independent databases by calling the API interface to obtain the non-ES retrieval data; performing search on the core databases according to the user permission set and the data access range of the core databases to obtain document data and non-document data not connected to the ES database. The document type data and the non-document type data are parsed and embedded with a permission tag to obtain the non-stored data.

6. The data retrieval method of claim 4, wherein, The target ES data is stored in a structure, and the non-ES retrieval data and the non-stored data are ES format adapted and assembled based on the target ES data storage structure to obtain a standardized permission metadata set. The target ES data is stored in a structure, and the non-ES retrieval data and the non-stored data are ES format adapted and assembled based on the target ES data storage structure to obtain a standardized permission metadata set. The target ES data is stored in a structure, and the non-ES retrieval data and the non-stored data are ES format adapted and assembled based on the target ES data storage structure to obtain a standardized permission metadata set.

7. The data retrieval method of claim 4, wherein, The retrieval filtering condition is converted into a query statement of the ES retrieval engine, the target retrieval data set is searched based on the query statement to obtain a plurality of retrieval data, the search text and the retrieval data are subjected to semantic extraction through an NLP model to obtain a semantic vector of the search text and a semantic vector of the retrieval data, the semantic vector of the search text and the semantic vector of the retrieval data are matched through a similarity matching algorithm to obtain a semantic similarity score between the search text and each retrieval data, and the retrieval data is screened based on the semantic similarity score to obtain the initial retrieval result. The retrieval filtering condition is converted into a query statement of the ES retrieval engine, the target retrieval data set is searched based on the query statement to obtain a plurality of retrieval data, the search text and the retrieval data are subjected to semantic extraction through an NLP model to obtain a semantic vector of the search text and a semantic vector of the retrieval data, the semantic vector of the search text and the semantic vector of the retrieval data are matched through a similarity matching algorithm to obtain a semantic similarity score between the search text and each retrieval data, and the retrieval data is screened based on the semantic similarity score to obtain the initial retrieval result. The retrieval filtering condition is converted into a query statement of the ES retrieval engine, the target retrieval data set is searched based on the query statement to obtain a plurality of retrieval data, the search text and the retrieval data are subjected to semantic extraction through an NLP model to obtain a semantic vector of the search text and a semantic vector of the retrieval data, the semantic vector of the search text and the semantic vector of the retrieval data are matched through a similarity matching algorithm to obtain a semantic similarity score between the search text and each retrieval data, and the retrieval data is screened based on the semantic similarity score to obtain the initial retrieval result. The retrieval filtering condition is converted into a query statement of the ES retrieval engine, the target retrieval data set is searched based on the query statement to obtain a plurality of retrieval data, the search text and the retrieval data are subjected to semantic extraction through an NLP model to obtain a semantic vector of the search text and a semantic vector of the retrieval data, the semantic vector of the search text and the semantic vector of the retrieval data are matched through a similarity matching algorithm to obtain a semantic similarity score between the search text and each retrieval data, and the retrieval data is screened based on the semantic similarity score to obtain the initial retrieval result. The retrieval filtering condition is converted into a query statement of the ES retrieval engine, the target retrieval data set is searched based on the query statement to obtain a plurality of retrieval data, the search text and the retrieval data are subjected to semantic extraction through an NLP model to obtain a semantic vector of the search text and a semantic vector of the retrieval data, the semantic vector of the search text and the semantic vector of the retrieval data are matched through a similarity matching algorithm to obtain a semantic similarity score between the search text and each retrieval data, and the retrieval data is screened based on the semantic similarity score to obtain the initial retrieval result. The retrieval filtering condition is converted into a query statement of the ES retrieval engine, the target retrieval data set is searched based on the query statement to obtain a plurality of retrieval data, the search text and the retrieval data are subjected to semantic extraction through an NLP model to obtain a semantic vector of the search text and a semantic vector of the retrieval data, the semantic vector of the search text and the semantic vector of the retrieval data are matched through a similarity matching algorithm to obtain a semantic similarity score between the search text and each retrieval data, and the retrieval data is screened based on the semantic similarity score to obtain the initial retrieval result.

8. The data retrieval method of claim 1, wherein, The retrieval filtering condition is converted into a query statement of the ES retrieval engine, the target retrieval data set is searched based on the query statement to obtain a plurality of retrieval data, the search text and the retrieval data are subjected to semantic extraction through an NLP model to obtain a semantic vector of the search text and a semantic vector of the retrieval data, the semantic vector of the search text and the semantic vector of the retrieval data are matched through a similarity matching algorithm to obtain a semantic similarity score between the search text and each retrieval data, and the retrieval data is screened based on the semantic similarity score to obtain the initial retrieval result. The retrieval filtering condition is converted into a query statement of the ES retrieval engine, the target retrieval data set is searched based on the query statement to obtain a plurality of retrieval data, the search text and the retrieval data are subjected to semantic extraction through an NLP model to obtain a semantic vector of the search text and a semantic vector of the retrieval data, the semantic vector of the search text and the semantic vector of the retrieval data are matched through a similarity matching algorithm to obtain a semantic similarity score between the search text and each retrieval data, and the retrieval data is screened based on the semantic similarity score to obtain the initial retrieval result. The retrieval filtering condition is converted into a query statement of the ES retrieval engine, the target retrieval data set is searched based on the query statement to obtain a plurality of retrieval data, the search text and the retrieval data are subjected to semantic extraction through an NLP model to obtain a semantic vector of the search text and a semantic vector of the retrieval data, the semantic vector of the search text and the semantic vector of the retrieval data are matched through a similarity matching algorithm to obtain a semantic similarity score between the search text and each retrieval data, and the retrieval data is screened based on the semantic similarity score to obtain the initial retrieval result. The retrieval filtering condition is converted into a query statement of the ES retrieval engine, the target retrieval data set is searched based on the query statement to obtain a plurality of retrieval data, the search text and the retrieval data are subjected to semantic extraction through an NLP model to obtain a semantic vector of the search text and a semantic vector of the retrieval data, the semantic vector of the search text and the semantic vector of the retrieval data are matched through a similarity matching algorithm to obtain a semantic similarity score between the search text and each retrieval data, and the retrieval data is screened based on the semantic similarity score to obtain the initial retrieval result.

9. A data retrieval system characterized by The retrieval filtering condition is converted into a query statement of the ES retrieval engine, the target retrieval data set is searched based on the query statement to obtain a plurality of retrieval data, the search text and the retrieval data are subjected to semantic extraction through an NLP model to obtain a semantic vector of the search text and a semantic vector of the retrieval data, the semantic vector of the search text and the semantic vector of the retrieval data are matched through a similarity matching algorithm to obtain a semantic similarity score between the search text and each retrieval data, and the retrieval data is screened based on the semantic similarity score to obtain the initial retrieval result. The retrieval filtering condition is converted into a query statement of the ES retrieval engine, the target retrieval data set is searched based on the query statement to obtain a plurality of retrieval data, the search text and the retrieval data are subjected to semantic extraction through an NLP model to obtain a semantic vector of the search text and a semantic vector of the retrieval data, the semantic vector of the search text and the semantic vector of the retrieval data are matched through a similarity matching algorithm to obtain a semantic similarity score between the search text and each retrieval data, and the retrieval data is screened based on the semantic similarity score to obtain the initial retrieval result. The retrieval filtering condition is converted into a query statement of the ES retrieval engine, the target retrieval data set is searched based on the query statement to obtain a plurality of retrieval data, the search text and the retrieval data are subjected to semantic extraction through an NLP model to obtain a semantic vector of the search text and a semantic vector of the retrieval data, the semantic vector of the search text and the semantic vector of the retrieval data are matched through a similarity matching algorithm to obtain a semantic similarity score between the search text and each retrieval data, and the retrieval data is screened based on the semantic similarity score to obtain the initial retrieval result. The retrieval filtering condition is converted into a query statement of the ES retrieval engine, the target retrieval data set is searched based on the query statement to obtain a plurality of retrieval data, the search text and the retrieval data are subjected to semantic extraction through an NLP model to obtain a semantic vector of the search text and a semantic vector of the retrieval data, the semantic vector of the search text and the semantic vector of the retrieval data are matched through a similarity matching algorithm to obtain a semantic similarity score between the search text and each retrieval data, and the retrieval data is screened based on the semantic similarity score to obtain the initial retrieval result. The retrieval filtering condition is converted into a query statement of the ES retrieval engine, the target retrieval data set is searched based on the query statement to obtain a plurality of retrieval data, the search text and the retrieval data are subjected to semantic extraction through an NLP model to obtain a semantic vector of the search text and a semantic vector of the retrieval data, the semantic vector of the search text and the semantic vector of the retrieval data are matched through a similarity matching algorithm to obtain a semantic similarity score between the search text and each retrieval data, and ​ 10. An electronic device, comprising: ​ ​ ​

Images