An acupuncture knowledge base hierarchical semantic retrieval method, system and device

Abstract

The application provides a kind of acupuncture knowledge base hierarchical semantic retrieval method, system and device, method includes: constructing hierarchical authorization module, designs dynamic permission control logic;Receive user query text and identity level, carry out intention recognition and pre-checking of authority;Based on multi-granularity knowledge base architecture, three-way hybrid retrieval is executed in parallel;Through multidimensional dynamic reordering mechanism, combined with user behavior mining agent, dynamic reordering is realized;Call generative large language model to generate explanatory summary containing dialectical basis and personalized suggestions.The application effectively solves the problems of lack of hierarchical management, insufficient retrieval accuracy and static lag of knowledge base in the prior art, significantly improves the accuracy, safety and timeliness of clinical data retrieval.

Description

Technical Field

[0001] This invention relates to the fields of database retrieval, medical data processing, and artificial intelligence, and in particular to a hierarchical semantic retrieval method, system, and device for acupuncture clinical acupuncture knowledge base. Background Technology

[0002] In the highly specialized field of Traditional Chinese Medicine (TCM) acupuncture, characterized by complex knowledge structures and strong clinical orientation, existing retrieval and hierarchical authorization technologies still have significant shortcomings, failing to meet the accuracy, safety, and personalization requirements of real-world clinical scenarios. Specifically, the existing technologies suffer from the following problems: Problem 1: Lack of in-depth understanding of TCM diagnostic logic. General search engines and traditional retrieval systems mainly rely on keyword matching or superficial semantic analysis, which cannot identify the unique diagnostic logic and treatment contraindications of TCM. This may lead to search results that violate diagnostic and treatment principles, posing a safety hazard. Problem 2: Mismatch between retrieval granularity and knowledge structure. Acupuncture knowledge has a significant hierarchical structure, covering multiple levels such as macroscopic treatment principles, mesoscopic disease-syndrome correspondences, and microscopic acupoint attributes. Existing systems mostly use a flat indexing method, which cannot simultaneously meet the needs of grasping the overall treatment principles and accurately querying specific acupoints, resulting in a structural gap between retrieval effectiveness and knowledge organization methods. Problem 3: Lack of tiered authorization and personalized services for different user roles. Different users have significantly different needs for knowledge content, depth, and expression methods, and some high-value clinical case data involves privacy or professional thresholds, requiring strict tiered authorization management. Existing technologies typically do not employ tiered sorting strategies, and the lack of effective access control for high-value clinical experience and sensitive data can easily lead to information overload, misuse, or privacy leaks. Question 4: Static sorting lacks dynamic adaptability. Most existing systems adopt static sorting strategies, which cannot dynamically adjust search results based on users' historical behavioral preferences, making it difficult to meet users' personalized needs and knowledge update requirements that change over time.

[0003] Therefore, developing a TCM acupuncture knowledge base with hierarchical authorization and high-match semantic retrieval, so as to fully integrate computer technology with TCM syndrome differentiation and treatment theory and apply it to actual clinical applications, has become an important research direction in the industry. Summary of the Invention

[0004] The purpose of this invention is to provide a hierarchical semantic retrieval method, system, and device for acupuncture knowledge base oriented towards clinical data, so as to solve the problems existing in the prior art.

[0005] Firstly, to solve the above-mentioned technical problems, the present invention provides a hierarchical semantic retrieval method for acupuncture knowledge base, comprising the following steps: Step 1: Construct a hierarchical authorization module: Define role levels and permission divisions, and design a dynamic permission control method based on knowledge sensitivity; Step 2, User Intent Recognition: Receive the query text and user role level input by the user. Based on the multi-dimensional question parsing intelligent agent module, perform intent recognition and permission pre-check, construct the initial context vector, and realize hierarchical authorization and accurate retrieval. Step 3: Construct a hierarchical multi-path retrieval system: Based on a multi-granularity knowledge base, using the initial context vector as input, perform multi-path hybrid retrieval in parallel to form an initial candidate set, thereby achieving high-precision semantic retrieval that conforms to the user's role. Step 4: Construct a multidimensional dynamic reordering: Based on the multidimensional clinical value assessment and combined with the user behavior mining intelligence agent, the initial candidate set is dynamically reordered according to clinical value priority to generate a sequence of preferred results. Step 5: Result Generation: Invoke the generative large language model, based on the preferred result sequence and user role level, generate and output an explanatory summary containing dialectical evidence and personalized suggestions.

[0006] In one feasible implementation, step 1 specifically includes: Step 11: Manage the acupuncture knowledge base based on role level and permission hierarchy; Step 12: Based on knowledge sensitivity, mark knowledge items with sensitivity tags to achieve access control to the acupuncture knowledge base; Step 13: Assign dynamic access permissions based on dynamic access control.

[0007] In one feasible implementation, step 2 specifically includes: Step 21: Based on the three-layer inference chain and permission constraints, perform scene classification, fine-grained recognition, requirement deconstruction and permission mapping in sequence to construct the initial context vector and generate a retrieval request with permission tags; Step 22: Integrate textual semantic information, user role levels, dynamic access control, and initial context vector. This allows for the creation of hierarchical and precise search requests.

[0008] In one feasible implementation, step 3 specifically includes: Step 31: Generate a set of multi-granularity sub-query vectors through the problem parsing module. And dynamically allocate retrieval weight vectors ; Step 32: Perform a search based on a hybrid search method, according to the question intent. Simultaneous fuzzy matching, BM25 exact matching, and vector semantic retrieval are performed. Step 33: Merge and deduplicate the results of the three parallel searches to obtain the initial candidate set. .

[0009] In one feasible implementation, step 4 specifically includes: Step 41: For the initial candidate set Each candidate document Normalized scores from four core dimensions are extracted to form a clinical value feature vector; Step 42: Based on user behavior, mine intelligent agents and perform full lifecycle user interest modeling and weight strategy management; Step 43: When responding to the user's current query, load the latest generated weighting strategy. The candidate set is rearranged to obtain the optimal result sequence.

[0010] In one feasible implementation, step 5 specifically includes: Step 51: Before generating the answer, optimize the sequence of results output in Step 4. Perform in-depth processing to construct a concise knowledge context that is highly consistent with user intent and conforms to the authorization level. ; Step 52: Invoke the generative large language model Combined with concise context Scene configuration Based on the original intent, generate the final response explanation text; Step 53: Based on log analysis, identify the knowledge items that need to be updated.

[0011] Secondly, based on the same inventive concept, this application also provides a hierarchical semantic retrieval system for acupuncture knowledge base, including a hierarchical authorization module, an intent parsing module, a hierarchical multi-path retrieval module, an agent dynamic rearrangement module, and a result generation module. The hierarchical authorization module, based on role-based knowledge base access control, enables refined management of permissions for users at different levels. The intent parsing module receives the query text and user role level input by the user, performs intent recognition and permission pre-check, and realizes hierarchical authorization and accurate retrieval. The hierarchical multi-path retrieval module achieves accurate knowledge matching under fuzzy query based on a three-layer knowledge architecture, and generates an initial candidate set according to user role permissions. The intelligent agent dynamic reordering module is used to dynamically allocate scoring weights according to user identity, sort the initial candidate set by relevance to obtain the preferred result sequence, and prioritize the display of high-value and highly relevant content. The result generation module is used to generate a personalized sorted list and explanatory summary, and to update the knowledge base log.

[0012] Preferably, the system further includes multiple intelligent agents, which include at least a multi-dimensional problem analysis intelligent agent and a user behavior mining intelligent agent.

[0013] Thirdly, based on the same inventive concept, this application also provides a hierarchical semantic retrieval device for acupuncture knowledge base, including a processor, a memory, and a bus. The memory stores instructions and data read by the processor, and the processor is used to call the instructions and data in the memory to execute the clinically oriented hierarchical authorization management and semantic retrieval method for acupuncture knowledge base as described above. The bus connects the various functional components for transmitting information.

[0014] By adopting the above technical solution, the present invention has the following beneficial effects: This solution constructs a hierarchical authorization module and a multi-dimensional problem-analysis intelligent agent to achieve dynamic access control based on user identity level and knowledge sensitivity. Combined with a Traditional Chinese Medicine (TCM) diagnostic reasoning model and rule-based logical filtering, it effectively avoids search results that violate diagnostic and treatment principles, significantly improving security. Based on a multi-granularity knowledge base architecture, divided into three layers, it can meet the diverse query needs of different users. Through multi-path hybrid retrieval strategies and a dynamic re-ranking mechanism by the intelligent agent, it improves the accuracy of search results and personalized recommendations. Simultaneously, it dynamically allocates retrieval strategies and ranking weights according to user role level, supporting multi-level access control and temporary authorization mechanisms to ensure the security and adaptability of knowledge access. Attached Figure Description

[0015] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0016] Figure 1 A flowchart of the retrieval method provided in an embodiment of the present invention; Figure 2 for Figure 1 Flowchart of the specific method for step 1; Figure 3 for Figure 1 Flowchart of the specific method for step 2; Figure 4 for Figure 1 Flowchart of the specific method for step 3; Figure 5 for Figure 1 Flowchart of the specific method for step 4; Figure 6 for Figure 1 Flowchart of the specific method for step 5; Figure 7 This is a schematic diagram of the semantic retrieval system structure provided in an embodiment of the present invention. Detailed Implementation

[0017] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0019] In the description of this invention, words such as "exemplarily," "for example," etc., are used to indicate that they are examples, illustrations, or descriptions. Any embodiment or design that is described as "exemplary" in this invention should not be construed as being more preferred or advantageous than other embodiments or design options. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of this invention, the meaning expressed by "and / or" can be both, or either one.

[0020] The present invention will be further explained below with reference to specific embodiments.

[0021] It should also be noted that the specific embodiments or implementation methods described below are a series of optimized settings listed by the present invention to further explain the specific content of the invention, and these settings can be combined or used in conjunction with each other.

[0022] Example 1: like Figure 1 As shown in this embodiment, the hierarchical semantic retrieval method for acupuncture knowledge base includes the following steps: Step 1: Construct a hierarchical authorization module: Define role levels and permission divisions, and design dynamic permission control logic based on knowledge sensitivity; Step 2, User Intent Recognition: Receive the query text and user role level input by the user. Based on the multi-dimensional question parsing intelligent agent module, perform intent recognition and permission pre-check, construct the initial context vector, and realize hierarchical authorization and accurate retrieval. Step 3: Construct a hierarchical multi-path retrieval system: Based on a multi-granularity knowledge base architecture, using the initial context vector as input, execute a multi-path hybrid retrieval strategy in parallel to form an initial candidate set, thereby achieving high-precision semantic retrieval that conforms to the user's role. Step 4: Construct a multidimensional dynamic reordering: For the initial candidate set, a multidimensional clinical value evaluation system is constructed, and combined with a user behavior mining agent, a dynamic reordering based on clinical value is achieved to generate a sequence of preferred results. Step 5: Result Generation: Invoke the generative large language model, based on the preferred result sequence and user role level, generate and output an explanatory summary containing dialectical evidence and personalized suggestions.

[0023] In this way, a hierarchical semantic retrieval method for acupuncture knowledge base oriented towards clinical practice can be constructed. Based on the hierarchical authorization management of user roles, it can provide targeted and secure acupuncture knowledge base query and retrieval, and provide a specific knowledge list with clinical operability.

[0024] Furthermore, such as Figure 2 As shown, the specific method of step 1 includes: Step 11: Based on role level and permission hierarchy, manage the acupuncture knowledge base and build permission levels and specific user permissions; Preferably, step 11 specifically includes: Step 111: Define user role levels, such as L0 for the general public, L1 for acupuncture students, L2 for licensed acupuncturists, and L3 for senior expert administrators, etc. Step 112: Set user permissions. Level L0 is for the public, authorized to access public data only; Level L1 is for acupuncture students, authorized to access the approved clinical knowledge base; Level L2 is for licensed acupuncturists, authorized to view the core knowledge base and apply for knowledge modification permissions; Level L3 is for senior expert administrators, authorized to modify the knowledge base, add unpublished experiences and approve knowledge changes. Step 12: Based on knowledge sensitivity, mark knowledge items with sensitivity tags to achieve access control to the acupuncture knowledge base; Preferably, step 12 specifically includes: Based on knowledge sensitivity, sensitive labels are defined as public level (basic knowledge), internal level (departmental routine), expert experience level (requires authorization to view), privacy level (including de-identified patient data) and core level (only modification permission is allowed); Step 13: Implement dynamic access control logic and assign dynamic access permissions; Preferably, step 13 specifically includes: Step 131: Authorize users to search the knowledge base based on their role level; Step 132: Allow user roles to temporarily authorize lower-level users to retrieve specific knowledge entries from the knowledge base based on their permissions; Furthermore, such as Figure 3 As shown, step 2 specifically includes: Step 21: Based on the three-layer inference chain and permission constraints, perform scene classification, fine-grained recognition, requirement deconstruction and permission mapping in sequence to construct the initial context vector and generate a retrieval request with permission tags; Preferably, step 21 specifically includes: Step 211: Based on the query text and user identity information entered by the user, determine the question tendency, match the knowledge sensitivity level, and perform the initial permission prediction; In this embodiment, the matching of knowledge sensitivity levels can be achieved by constructing a hierarchical sensitive word library and knowledge graph tags. For example, it can be divided into L1 (popular science level), L2 (professional level), and L3 (danger level). The user query text is segmented, and different hit levels are matched with low, medium, and high sensitivity. Step 212: When the user's role level is lower than the minimum sensitivity level required for the query content, the system intercepts or marks it as "to be downgraded" in advance, and triggers a permission upgrade prompt or temporary authorization application entry. Step 213: Based on the user's question and the preset scenario, identify the user's deep-seated needs, including explicit intentions and implicit needs, and inject role permission features to ensure that subsequent searches are only conducted within the scope of their authorized knowledge base; In identifying users' deeper needs, explicit intents are directly extracted from the query content, including entities (such as symptoms, medicinal herbs, and acupoints) and actions (such as querying, treating, and identifying). Implicit needs are identified using a large language model combined with the query context. Through deep need identification, it can be determined whether the user requires popular science explanations or professional literature.

[0025] To ensure retrieval within authorized scope, the knowledge base is divided into physically or logically isolated index shards based on sensitivity tags. When generating search instructions, injected role-based access permissions are used as filtering conditions to block index shards that users are not authorized to access at the database level, ensuring that search results absolutely do not contain unauthorized information.

[0026] Step 214: Based on the parsed intent, user role level, and dynamic authorization context, automatically complete the search dimensions, update the weights by combining the user's historical behavior, and extract the constraints required for the search. The weighting of user historical behavior is updated by dynamically adjusting the search strategy based on user characteristics. The system first sets basic weight preferences based on user roles, adjusting the weight of each search dimension for different roles. Building on this, the system dynamically adjusts the weight coefficients of each search dimension by combining the user's recent historical behavior records, such as clicks, likes, shares, and disinterested actions, to make the search results more relevant to the user's current needs.

[0027] Step 22: Integrate textual semantic information, user role levels, dynamic permission control logic, and initial context vector. Construct hierarchical retrieval requests for acupuncture knowledge base; Furthermore, such as Figure 4 As shown, step 3 specifically includes: Step 31: Generate a set of multi-granularity sub-query vectors through the problem parsing module. And dynamically allocate retrieval weight vectors ; Preferably, step 31 specifically includes: Step 311: Multi-granularity knowledge base hierarchical division based on clinical semantics: paragraph level It stores diagnostic and treatment principles and mechanisms of action, suitable for answering "why" and "what are its effects" type of questions; sentence level. Stores precise facts, contraindications, and dosages, suitable for answering "can / cannot" and "how much" type questions; ternary level. It stores triple structures to support relational reasoning problems; Step 312: Generate functions based on subqueries For different user roles such as doctors, students, and the general public, subqueries are mapped to semantic vectors to form multi-granularity subquery vectors. gather:

[0028] Where GenQuery represents the subquery generation function. The original query text. For user role level, For the target knowledge granularity level, corresponding to the document... The levels are, in order, paragraph level, sentence level, and triplet level. Step 32: Perform a search based on a hybrid search method, according to the question intent. Simultaneous fuzzy matching, BM25 exact matching, and vector semantic retrieval are performed. Preferably, step 32 specifically includes: Step 321, Fuzzy Matching: Using the Edit Distance Algorithm Search for similar words in the index to recall a set of related document fragments. :

[0029] in, Each of the document fragments to be retrieved is located in the acupuncture knowledge base. Iterate through the data. The original query text. Indicates from document fragment Extracted key terms or vocabulary This is the threshold for fuzzy matching.

[0030] Step 322, BM25 exact matching: Based on the inverted index, perform sentence-level matching. and paragraph level Perform high-frequency technical terminology matching to retrieve a Top-K document set. :

[0031] in, This represents a document fragment. User query The degree of relevance, For the fusion weights of BM25, For the query term, Indicated as query terms Distinguishing capabilities across the entire knowledge base For word frequency saturation parameters, This is a normalization parameter for document length. This represents the average document length. Indicated as query terms In the current document fragment The number of times it appears in the text.

[0032] Step 323, Vector Semantic Retrieval: Utilizing a dedicated semantic model pre-trained based on a corpus of traditional Chinese medicine. Perform vector semantic retrieval and calculate subqueries. Documents at various granularities Cosine similarity is used to capture deep semantic relationships. The calculation method for cosine similarity is as follows:

[0033] Combined with security mask function High-risk operations are blocked for general users, and a comprehensive semantic score is calculated:

[0034] in, For the fusion weights of vector retrieval, The document fragment to be retrieved. For cosine similarity, This represents the current user's role level.

[0035] The recall of Top-K documents constitutes a set ; Step 33: Merge and deduplicate the three recall results to obtain the initial candidate set. ; Preferably, step 33 specifically includes: Step 331: Merge the three recall results:

[0036] for Calculate the weighted fusion score:

[0037] in Let be the path fusion coefficient, and satisfy... Its value can be dynamically adjusted based on the intent parsing in step 1. When the query is somewhat ambiguous, Improve; when searching for more precise terms, Improvement; when querying with semantic understanding Improvement, and finally output the initial candidate set. :

[0038] SortByScore means sorting the set according to Sort in descending order from highest to lowest. This represents the final total score obtained after weighted calculation of the four recall results. This is a document fragment awaiting rating. This is the initial candidate set.

[0039] Furthermore, such as Figure 5 As shown, step 4 specifically includes: Step 41: For the initial candidate set Each candidate document Normalized scores from four core dimensions are extracted to form a clinical value feature vector; Preferably, step 41 specifically includes: Step 411: Construct a knowledge academic credibility vector: authority Calculated based on the source level of the document, the author's professional title, and the institution's level:

[0040] in, These are the weighting coefficients for each dimension. Objective ratings for each dimension; Step 412: Construct a vector of knowledge novelty and clinical applicability: timeliness According to the release time calculate: ,in The attenuation coefficient is... The current system time is used; knowledge update markers are also introduced: if a document has been recently updated or clinically validated, an additional timeliness bonus is given; Step 413: Construct a vector of clinical acceptance and widespread application of knowledge: high frequency in clinical practice. Calculated based on the number of times it is cited and used in the knowledge base:

[0041] in, This represents the actual number of times the current knowledge has been cited. The highest number of times it has been cited in the entire knowledge base; Step 414: Construct the relevance vector of the user query: basic semantic matching degree The vector similarity is calculated based on the initial screening stage: in, Query vectors for users, For document vectors, Multi-granularity consistency bonus: If the document highly matches the query at the paragraph, sentence, and triple levels, the semantic matching score is improved; Step 42: Learn personalized reordering strategies for intelligent agents based on user behavior, and be responsible for user interest modeling and weight strategy management throughout the entire lifecycle; Preferably, step 42 specifically includes: Step 421: The user behavior mining agent periodically accesses the user behavior log database and extracts sequence data.

[0042] in, For users Behavioral log sequences, For the first The query terms during the next interaction For the first The document clicked during the next interaction. Duration of stay For the first The type of explicit feedback action. , This is standard feedback. As a form of feedback, To share feedback, Feedback for those who are not interested; Step 422: Analyze using time series attention mechanisms To uncover user preference trends across different dimensions of features and define user interest vectors. :

[0043] in, As for the weight of authority, For time-sensitive components, This is a frequently used component in clinical settings. It is the basic semantic matching degree component.

[0044] Specifically, if a user has recently been frequently clicking on highly timely documents and has spent a considerable amount of time on them, and the action is clicking, then The quantity has increased significantly; if the user frequently collects classic ancient books, then Increased quantity; Step 423: User behavior mining agent based on user interest vectors User basic character profile and clinical scenario characteristics Generate and persist the user's personalized reordering weight strategy:

[0045] in This is the behavioral adjustment coefficient, which controls the influence of historical interests on the weights. : Scene adjustment coefficient. This weighting method is used for calculation. It is stored in the user configuration center for later real-time retrieval and retrieval; Step 43: When responding to the user's current query, load the latest weighting strategy generated in step 423. The candidate set is scored and rearranged. Preferably, step 43 specifically includes: Step 431: Process the candidate set Each document in Calculate the final score: according to Sort in descending order to obtain the optimal result sequence list. :

[0046] M represents the maximum number of optimal results ultimately returned to the user.

[0047] Step 432, in Each document is tagged with a clinical value tag, and content with high clinical value is displayed first. Furthermore, such as Figure 6As shown, step 5 specifically includes: Step 51: Before generating the answer, review the list of preferred results output in Step 4. Perform in-depth processing to construct a concise knowledge context that is highly consistent with user intent and conforms to the authorization level. ; Preferably, step 51 specifically includes: Step 511, Traversal In the document fragments, semantic similarity calculation is used to remove fragments with duplicate content and retain highly matched search fragments; Step 512: Based on the task intent and task requirements parsed in Step 1, define the extraction template. The extracted high-density information is then reorganized into a structured sequence of knowledge blocks: in Includes: core conclusions, dialectical basis, data sources, and applicable scenario markings; This represents structured reorganization, transforming unstructured or semi-structured raw text into a standardized format. This process ensures that the context of the input large model is both concise and noise-free. Step 52: Invoke the generative large language model Combined with concise context Scene configuration Based on the original intent, generate the final response explanation text; Preferably, step 52 specifically includes: Step 53: Based on log analysis, identify the knowledge items that need to be updated; Preferably, step 53 specifically includes: Step 531: For knowledge base content that is frequently searched but has a consistently low click-through rate, trigger the knowledge review process, where an L3-level senior expert administrator will assess whether updates or optimizations are needed.

[0048] in, This is a list of documents to be evaluated and optimized. For document fragments The frequency of being searched This is the high-frequency retrieval threshold. For document fragments Click-through rate This is the threshold for low click-through rate.

[0049] Step 532: Under semantic fuzzy query, content that is frequently adopted and has a long dwell time is considered to have high clinical value, and its weight is increased in the ranking model:

[0050] in, For document The semantic clinical value of the classification For query statements, A collection of semantic fuzzy queries. This indicates that the user inputs a fuzzy search. At that time, document The number of times it is clicked by users. This indicates that the user is querying... Click to access the document The duration of stay afterwards.

[0051] Example 2: like Figure 7 As shown, this embodiment provides a hierarchical semantic retrieval system for acupuncture knowledge base, including a hierarchical authorization module, an intent parsing module, a hierarchical multi-path retrieval module, an agent dynamic rearrangement module, and a result generation module; The hierarchical authorization module, based on role-based knowledge base access control, enables refined management of permissions for users at different levels. The intent parsing module receives the query text and user role level input by the user, performs intent recognition and permission pre-check, and realizes hierarchical authorization and accurate retrieval. The hierarchical multi-path retrieval module achieves accurate knowledge matching under fuzzy query based on a three-layer knowledge architecture, and generates an initial candidate set according to user role permissions. The intelligent agent dynamic reordering module is used to dynamically allocate scoring weights according to user identity, sort the initial candidate set by relevance to obtain a list of preferred results, and prioritize displaying high-value and highly relevant content. The result generation module is used to generate a personalized sorted list and explanatory summary, and to update the knowledge base log.

[0052] Furthermore, the system can contain multiple agents, such as a multi-dimensional problem analysis agent and a user behavior mining agent.

[0053] Example 3: This embodiment provides a hierarchical semantic retrieval device for an acupuncture knowledge base, including a processor, a memory, and a bus. The memory is used to store instructions and data, and the processor is used to call and execute the instructions and data in the memory to realize the clinically oriented hierarchical authorization management and semantic retrieval method for an acupuncture knowledge base as described above. The bus is used to connect various functional components to transmit information.

[0054] The processor executes the various methods and processes described above. These methods can be implemented as software programs tangibly contained in a machine-readable medium, such as memory. In some embodiments, part or all of the software program may be loaded and / or installed via memory and / or a communication interface. When the software program is loaded into memory and executed by the processor, one or more steps of the methods described above may be performed. In other embodiments, the processor may also be configured to execute the methods described above via firmware or the like.

[0055] This device can be implemented using a bus architecture. The bus architecture includes any number of interconnect buses and bridges, depending on the hardware application and overall design constraints. The bus connects various circuits, including one or more processors, memory, and / or hardware modules, and connects peripheral devices, voltage regulators, power management circuits, external antennas, and other circuits.

[0056] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A hierarchical semantic retrieval method for acupuncture knowledge base, characterized in that, include: Step 1: Construct a hierarchical authorization module: Define role levels and permission divisions, and design a dynamic permission control method based on knowledge sensitivity; Step 2, User Intent Recognition: Receive the query text and user role level input by the user. Based on the multi-dimensional question parsing intelligent agent module, perform intent recognition and permission pre-check, construct the initial context vector, and realize hierarchical authorization and accurate retrieval. Step 3: Construct a hierarchical multi-path retrieval system: Based on a multi-granularity knowledge base, using the initial context vector as input, perform multi-path hybrid retrieval in parallel to form an initial candidate set; Step 4: Construct a multidimensional dynamic reordering: Based on the multidimensional clinical value assessment and combined with the user behavior mining intelligence agent, the initial candidate set is dynamically reordered according to clinical value priority to generate a sequence of preferred results. Step 5: Result Generation: Invoke the generative large language model, based on the preferred result sequence and user role level, generate and output an explanatory summary containing dialectical evidence and personalized suggestions.

2. The method according to claim 1, characterized in that, Step 1 specifically includes: Step 11: Manage the acupuncture knowledge base based on role level and permission hierarchy; Step 12: Based on knowledge sensitivity, mark knowledge items with sensitivity tags to achieve access control to the acupuncture knowledge base; Step 13: Assign dynamic access permissions based on dynamic access control.

3. The method according to claim 1, characterized in that, Step 2 specifically includes: Step 21: Based on the three-layer inference chain and permission constraints, perform scene classification, fine-grained recognition, requirement deconstruction and permission mapping in sequence to construct the initial context vector and generate a retrieval request with permission tags; Step 22: Integrate textual semantic information, user role levels, dynamic permission control, and initial context vectors to construct hierarchical authorization and accurate retrieval requests.

4. The method according to claim 1, characterized in that, Step 3 specifically includes: Step 31: Generate a set of multi-granularity sub-query vectors through the problem parsing module. And dynamically allocate retrieval weight vectors ; Step 32: Perform a search based on a hybrid search method, according to the question intent. Simultaneous fuzzy matching, BM25 exact matching, and vector semantic retrieval are performed. Step 33: Merge and deduplicate the results of the three parallel searches to obtain the initial candidate set. .

5. The method according to claim 1, characterized in that, Step 4 specifically includes: Step 41: For the initial candidate set Each candidate document Normalized scores from four core dimensions are extracted to form a clinical value feature vector; Step 42: Based on user behavior, mine intelligent agents and perform full lifecycle user interest modeling and weight strategy management; Step 43: When responding to the user's current query, load the latest generated weighting strategy. The candidate set is rearranged to obtain the optimal result sequence.

6. The method according to claim 1, characterized in that, Step 5 specifically includes: Step 51: Before generating the answer, optimize the sequence of results output in Step 4. Perform in-depth processing to construct a concise knowledge context that is highly consistent with user intent and conforms to the authorization level. ; Step 52: Invoke the generative large language model and combine it with the condensed context. Scene configuration Based on the original intent, generate the final response explanation text; Step 53: Based on log analysis, identify the knowledge items that need to be updated.

7. A hierarchical semantic retrieval system for acupuncture knowledge base, characterized in that, The system is used to perform the method as described in any one of claims 1-6, and the system includes a hierarchical authorization module, an intent parsing module, a hierarchical multi-path retrieval module, an agent dynamic rearrangement module, and a result generation module; The hierarchical authorization module uses role-based knowledge base access control to manage permissions for users at different levels. The intent parsing module receives the query text and user role level input by the user, and performs intent recognition and permission pre-check. The hierarchical multi-path retrieval module achieves accurate knowledge matching under fuzzy search based on a three-layer knowledge architecture, and generates an initial candidate set based on user role permissions. The agent dynamic reordering module is used to dynamically allocate scoring weights according to user identity and sort the initial candidate set by relevance to obtain the optimal result sequence; The results generation module is used to generate personalized sorted lists and explanatory summaries, and update the knowledge base log.

8. The system according to claim 7, characterized in that, The system also includes multiple intelligent agents, including at least a multi-dimensional problem analysis intelligent agent and a user behavior mining intelligent agent.

9. A hierarchical semantic retrieval device for an acupuncture knowledge base, characterized in that, It includes a processor, a memory, and a bus. The memory stores instructions and data read by the processor. The processor is used to call the instructions and data in the memory to execute the method as described in any one of claims 1-6. The bus connects the functional components for transmitting information.

Images