An AI knowledge base directional retrieval method, device, equipment and medium

Abstract

The application discloses an AI knowledge base directional retrieval method and device, equipment and medium. It belongs to the technical field of AI application, and specifically relates to an AI knowledge base retrieval method. The method comprises the following steps: acquiring a knowledge homepage after receiving a user task, wherein the knowledge homepage is used for providing an index of an AI retrieval knowledge base; directionally calling knowledge content from the knowledge base according to the index information; and completing the task and feeding back based on the called knowledge content. The knowledge base adopts a hierarchical structure, comprising a partition, a classified knowledge base, a knowledge file and a knowledge segment, and each level has a globally unique knowledge identifier. The knowledge homepage contains knowledge entries corresponding to the knowledge content one by one, and the entries and the corresponding content have the same knowledge identifier. Directional calling is realized through an AI callable tool function, and the same identifier always returns the same result. The application makes the result of AI knowledge acquisition certain and reproducible, accurately acquires the knowledge according to the identifier without fragmentation, and effectively reduces the context pressure.

Description

Technical Field

[0001] This invention belongs to the field of AI application technology, specifically relating to a retrieval method for an AI knowledge base. Background Technology

[0002] Currently, artificial intelligence applications across various industries frequently rely on external knowledge bases to support the execution of knowledge-intensive tasks. To support AI in completing standardization work, most industries adopt external knowledge bases combined with retrieval architectures such as RAG to provide proprietary content such as business specifications and knowledge standards.

[0003] First, traditional fuzzy search results are highly random. Search effectiveness depends heavily on user or AI input suggestions, word segmentation algorithms, vector model matching accuracy, and similarity threshold configuration. Even for the same query and the same knowledge base, different expressions used in the search will result in different returned knowledge content. This makes it impossible to guarantee consistent, reproducible results for the same query across different search scenarios, leading to unstable AI output and difficulty in meeting the consistent requirements of standardized operations.

[0004] Secondly, conventional RAG retrieval is prone to knowledge fragmentation. The retrieval process typically extracts local knowledge fragments based on matching degree, failing to retrieve all related content corresponding to a specification or process. AI obtains incomplete information, which can lead to misunderstandings and results that do not meet established standards. Simultaneously, the full-scale retrieval mechanism returns a large amount of irrelevant content from the knowledge base, significantly increasing the burden on the AI's context. In complex tasks with long processes and multiple stages, this can easily trigger context overload issues in large models, limiting AI execution capabilities in complex business scenarios.

[0005] Furthermore, mainstream third-party plug-in knowledge base tools on the market face deployment and integration challenges. These tools are all standalone applications, requiring separate installation and configuration, and cannot be deeply integrated with the main system, resulting in poor data exchange and functional interoperability. Moreover, third-party knowledge bases possess independent databases, posing risks of data migration and backup difficulties, as well as internal network data leakage for enterprise private business knowledge, making them unsuitable for the needs of enterprise-wide private deployments.

[0006] Finally, traditional knowledge repositories suffer from content redundancy and maintenance inconvenience. For standard knowledge points shared by multiple business processes, existing technologies can only achieve reuse by copying content from multiple sources. This not only further increases the volume of contextual data but also requires modifying all copied content one by one when knowledge points need to be updated, resulting in low maintenance efficiency and inconsistencies. Meanwhile, traditional retrieval methods cannot limit the scope of AI's knowledge access, cannot force AI to use only specified standard content, and make it difficult to control and trace the source of AI's knowledge, hindering the standardized management of work processes.

[0007] Therefore, there is an urgent need for an AI knowledge base targeted retrieval method that provides stable search results, complete knowledge retrieval, accurate traceability, and is compatible with integrated platform deployment.

[0008] It should be noted that the above content is only used to help understand the technical solution of the present invention and does not necessarily constitute prior art. Summary of the Invention

[0009] The main objective of this invention is to provide an AI knowledge base targeted retrieval method, apparatus, device, and storage medium, which aims to solve one or more technical problems in the prior art.

[0010] To achieve the above objectives, the present invention provides an AI knowledge base targeted retrieval method, comprising: After receiving the user task, S100 obtains the knowledge homepage according to the user task. The knowledge homepage is used to provide an index for AI to retrieve the knowledge base. S200 Retrieves knowledge content from the knowledge base according to the information in the index; The S300 completes the current task based on the retrieved knowledge content and provides feedback to the user.

[0011] Optionally, the knowledge base is a structured knowledge base, including: At least one partition, each partition contains at least one category knowledge base, each category knowledge base contains at least one knowledge file, and each knowledge file contains at least one knowledge fragment; Each level of knowledge content has a globally unique knowledge identifier.

[0012] Optionally, the knowledge homepage contains knowledge entries that correspond one-to-one with the knowledge content of the knowledge base, and each knowledge entry has a globally unique knowledge identifier that is consistent with the corresponding knowledge content.

[0013] Optionally, the knowledge homepage is stored as knowledge content in the knowledge base and has a globally unique knowledge identifier.

[0014] Optionally, step S200 specifically involves the AI ​​obtaining the knowledge identifier of the required knowledge item from the knowledge homepage according to the user's task requirements, and then selectively retrieving the knowledge content corresponding to the knowledge identifier from the knowledge base.

[0015] Optionally, the targeted retrieval is implemented through AI-callable utility functions.

[0016] Optionally, the knowledge content retrieved by the AI ​​includes references to other knowledge identifiers. The AI ​​can continuously retrieve knowledge content in a targeted manner along the referenced knowledge identifiers, and obtain deeper knowledge content step by step.

[0017] Optionally, when the required knowledge content cannot be obtained through knowledge identifiers or the obtained knowledge content is insufficient, a supplementary retrieval tool is invoked to search for knowledge content from the knowledge base according to a content-matching retrieval method.

[0018] Furthermore, to achieve the above objectives, the present invention also proposes an AI knowledge base targeted retrieval device, comprising: The knowledge homepage management module is used to provide an index for AI-based knowledge base retrieval. The knowledge base management module is used to manage knowledge content stored in a hierarchical structure. The identifier-based targeted retrieval module is used to return corresponding knowledge content based on knowledge identifiers; The task processing module is used to receive user tasks and output the task execution results.

[0019] Furthermore, to achieve the above objectives, the present invention also proposes an AI knowledge base targeted retrieval device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the above-mentioned AI knowledge base targeted retrieval method.

[0020] Furthermore, to achieve the above objectives, the present invention also proposes a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described AI knowledge base targeted retrieval method.

[0021] By employing the aforementioned AI knowledge base targeted retrieval method, the beneficial technical effects of this invention are as follows: (1) The search results are certain and reproducible. AI performs precise targeted retrieval based on knowledge identifiers. For the same knowledge identifier, it always returns the same knowledge content, without relying on semantic matching or fuzzy search, thus fundamentally eliminating the problem of randomness in traditional search results.

[0022] (2) Complete and unfragmented knowledge retrieval. Knowledge tags can point to knowledge units of any granularity (partitions, category knowledge bases, knowledge files, knowledge fragments). AI retrieves complete knowledge content according to the tags, avoiding the knowledge fragmentation problem caused by traditional RAG retrieval, and ensuring that the information obtained by AI is complete and accurate.

[0023] (3) Controllable context pressure. The knowledge homepage provides an indexing mechanism, and the AI ​​retrieves knowledge content level by level as needed, loading only the information required for the current step, effectively avoiding the context overload problem caused by full retrieval.

[0024] (4) The retrieval scope is traceable. Knowledge identification accurately limits the scope of AI's knowledge access, preventing AI from using unauthorized or irrelevant knowledge content, and enabling control and traceability of the source of AI knowledge.

[0025] (5) Supports integrated deployment. This invention can be implemented using a self-developed architecture, without relying on third-party knowledge base tools, eliminating the difficulties of data migration and the risk of internal network data leakage, and adapting to the private deployment needs of enterprises. Attached Figure Description

[0026] Figure 1 This is a flowchart illustrating an AI knowledge base targeted retrieval method according to the present invention; The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0027] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0028] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below. Example 1

[0029] This invention provides an AI-driven targeted knowledge base retrieval method. Implementing this method first requires constructing a knowledge homepage and a knowledge base. There is no strict order in which the knowledge homepage and knowledge base are constructed: either the knowledge base can be constructed first, and then the knowledge homepage generated based on its structure; or the knowledge homepage can be constructed first, and then the knowledge base content filled in based on its structure; or both can be completed in parallel iteratively. The knowledge base can be derived from an existing knowledge base through structural modification, or it can be built from scratch according to business needs. This embodiment uses the example of first constructing a self-developed enterprise knowledge base, and then generating the knowledge homepage based on the structure of that knowledge base for illustration.

[0030] A knowledge base is a structured collection of knowledge. It can be understood that different knowledge bases are like different books, each serving different task scenarios, with its content and structure organized around a specific task domain. An enterprise system can run multiple knowledge bases simultaneously, each targeting different business domains.

[0031] In this embodiment, the knowledge base adopts a hierarchical storage structure. The first layer is partitions, which can be set up as enterprise knowledge base partitions, personal knowledge base partitions, system knowledge base partitions, etc. The second layer is a category knowledge base, under which one or more category knowledge bases can be created. For example, under the enterprise knowledge base partition, software design specification library, project management process library, interface design standard library, etc., can be created. The third layer is knowledge files, under which one or more knowledge files can be created under each category knowledge base. For example, under the software design specification library, data table design specification files, module division standard files, etc., can be created. The fourth layer is knowledge fragments, under which one or more knowledge fragments can be created. For example, under the data table design specification file, field naming rules knowledge fragments, data type definition standard knowledge fragments, table relationship specification knowledge fragments, etc., can be created. All knowledge content at each level has its own globally unique knowledge identifier for precise location.

[0032] Knowledge content refers to knowledge units stored at various levels in a knowledge base, including but not limited to the entire categorized knowledge base, knowledge files, and knowledge fragments. Each piece of knowledge content has a globally unique knowledge identifier, which serves as the sole credential for AI-targeted retrieval. In this application, the knowledge identifier is identification information used to uniquely point to specific knowledge content in the knowledge base. Based on this knowledge identifier, AI can accurately locate and retrieve the corresponding knowledge content. Knowledge identifiers can be implemented in various forms, including but not limited to: numeric numbers, string encoding, UUIDs, hierarchical path names, hash values, URL links, aliases, composite identifiers, and any combination thereof. Any identification information that can uniquely point to and accurately locate knowledge content falls within the scope of the knowledge identifiers described in this application. This embodiment will subsequently use knowledge IDs (i.e., knowledge identifiers in numeric or character encoding forms) as a specific implementation of knowledge identifiers for detailed explanation, but the scope of protection of this application is not limited to this.

[0033] The so-called knowledge homepage is an index of the knowledge base, much like the table of contents of a book—just as the table of contents tells the reader the name and location of each chapter, the knowledge homepage tells the AI ​​what knowledge content is included in the knowledge base and how to access each piece of content. Different tasks correspond to different knowledge bases, so the form of the knowledge homepage varies depending on the task type. In some scenarios, the index structure of the knowledge homepage itself can serve as an operation guide, allowing the AI ​​to complete the task sequentially based on the entries in the index.

[0034] The knowledge homepage contains knowledge entries that correspond one-to-one with the knowledge content in the knowledge base. Each knowledge entry has a globally unique knowledge identifier that is consistent with the corresponding knowledge content.

[0035] The following are some typical homepage formats for knowledge resources: In software development task scenarios, the knowledge homepage is presented as a sequence of steps, such as requirements collection and analysis → business module division → data table design → interface creation → permission configuration → data statistics configuration. Each step corresponds to a knowledge item and is marked with a unique knowledge identifier.

[0036] • In the scenario of querying rules and regulations, the knowledge homepage is presented as a category directory of rules and regulations, such as personnel system → financial system → administrative system → IT system. Each system corresponds to a knowledge entry and is marked with a unique knowledge identifier.

[0037] In clinical diagnostic and treatment assistance scenarios, the knowledge homepage appears as a table comparing symptoms with diagnostic and treatment guidelines. For example, the knowledge entry for "fever with cough" points to a section of fever diagnosis and treatment guidelines, and the knowledge entry for "chest pain" points to a section of emergency chest pain treatment procedures. Doctors or AI assistants can then use this information to perform actions such as consultation, examination, and medication administration. In this case, the knowledge homepage serves as both an index from symptoms to knowledge and a practical operational guide—the index format naturally varies depending on the task requirements.

[0038] • In legal and regulatory search scenarios, the knowledge homepage is presented as a legal provision classification index, organized by section, chapter, and subsection. Each part corresponds to a knowledge entry and is marked with a unique knowledge identifier.

[0039] • In product operation guidance scenarios, the knowledge homepage is presented as a list of functional modules, such as login / registration → main interface → settings → advanced functions. Each function corresponds to a knowledge item and is marked with a unique knowledge identifier.

[0040] The knowledge homepage itself is stored as knowledge content in the knowledge base, and it also has a globally unique knowledge identifier, just as the table of contents is part of a book. AI can retrieve the knowledge homepage from the knowledge base through identifier-based targeted retrieval, a process similar to a reader turning to the table of contents to find the desired content. It should be noted that the knowledge homepage is not limited to being stored in the knowledge base. In other implementations, the knowledge homepage can also be stored in an independent database, in the device's local storage, or provided as a service through an external interface. Correspondingly, the way AI retrieves the knowledge homepage is not limited to identifier-based targeted retrieval tool functions; it can also be achieved through dedicated API calls, database queries, configuration file loading, etc. This embodiment of the invention adopts the method of storing the knowledge homepage as knowledge content in the knowledge base and retrieving it through identifier-based targeted retrieval tool functions. This method unifies the storage system of the knowledge homepage and the knowledge base, facilitating integrated management and maintenance of the system.

[0041] After the knowledge homepage and knowledge base are built, the system implements a tool function for targeted retrieval, which is then called by the AI. The AI ​​model can then determine when to call the appropriate tool function based on the task requirements.

[0042] In this embodiment, identifier-based targeted retrieval is implemented through three independent utility functions, each corresponding to one of the three core levels of the knowledge base (using knowledge ID as an example below): The first utility function is used for knowledge fragment-level retrieval. This function takes a knowledge ID and, when that ID points to a specific knowledge fragment, returns the content of that fragment. For example, if the ID is FRAG_REQ_COLLECT, it returns the complete content of the requirements collection standard process and specification question list.

[0043] The second utility function is used for knowledge file-level retrieval. This function takes a knowledge ID, and when that ID points to a specific knowledge file, it returns the content of all knowledge fragments within that file. For example, if the ID is FILE_DESIGN_SPEC, it returns all specification fragments contained in the design specification file.

[0044] The third utility function is used for category-based knowledge base retrieval. This function takes a knowledge ID, and when the ID points to a specific category knowledge base, it returns the content of all knowledge files and their knowledge fragments within that category knowledge base.

[0045] Designing the three levels as independent utility functions allows AI to select the appropriate tool based solely on the needs of the current step without needing to understand the semantics of the level parameters. Each of the three utility functions has its own independent search entry point and storage space, and the search is executed only within the corresponding level, resulting in precise query scope and high response efficiency.

[0046] In actual deployment, those skilled in the art can further select various implementation methods based on system architecture and business requirements. Several typical variations are listed below: 1. Three independent utility functions (preferred in this embodiment). Each of the three utility functions is independently registered as an interface that the AI ​​can call. Each utility function has a unique name and description, allowing the AI ​​to determine its purpose simply by its name. This approach is suitable for scenarios with clear hierarchical boundaries and significant differences in data volume and access patterns across different levels.

[0047] 2. Unified Utility Function. This approach merges the three levels of retrieval into a single unified utility function, distinguishing the retrieval levels by parameters. For example, the unified utility function accepts two parameters—a knowledge ID and a level identifier (optional values ​​are knowledge fragment, knowledge file, and category knowledge base). Internally, the function performs the retrieval within the corresponding range based on the level identifier value. This method is more streamlined in terms of the number of interfaces and is suitable for scenarios where all levels share the same retrieval logic and storage architecture. There is no fundamental difference in the implementation effect between the two methods described above.

[0048] 3. On-demand Combination. The granularity of utility functions can be flexibly combined according to actual business scenarios. For example, for scenarios that frequently perform fragment retrieval but rarely perform category-level knowledge base retrieval, only the first and second utility functions can be implemented, downgrading the category-level knowledge base retrieval to obtaining it layer by layer through the second utility function and recursive references; conversely, for scenarios that require batch retrieval of complete knowledge base content, only the higher-level utility functions can be implemented. The granularity of tool division can be flexibly adjusted according to the AI ​​task's calling pattern and system performance requirements.

[0049] 4. Caching Strategy. Building upon any of the above methods, a caching mechanism can be added to utility functions. Search results for frequently used knowledge IDs can be cached locally. When the AI ​​repeatedly calls the same ID within the same session, the result is directly returned from the cache, avoiding duplicate searches without affecting the determinism and reproducibility of the search results.

[0050] All of the above-described implementation methods can achieve the core function of identifier-oriented retrieval and fall within the protection scope of this invention. This embodiment uses three independent utility functions as preferred solutions for detailed description, but this does not constitute a limitation on the scope of protection.

[0051] The core feature of the identifier-oriented retrieval tool function is that it does not perform semantic matching or fuzzy search, and always returns the exact same result for the same knowledge identifier, thereby making the knowledge acquired by AI deterministic and reproducible.

[0052] like Figure 1 As shown, after the knowledge homepage and knowledge base are built, the system enters the execution phase.

[0053] After receiving a user task, the AI ​​first retrieves the knowledge homepage. In this embodiment, the knowledge homepage is retrieved using a targeted retrieval tool function (the knowledge ID will continue to be used as the knowledge identifier in the following example). Taking a software development task as an example, the knowledge homepage corresponding to this task is stored in the knowledge base, and its knowledge ID is FILE_HOMEPAGE_SOFTWARE_DEV. The AI ​​can retrieve the entire content of the knowledge homepage by calling the corresponding tool function and passing in this ID.

[0054] After obtaining the knowledge homepage, the AI ​​determines how to utilize the knowledge items within it based on the characteristics of the current task. Depending on the task type, the AI ​​can choose to retrieve and execute the knowledge items from the homepage sequentially as task steps, or it can identify the necessary knowledge items from the homepage to complete the user's task and retrieve the corresponding knowledge content based on their knowledge IDs; or it can combine both methods to complete the user's task. Method 1: Sequential Execution. When a task requires completing a full process in fixed steps, the AI ​​executes the knowledge items sequentially according to their order on the knowledge homepage. In this embodiment, the software development task is a typical scenario, and its knowledge homepage contains the following knowledge items arranged in execution order: The first knowledge entry is "Requirements Gathering and Analysis," with the corresponding knowledge ID FRAG_REQ_COLLECT. The second knowledge item is "Business Module Division," with the corresponding knowledge ID FRAG_MODULE_DIVIDE. The third knowledge item is "Data Table Design", with the corresponding knowledge ID FRAG_TABLE_DESIGN. The fourth knowledge item: "Interface Creation", corresponding to the knowledge ID FRAG_UI_CREATE The fifth knowledge item: "Permission Configuration", corresponding to the knowledge ID FRAG_PERMISSION The sixth knowledge item is "Data Statistics Configuration," with the corresponding knowledge ID FRAG_STATISTICS. Starting with the first knowledge item, the AI ​​reads the knowledge ID of that item and calls the corresponding ID-oriented retrieval tool function to retrieve the knowledge content corresponding to that ID from the knowledge base. Taking the first item as an example, after obtaining the knowledge ID FRAG_REQ_COLLECT, the AI ​​calls the first tool function with that ID as input to obtain the standard process and list of standardized questions for requirements collection. It then combines this with user needs to complete requirements analysis and organize it into a requirements document. After completing the current item, the AI ​​moves on to the next knowledge item and repeats the above process—retrieving the knowledge ID from the knowledge homepage, calling the tool function to retrieve the knowledge content, and performing operations based on the knowledge content. After completing all knowledge items in sequence, the AI ​​returns the results to the user.

[0055] Method 2: Selective Retrieval on Demand. When the task objective is clear and only a portion of the knowledge content needs to be retrieved, the AI ​​analyzes the user's task, identifies knowledge items directly related to the current task from the knowledge homepage, and only retrieves the knowledge IDs corresponding to these items for targeted knowledge content retrieval, without executing all items. For example, in a regulations query scenario, if a user asks about the "sick leave application process," the AI, after obtaining the knowledge homepage, identifies the item "Personnel Regulations → Attendance and Leave Management" from the regulations category directory as relevant to the question. It only retrieves the complete specifications of the sick leave process by calling the knowledge ID corresponding to this item, without executing other irrelevant items. Through selective execution, the AI ​​avoids invalid calls, improves efficiency while maintaining the certainty of the search results, and effectively reduces contextual limitations.

[0056] Method three is a combination of the two methods mentioned above, and will not be elaborated further here.

[0057] The core path of the above workflow is as follows: the knowledge homepage provides an index → ​​AI reads knowledge entries → obtains the knowledge ID → calls tool functions to retrieve knowledge content in a targeted manner → performs operations based on the knowledge → moves to the next knowledge entry or completes execution as needed.

[0058] The knowledge content retrieved by AI may contain references to other knowledge IDs. For example, the knowledge fragment "module partitioning specification" (ID: FRAG_MODULE_DIVIDE) may contain the following: "Module partitioning should follow the principles of business independence (see FRAG_MODULE_DIVIDE_DETAIL_1) and data closure (see FRAG_MODULE_DIVIDE_DETAIL_2)".

[0059] After reading the knowledge fragment, if the AI ​​deems it necessary to further understand the specific principles, it can continue to call ID-oriented retrieval tool functions along the referenced knowledge IDs to obtain deeper knowledge content level by level. Through this recursive referencing mechanism, the knowledge homepage only contains top-level entry information, and each level of knowledge content only contains the information required at that level. The AI's working context remains controllable at all times, while the chained reference of IDs ensures the integrity and accessibility of information.

[0060] When AI cannot obtain sufficient knowledge content during targeted retrieval, supplementary retrieval tools can be invoked for fallback searches. Supplementary retrieval is triggered in scenarios including, but not limited to, the following two: First, the knowledge homepage does not contain a knowledge entry corresponding to the current task, and AI cannot obtain a valid knowledge identifier; second, after obtaining knowledge content through a knowledge identifier, AI deems the obtained information insufficient to fully execute the current operation and needs to further retrieve more relevant information to complete the knowledge.

[0061] The supplementary search tool supports content-based search methods, including but not limited to one or more of keyword search, full-text search, vector search, and semantic search. This embodiment employs a hybrid search method, where the tool performs full-text search scoring and vector relevance scoring separately, and calculates a comprehensive score using the following formula: Full-text search score × 0.3 + Vector relevance score × 0.7 Only knowledge fragments with scores higher than a preset threshold are returned, and they are sorted in descending order of score. In other implementations, keyword retrieval, full-text retrieval, vector retrieval, or semantic retrieval can be used individually according to actual business needs, or the above retrieval methods and their weights can be flexibly combined and configured.

[0062] For example, when AI is creating an interface for a specific industry, if there is no corresponding knowledge entry on the knowledge homepage, it can use the supplementary search tool to input relevant content such as "medical industry interface design specifications" to obtain relevant snippets from the knowledge base for reference. Similarly, if AI has already obtained module division specifications through knowledge identifiers but believes that the examples are insufficient to cover the current complex business scenarios, it can further use the supplementary search tool to input "multi-level organizational structure module division principles" to obtain more reference information.

[0063] Examples of devices, equipment and media Based on the above method, the present invention also provides an AI knowledge base targeted retrieval device, comprising: The knowledge homepage management module is used to provide an index for AI to retrieve knowledge bases. The index contains knowledge entries that correspond one-to-one with the knowledge content of the knowledge base. Each knowledge entry has a globally unique knowledge identifier that is consistent with the corresponding knowledge content. The knowledge base management module is used to manage knowledge content stored in a hierarchical structure. The hierarchy includes at least a classified knowledge base, knowledge files, and knowledge fragments. Each level of knowledge content has a globally unique knowledge identifier. The identifier-oriented retrieval module is used to receive knowledge identifiers and return the corresponding knowledge content based on the knowledge identifiers. This module can be implemented as an AI-callable tool function. In this embodiment, the knowledge identifier is in the form of a knowledge ID, that is, it is implemented as an ID-oriented retrieval module, including a first tool function, a second tool function, and a third tool function that respectively perform retrieval at the knowledge fragment level, the knowledge file level, and the classification knowledge base level. The task processing module is used to receive user tasks, call the knowledge homepage management module to obtain the knowledge homepage according to the user tasks, and generate task execution results based on the knowledge content returned by the identifier-oriented retrieval module and feed them back to the user.

[0064] Optionally, the device may further include a supplementary retrieval module, used to search for knowledge content from the knowledge base and return the results when the required content cannot be obtained by the identifier-oriented retrieval.

[0065] Based on the above method, the present invention also provides an AI knowledge base targeted retrieval device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the above-described AI knowledge base targeted retrieval method.

[0066] Based on the above method, the present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described AI knowledge base targeted retrieval method.

[0067] This invention solves the problems of random search results, fragmented knowledge, context overload, and reliance on third-party tools in traditional AI plug-in knowledge bases by providing an index on the knowledge homepage, enabling precise targeted retrieval through knowledge tags, achieving structured storage through a hierarchical knowledge base, and employing a dual strategy of prioritizing knowledge tag retrieval and providing supplementary retrieval as a fallback.

[0068] Compared with existing technologies, the present invention has the following advantages: the results of AI knowledge acquisition are fixed and reproducible, ensuring stable and consistent output; complete knowledge fragments are accurately acquired according to the identifier, without fragmentation or loss of related information; only the content required for the current step is loaded, reducing the AI ​​context pressure; the retrieval scope is controllable and the source is traceable, improving the standardization of AI's use of knowledge; the self-developed integrated architecture is simple to deploy, data secure, and has strong platform interoperability.

[0069] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for targeted retrieval of AI knowledge bases, characterized in that, include: After receiving the user task, S100 obtains the knowledge homepage according to the user task. The knowledge homepage is used to provide an index for AI to retrieve the knowledge base. S200 Retrieves knowledge content from the knowledge base according to the information in the index; The S300 completes the current task based on the retrieved knowledge content and provides feedback to the user.

2. The AI ​​knowledge base targeted retrieval method according to claim 1, characterized in that, The knowledge base is a structured knowledge base, including: At least one partition, each partition contains at least one category knowledge base, each category knowledge base contains at least one knowledge file, and each knowledge file contains at least one knowledge fragment; Each level of knowledge content has a globally unique knowledge identifier.

3. The AI ​​knowledge base targeted retrieval method according to claim 2, characterized in that, The knowledge homepage contains knowledge entries that correspond one-to-one with the knowledge content of the knowledge base, and each knowledge entry has a globally unique knowledge identifier that is consistent with the corresponding knowledge content.

4. The AI ​​knowledge base targeted retrieval method according to claim 3, characterized in that, The knowledge homepage is stored as knowledge content in the knowledge base and has a globally unique knowledge identifier.

5. The AI ​​knowledge base targeted retrieval method according to claim 4, characterized in that, Specifically, step S200 involves the AI ​​obtaining the knowledge identifier of the required knowledge item from the knowledge homepage based on the user's task requirements, and then retrieving the knowledge content corresponding to the knowledge identifier from the knowledge base.

6. The AI ​​knowledge base targeted retrieval method according to claim 5, characterized in that, The targeted retrieval is achieved through AI-callable utility functions.

7. The AI ​​knowledge base targeted retrieval method according to claim 6, characterized in that, The knowledge content retrieved by AI includes references to other knowledge identifiers. AI can continuously and directionally retrieve knowledge content along the referenced knowledge identifiers, and gradually obtain deeper knowledge content.

8. The AI ​​knowledge base targeted retrieval method according to any one of claims 1-7, characterized in that, When the required knowledge content cannot be obtained through knowledge identifiers or the obtained knowledge content is insufficient, a supplementary retrieval tool is invoked to search for knowledge content from the knowledge base according to the content-matching retrieval method.

9. An AI knowledge base targeted retrieval device, characterized in that, include: The knowledge homepage management module is used to provide an index for AI-based knowledge base retrieval. The knowledge base management module is used to manage knowledge content stored in a hierarchical structure. The identifier-based targeted retrieval module is used to return corresponding knowledge content based on knowledge identifiers; The task processing module is used to receive user tasks and output the task execution results.

10. An AI knowledge base targeted retrieval device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the method according to any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that, The device contains a computer program that, when executed by a processor, implements the steps of the method according to any one of claims 1 to 8.

Images