Information retrieval method and system based on large model and vector knowledge base

Abstract

This invention provides an information enhancement retrieval method and system based on a large model and vector knowledge base, applied to the field of medical supply chain management. By receiving retrieval requests initiated by users through medical SPD (Supply Chain Device) business terminals, a medical SPD business scenario perception model is triggered. This model has built-in mapping rules between business processes and retrieval intents, and can output scenario adaptation parameters matching the retrieval request. A medical SPD vector knowledge base storing structured and unstructured knowledge units of the entire medical SPD business process is invoked, and a dynamic association network between retrieval intents and knowledge units is constructed using the scenario adaptation parameters. This dynamic association network is input into a large-scale medical SPD business enhancement model jointly trained with medical SPD business data and scenario cases to generate a knowledge enhancement set. This set then generates a medical SPD business execution plan containing operational steps, related knowledge bases, and data support sources, which is then pushed to the business terminal, improving the accuracy and comprehensiveness of medical SPD business information retrieval.

Description

Technical Field

[0001] This invention relates to the field of medical supply chain management technology, and more specifically, to an information enhancement retrieval method and system based on a large model and vector knowledge base. Background Technology

[0002] In the field of healthcare supply chain management (SPD), the increasing complexity and sophistication of business operations have placed higher demands on information retrieval and utilization. Traditional healthcare SPD business retrieval methods mainly rely on simple keyword matching, which has several limitations. On the one hand, they cannot accurately understand the business operational needs behind user search requests and the associated healthcare SPD business context, resulting in significant discrepancies between search results and actual needs. For example, in the procurement of medical consumables, if a user initiates a search for a certain consumable, traditional retrieval may only return basic information about that consumable, failing to provide more accurate information by combining current procurement plans, inventory status, and other business contexts. On the other hand, traditional retrieval methods struggle to build deep connections between search intent and knowledge units, failing to fully utilize the rich structured and unstructured knowledge throughout the entire healthcare SPD business process, resulting in search results lacking comprehensiveness and relevance. Furthermore, the business execution plans generated by traditional retrieval often lack knowledge basis, making it difficult to meet the stringent accuracy and reliability requirements of healthcare SPD operations. Summary of the Invention

[0003] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide an information enhancement retrieval method based on a large model and a vector knowledge base, the method comprising:

[0004] The system receives a search request initiated by a user through a medical SPD business terminal, triggering a medical SPD business scenario perception model. The search request includes business operation requirements and associated medical SPD business context. The medical SPD business scenario perception model has built-in mapping rules between business processes and search intent.

[0005] The medical SPD business scenario perception model outputs scenario adaptation parameters that match the retrieval request, and then calls the medical SPD vector knowledge base. The scenario adaptation parameters include business process identifiers, core data dimensions, and related business rules. The medical SPD vector knowledge base stores structured and unstructured knowledge units of the entire medical SPD business process.

[0006] Using the scenario adaptation parameters as the core elements, a dynamic association network between retrieval intent and knowledge units is constructed in the medical SPD vector knowledge base. Each node in the dynamic association network corresponds to a knowledge unit or retrieval intent element, and the connections between nodes represent business logic associations.

[0007] The dynamic association network is input into the medical SPD business enhancement model to generate a knowledge enhancement set. The medical SPD business enhancement model is jointly trained with medical SPD business data and scenario cases and has the ability to complete knowledge and deduce business logic.

[0008] Based on the knowledge enhancement set, a medical SPD business execution plan is generated, and the business execution plan is pushed to the medical SPD business terminal. The business execution plan includes operation steps, related knowledge basis, and data support sources.

[0009] Furthermore, embodiments of the present invention also provide an information enhancement retrieval system based on a large model and a vector knowledge base, comprising:

[0010] A processor; a machine-readable storage medium for storing machine-executable instructions of the processor; wherein the processor is configured to execute the aforementioned information augmentation retrieval method based on a large model and vector knowledge base by executing the machine-executable instructions.

[0011] In another aspect, embodiments of the present invention also provide a computer program product, the computer program product including machine-executable instructions, the machine-executable instructions being stored in a computer-readable storage medium, a processor of a computer device reading the machine-executable instructions from the computer-readable storage medium, the processor executing the machine-executable instructions, causing the computer device to execute the aforementioned information enhancement retrieval method based on large models and vector knowledge bases.

[0012] Based on the above, by receiving retrieval requests initiated by users from medical SPD business terminals, a medical SPD business scenario awareness model with built-in business process and retrieval intent mapping rules is triggered. This model can accurately capture users' business operation needs and related business contexts. When calling the medical SPD vector knowledge base, a dynamic association network between retrieval intent and knowledge units is constructed with scenario adaptation parameters as the core. This network intuitively presents business logic relationships through nodes and connections, integrating structured and unstructured knowledge from the entire medical SPD business process, achieving in-depth knowledge mining and efficient utilization. The dynamic association network is input into a large-scale medical SPD business enhancement model jointly trained with medical SPD business data and scenario cases to generate a knowledge enhancement set. Leveraging the large-scale model's knowledge completion and business logic deduction capabilities, the retrieval results are further enriched, improving the completeness and accuracy of knowledge. Finally, the medical SPD business execution plan generated based on the knowledge enhancement set includes operation steps, related knowledge basis, and data support sources, providing users with comprehensive, reliable, and operable guidance, effectively improving the execution efficiency and accuracy of medical SPD business and reducing business risks. Attached Figure Description

[0013] Figure 1 This is a schematic diagram of the execution flow of the information enhancement retrieval method based on a large model and vector knowledge base provided in an embodiment of the present invention.

[0014] Figure 2 This is a schematic diagram of exemplary hardware and software components of the information enhancement retrieval system based on a large model and vector knowledge base provided in an embodiment of the present invention. Detailed Implementation

[0015] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating an information enhancement retrieval method based on a large model and vector knowledge base, provided in one embodiment of the present invention. The following is a detailed description of this information enhancement retrieval method based on a large model and vector knowledge base.

[0016] Step S110: Receive a search request initiated by the user through the medical SPD business terminal and trigger the medical SPD business scenario perception model. The search request includes business operation requirements and associated medical SPD business context. The medical SPD business scenario perception model has built-in mapping rules between business processes and search intent.

[0017] This embodiment uses the example of a medical SPD (Service Provider Development) staff member in the orthopedic ward of a hospital needing to process a batch of high-value consumable requisition requests for emergency surgeries. When the staff member initiates a relevant search request on the medical SPD terminal, the medical SPD scenario awareness model is triggered. This search request explicitly includes the business operation requirement of requisitioning high-value consumables, and the associated medical SPD business context includes the current surgical schedule of the orthopedic ward, patient basic information, and existing consumable inventory. The medical SPD scenario awareness model has a large number of pre-set mapping rules between business processes and search intents. For example, when keywords such as "high-value consumables" and "emergency requisition" appear in the search request, the model can correspond to the business process of emergency requisition of high-value consumables.

[0018] Step S111: Receive user input information through the interactive interface of the medical SPD business terminal. The interactive interface supports text input, voice input, and business form import. Convert user input information in different forms into raw search text in a unified format. During the conversion process, retain the tone and pause features of voice input and the field associations of imported forms.

[0019] In the aforementioned emergency requisition scenario for high-value orthopedic consumables, the interactive interface of the medical SPD business terminal provides business personnel with multiple input methods. Assuming a business personnel uses voice input, stating, "There are three surgeries in the orthopedic ward tomorrow morning, and we urgently need to requisition a batch of high-value consumables, including artificial joints and bone screws," the interactive interface, upon receiving this voice input, will convert it into raw search text in a standardized format. During the conversion process, the interface will retain the intonation and pauses characteristic of the voice input, such as the natural pauses at phrases like "three surgeries," "high-value consumables," and "artificial joints." These pauses help to more accurately understand the business semantics later. If the business personnel choose to import data via a business form, and the imported form contains fields such as surgery number, surgery name, required consumable name, and quantity, the interactive interface will retain the relationships between these fields in the raw search text, ensuring accurate identification of the logical connections between the fields during subsequent processing.

[0020] Step S112: Perform business semantic cleaning on the original search text to obtain core text segments containing business requirements. The division of core text segments is based on sentence-end punctuation and semantic pauses.

[0021] After obtaining the original search text, business semantic cleaning is required. For example, the original search text may contain some casual remarks or repetitive expressions unrelated to business needs. The business semantic cleaning process will remove such content, retaining only the core content related to the application for high-value consumables. The division of core text segments is based on sentence-ending punctuation and semantic pauses. For example, "There are three surgeries in the orthopedic ward tomorrow morning, and we urgently need to apply for a batch of high-value consumables, including artificial joints and bone screws," is divided into core text segments such as "There are three surgeries in the orthopedic ward tomorrow morning," "We urgently need to apply for a batch of high-value consumables," and "Including artificial joints and bone screws." Each segment revolves around a specific business semantic point.

[0022] Step S113: Extract medical SPD business feature words from the core text fragment. The business feature words cover medical supply names, business operation verbs, business process names, data requirement types, and responsible entity identifiers. Each business feature word corresponds to a unique business attribute code.

[0023] For the segmented core text fragments, it is necessary to extract the medical SPD (Special Purpose Device) business feature words. In the core text fragment "urgently need to apply for a batch of high-value consumables, including artificial joints, bone nails, etc.", "apply" is a business operation verb, while "high-value consumables," "artificial joints," and "bone nails" are names of medical supplies. Each business feature word has a unique business attribute code. For example, "apply" corresponds to code OP001, "high-value consumables" corresponds to code MT005, "artificial joints" corresponds to code MT00501, and "bone nails" corresponds to code MT00502, etc. The above codes are formulated according to the standard classification system of medical SPD business to facilitate unified business processing and analysis in the future.

[0024] Step S1131: Call the preset medical SPD business feature word dictionary. The medical SPD business feature word dictionary contains commonly used business feature words in the medical SPD field and their corresponding business categories, expression variations and semantic weights. The business categories include material information, operation behavior, business process and data requirement.

[0025] When extracting business feature words, the system first calls a pre-defined medical SPD business feature word dictionary. This dictionary is built based on professional knowledge and business practices in the medical SPD field and contains a large number of commonly used business feature words. For example, in the material information category, in addition to "high-value consumables," "artificial joints," and "bone nails," it also includes "syringe" and "gauze"; the operational behavior category includes "requisition," "warehousing," "outbound," and "inventory"; the business process category includes "procurement plan," "inventory management," and "consumable delivery"; and the data requirement category includes "inventory quantity," "purchase price," and "usage record." Each business feature word corresponds to a business category, expression variants, and semantic weight. For example, the expression variants of "requisition" may include "apply for receipt" and "request," and its semantic weight is set according to its importance in the business process.

[0026] Step S1132: Perform word segmentation on the core text segment, match the word segmented vocabulary units with the medical SPD business feature word dictionary, select the successfully matched vocabulary units as candidate business feature words, and record the business category and semantic weight corresponding to each candidate business feature word.

[0027] The core text segment "There are three surgeries in the orthopedic ward tomorrow morning, and we urgently need to apply for a batch of high-value consumables, including artificial joints and bone nails, etc." is segmented into word units, resulting in lexical units such as "orthopedic ward," "tomorrow morning," "have," "three surgeries," "surgeries," "urgently need," "apply," "batch," "high-value consumables," "including," "artificial joints," "bone nails," and "etc." These lexical units are then matched with a medical SPD business feature word dictionary, selecting successfully matched lexical units such as "surgeries," "apply," "high-value consumables," "artificial joints," and "bone nails" as candidate business feature words. Simultaneously, the business category corresponding to each candidate business feature word is recorded; for example, "surgeries" belongs to the business process category, "apply" belongs to the operational behavior category, and "high-value consumables," "artificial joints," and "bone nails" belong to the material information category, with their respective semantic weights also recorded.

[0028] Step S1133: Deduplicate the candidate business feature words, use the deduplicated candidate business feature words as network nodes, establish connections between nodes according to the semantic relationships in the core text fragments, and the thickness of the connections represents the tightness of the semantic relationships to construct a business feature word association network.

[0029] The candidate business feature words obtained after matching may contain duplicates. For example, if "high-value consumables" appears in multiple core text segments, deduplication is required to ensure that only one of each candidate business feature word is retained. The deduplicated candidate business feature words, such as "surgery," "application," "high-value consumables," "artificial joint," and "bone nail," will serve as nodes in the business feature word association network. Connections are established between these nodes based on the semantic relationships in the core text segments. For example, "surgery" and "application" are associated because of the application behavior arising from surgical needs; "application" and "high-value consumables" are associated as an operation and an object; and "high-value consumables" and "artificial joint" and "bone nail" have an inclusion relationship. The thickness of the connections is determined by the tightness of the semantic relationship. For example, "application" and "high-value consumables" have a relatively strong association, so the connection is thicker; the association between "surgery" and "artificial joint" is indirectly achieved through "application" and "high-value consumables," so the association is relatively weaker, and the connection is thinner.

[0030] Step S1134: Analyze the importance of nodes in the business feature word association network, and determine the core feature words by calculating the degree centrality of the nodes. The higher the degree centrality, the wider the association range of the feature word in the business feature word association network, and the more critical its role in expressing the retrieval intent.

[0031] In constructing a business feature word association network, it is necessary to analyze the importance of each node. The degree centrality of a node refers to the number of direct associations between that node and other nodes in the network. For example, the node "high-value consumables" is directly associated with nodes such as "request," "artificial joint," and "bone nail," and its degree centrality is relatively high; the node "surgery" is mainly directly associated with the "request" node, and its degree centrality is relatively low. By calculating the degree centrality of each node, nodes with high degree centrality are identified as core feature words. These core feature words play a crucial role in expressing the search intent. In this scenario, "request" and "high-value consumables" are likely to be identified as core feature words.

[0032] For example, step S11341: Define the meaning of the degree centrality of nodes in the business feature word association network. The degree centrality of a node is specifically manifested as the number of direct associations between the node and other nodes in the business feature word association network. The more direct associations a node has with other nodes, the more obvious its degree centrality feature is.

[0033] In a business feature term association network, the degree centrality of a node is defined as the number of direct associations between that node and other nodes. A higher number of direct associations indicates a more extensive network connectivity for the node, and better reflects the core content of the business request. For example, in the aforementioned business feature term association network for the orthopedic high-value consumables requisition scenario, the node "high-value consumables" directly associates with multiple nodes such as "requisition," "artificial joint," and "bone nail," exhibiting a large number of direct associations and a clear degree centrality, indicating that "high-value consumables" is a crucial core concept in this search request.

[0034] Step S11342: Construct a network adjacency matrix. The rows and columns of the network adjacency matrix correspond to nodes in the network associated with business feature words. A matrix element value of 1 indicates that there is an association link between the nodes in the corresponding row and column, and an element value of 0 indicates that there is no association link.

[0035] To more clearly calculate the degree centrality of nodes, a network adjacency matrix is ​​constructed. Assuming the nodes in the business feature word association network are "surgery," "application," "high-value consumables," "artificial joint," and "bone nail," then the rows and columns of the adjacency matrix will contain these five nodes. If there is an association link between "surgery" and "application," then the element in the "surgery" row and "application" column of the matrix will be 1; otherwise, it will be 0. Similarly, if there is an association link between "application" and "high-value consumables," the corresponding element will be 1. If there is an association link between "high-value consumables" and "artificial joint" or "bone nail," the corresponding element will be 1. Other nodes without direct association will have elements with a value of 0.

[0036] Step S11343: Calculate the degree value of each node. The degree value is the sum of the element values ​​of the corresponding row in the network adjacency matrix. The degree value is used to reflect the number of associations between a node and other nodes.

[0037] For the adjacency matrix above, calculate the degree value of each node. Taking the "High-Value Consumables" node as an example, the sum of the element values ​​in its row is 1+1+1=3, which is associated with "Application", "Artificial Joint", and "Bone Screw", meaning the degree value of the "High-Value Consumables" node is 3. The degree value of the "Application" node might be 1+1=2, which is associated with "Surgery" and "High-Value Consumables". By calculating the degree value, the number of associations between each node and other nodes can be intuitively reflected.

[0038] Step S11344: Normalize the degree values ​​of all nodes, and correct the normalized degree centrality values ​​by combining the semantic weights of medical SPD business. After sorting all nodes from high to low according to the corrected degree centrality values, select the nodes with corrected degree centrality values ​​higher than the screening threshold as core feature words, and finally determine the core feature word set.

[0039] The calculated degree values ​​of each node are normalized, mapping them to the range of 0-1 for comparison and analysis. For example, assuming the largest degree value in the network is 5 and the degree value of "high-value consumables" is 3, the normalized degree centrality is 3 / 5 = 0.6. Then, the normalized degree centrality is corrected by incorporating the semantic weights of the medical SPD business. Nodes with higher semantic weights will receive a certain improvement during the correction. For example, "application" has a high semantic weight in the business process; assuming its normalized degree centrality is 0.5 and the semantic weight correction coefficient is 1.2, the corrected degree centrality is 0.5 × 1.2 = 0.6. All nodes are sorted from highest to lowest according to their corrected degree centrality values. A filtering threshold, such as 0.5, is set, and nodes with corrected degree centrality values ​​higher than this threshold are selected as core feature words. Finally, "application" and "high-value consumables" are determined as the core feature word set.

[0040] Step S1135: Group the core feature words that meet the degree centrality standard according to business categories to form material information feature group, operation behavior feature group, business process feature group and data demand feature group. The feature words in each group are sorted according to semantic weight.

[0041] The identified core feature words are grouped according to business categories. In this scenario, "high-value consumables," "artificial joints," and "bone nails" belong to the material information feature group; "application" belongs to the operational behavior feature group; and "surgery" belongs to the business process feature group. Feature words within each group are sorted according to semantic weight. For example, in the material information feature group, the semantic weight of "high-value consumables" may be higher than that of "artificial joints" and "bone nails," so it is ranked first; in the operational behavior feature group, "application" is the only feature word and is ranked first; and in the business process feature group, "surgery" is ranked first.

[0042] Step S1136: Add business attribute codes to the feature words in each group. The business attribute codes include business classification identifiers, semantic weight levels, and associated business scenario codes. The grouped feature words and corresponding business attribute codes together constitute the medical SPD business feature word set.

[0043] Add business attribute codes to the feature words within each group. For example, the business classification identifier for the material information feature group is MT, the semantic weight level of "high-value consumables" is level 1 (the highest level), and the associated business scenario code is YGSC001 (high-value consumables application scenario), then its business attribute code is MT-1-YGSC001; the semantic weight level of "artificial joint" is level 2, the associated business scenario code is also YGSC001, and the business attribute code is MT-2-YGSC001; the semantic weight level of "bone nail" is level 2, and the business attribute code is MT-2-YGSC001. The business classification identifier for the operational behavior feature group is OP, the semantic weight level of "application" is level 1, the associated business scenario code is YGSC001, and the business attribute code is OP-1-YGSC001. The business classification identifier for the business process feature group is BH, the semantic weight level of "surgery" is level 1, the associated business scenario code is YGSC001, and the business attribute code is BH-1-YGSC001. These grouped feature words and their corresponding business attribute codes together constitute the medical SPD business feature word set.

[0044] Step S114: Call the context association module of the medical SPD business scenario perception model to read the business permission information, historical operation records and currently associated medical SPD business document data of the currently logged-in user of the medical SPD business terminal, and generate a business context feature set.

[0045] The context association module of the medical SPD business scenario awareness model reads relevant information about the currently logged-in user. In the orthopedic high-value consumable requisition scenario, the currently logged-in user is a business personnel in the orthopedic ward. Their business permission information may include approval authority for high-value consumable requisitions and the scope of consumables that can be requisitioned; historical operation records may include records of high-value consumable requisitions by this business personnel in the past month, the quantity requisitioned, and the surgeries involved; the currently associated medical SPD business document data may include surgical notification slips for three surgeries tomorrow morning, existing high-value consumable inventory documents, etc. The context association module integrates the above information to generate a business context feature set, which includes features from multiple dimensions such as user permission features, historical behavior features, and current document features.

[0046] Step S115: The business feature words and the business context feature set are fused to generate a search intent feature vector. During the fusion process, the feature words associated with the current business document data are given higher weights. The weight allocation is determined based on the urgency and relevance of the business document.

[0047] The set of business feature words and the set of business context feature words are fused together. During the fusion process, feature words associated with the current business document data (such as a surgery notification) are given higher weights because they are directly related to the urgent requisition of high-value consumables. For example, feature words such as "surgery" and "urgently needed" associated with information like "tomorrow morning" and "three surgeries" mentioned in the surgery notification will have higher weights than other feature words. The weight allocation is determined based on the urgency and relevance of the business document. The surgery notification has a high degree of urgency and a high degree of relevance to the requisition of high-value consumables, so the corresponding feature words have higher weights. Through the above weighted fusion process, a search intent feature vector that accurately reflects the user's search intent is generated.

[0048] Step S116: Call the scenario rule engine of the medical SPD business scenario perception model to store the matching rules between the retrieval intent feature vector and the business scenario constructed by medical SPD business experts. Each matching rule includes feature vector matching conditions, scenario identifier and scenario description information.

[0049] The system utilizes the scenario rule engine of the healthcare SPD business scenario awareness model. This engine stores a large number of matching rules constructed by healthcare SPD business experts based on their business experience and professional knowledge. Each matching rule includes feature vector matching conditions, scenario identifiers, and scenario descriptions. For example, a matching rule might have feature vector matching conditions containing keywords such as "request," "high-value consumables," "surgery," and "urgent need," with the weight of the "surgery" feature word reaching a certain threshold. Its scenario identifier would be YGSC001, and its scenario description would be "urgent request scenario for high-value consumables." When the generated search intent feature vector satisfies the feature vector matching conditions of this matching rule, the model can determine the corresponding business scenario.

[0050] Step S117: Input the search intent feature vector into the scene rule engine, and generate scene matching results through rule matching. The scene matching results include the scene identifier of the successfully matched scene, the matching degree score, and the associated business process nodes.

[0051] The generated search intent feature vector is input into the scenario rule engine, which compares the feature vector with the stored matching rules one by one. In this scenario, the search intent feature vector contains feature words such as "request," "high-value consumables," "surgery," and "urgent need," and its weight meets the requirements, matching the matching rule with scenario identifier YGSC001. After rule matching, a scenario matching result is generated, where the scenario identifier is YGSC001, and the matching score is calculated based on the degree of matching between the feature vector and the rule, such as 0.9 (out of 1.0). The associated business process nodes include "submitting the requisition form," "department review," "consumables management department approval," "warehouse issuance," and "delivery to the operating room," etc.

[0052] Step S118: Select the main scene identifier and the associated scene identifier according to the matching score. The main scene identifier corresponds to the core business requirement of the retrieval request, and the associated scene identifier corresponds to the extended business requirement related to the core requirement. The main scene identifier and the associated scene identifier are sent together to the parameter generation module of the medical SPD business scene perception model to trigger the parameter generation module to start running.

[0053] The main scenario identifier and associated scenario identifiers are selected based on the matching score in the scenario matching results. Since the retrieval intent feature vector in this scenario has the highest matching score (0.9) with scenario YGSC001, the main scenario identifier is YGSC001, corresponding to the core business requirement "emergency application for high-value consumables". Simultaneously, there may be some associated scenarios, such as "high-value consumables inventory query scenario" (matching score 0.7) and "high-value consumables usage record query scenario" (matching score 0.6), etc. These associated scenario identifiers correspond to extended business needs related to the core requirement, such as checking inventory sufficiency and viewing the historical usage of this type of consumable. The main scenario identifier and associated scenario identifiers are jointly sent to the parameter generation module of the medical SPD business scenario perception model, triggering the module to start running.

[0054] Step S120: The scenario adaptation parameters that match the retrieval request are output by the medical SPD business scenario perception model, and the medical SPD vector knowledge base is called. The scenario adaptation parameters include business link identifiers, core data dimensions and related business rules. The medical SPD vector knowledge base stores structured and unstructured knowledge units of the entire medical SPD business process.

[0055] After determining the main scenario identifier YGSC001 and related scenario identifiers, the parameter generation module of the medical SPD business scenario perception model outputs scenario adaptation parameters that match the retrieval request. These scenario adaptation parameters include business process identifiers, such as "YGSC001-01" (submitting an application form) and "YGSC001-02" (department review); core data dimensions, such as "name of consumables applied for," "specifications," "quantity," "urgency level," "applying department," and "surgery time"; and related business rules, such as "high-value consumables applications require department head review and signature" and "urgent applications must be responded to within 2 hours." Simultaneously, the medical SPD vector knowledge base is invoked. This knowledge base stores structured and unstructured knowledge units covering the entire medical SPD business process. Structured knowledge includes various business process specifications, data dictionaries, and approval rules, such as standard process documents for high-value consumables applications and tables of consumable specifications and codes. Unstructured knowledge includes expert experience summaries, solutions to difficult problems, and interpretations of relevant policies.

[0056] Step S130: Using the scenario adaptation parameters as the core elements, construct a dynamic association network of retrieval intent and knowledge units in the medical SPD vector knowledge base. Each node in the dynamic association network corresponds to a knowledge unit or retrieval intent element, and the connections between nodes represent business logic associations.

[0057] Using business process identifiers, core data dimensions, and related business rules from the scenario adaptation parameters as core elements, a dynamic association network is constructed in the medical SPD vector knowledge base. Nodes in the dynamic association network include retrieval intent elements (such as "emergency application for high-value consumables" and "orthopedic surgery") and knowledge units (such as the high-value consumables application process knowledge unit, commonly used high-value orthopedic consumables knowledge unit, and emergency application approval rule knowledge unit). Connections between nodes represent business logic relationships; for example, the "emergency application for high-value consumables" node is connected to the high-value consumables application process knowledge unit node, the "orthopedic surgery" node is connected to the commonly used high-value orthopedic consumables knowledge unit node, and the high-value consumables application process knowledge unit node is connected to the emergency application approval rule knowledge unit node, etc.

[0058] Step S131: Analyze the scenario adaptation parameters, extract the business process identifier, core data dimension and related business rules as the core elements of network construction, and add business attribute descriptions and weight ratios in the search intent for each core element.

[0059] The scenario adaptation parameters are analyzed to extract business process identifiers, such as "YGSC001-01", "YGSC001-02", and "YGSC001-03"; core data dimensions, such as "name of consumables requested", "specifications", "quantity", "urgency level", "requesting department", and "surgery time"; and associated business rules, such as "requests for high-value consumables require approval and signature from the department head", "emergency request forms must be responded to within 2 hours", and "requests exceeding a certain limit require approval from the hospital director". Business attribute descriptions are added to each core element. For example, the business attribute description for the business process identifier "YGSC001-01" is "submitting an emergency request form for high-value consumables", the business attribute description for the core data dimension "urgency level" is "identifying the urgency level of the request, divided into general, emergency, and urgent", and the business attribute description for the associated business rule "emergency request forms must be responded to within 2 hours" is "stipulating the response time requirements for high-value consumables requests in emergency situations". At the same time, a weight percentage is added to each core element in the search intent. The weight percentages of business process identifiers and associated business rules are relatively high, such as 0.35 for both, while the weight percentage of core data dimensions is 0.3, in order to highlight their importance in network construction.

[0060] Step S132: Call the knowledge organization engine of the medical SPD vector knowledge base. The knowledge organization engine divides the medical SPD business process into multiple knowledge domains. Each knowledge domain contains several knowledge units, and each knowledge unit corresponds to a specific business knowledge point.

[0061] The system utilizes the knowledge organization engine of the medical SPD vector knowledge base. This engine divides the medical SPD business processes into multiple knowledge domains based on different stages and content. Examples include "Consumables Procurement Management," "Inventory Management," "Consumables Requisition and Issuance," and "Fee Settlement." Each knowledge domain contains several knowledge units, each corresponding to a specific business knowledge point. For instance, the "Consumables Requisition and Issuance" knowledge domain includes knowledge units such as "High-Value Consumables Requisition Process," "Ordinary Consumables Requisition Process," "Emergency Requisition Processing Standards," and "Consumables Issuance Verification Requirements." Each knowledge unit has a unique identifier and a detailed content description.

[0062] Step S133: Based on the business process identifier, locate the corresponding target knowledge domain, and read the metadata information of all knowledge units in the target knowledge domain. The metadata information includes knowledge unit identifier, business attribute tag, associated business rules and data dimension description.

[0063] Based on the business process identifiers "YGSC001-01" and "YGSC001-02" in the scenario adaptation parameters, the corresponding target knowledge domain is located as "consumable requisition and issuance". Then, the metadata information of all knowledge units within this target knowledge domain is read. For example, in the metadata information of the "High-value consumable requisition process" knowledge unit, the knowledge unit identifier is ZSCL-SL-001, the business attribute tags are "high-value consumables", "requisition process", and "routine", and the associated business rules include "department review" and "consumable management department approval", etc., and the data dimension descriptions include "requisition form number", "requisitioning department", "consumable name", "specification", "quantity", and "requisition date", etc.; in the metadata information of the "emergency requisition processing specifications" knowledge unit, the knowledge unit identifier is ZSCL-SL-002, the business attribute tags are "high-value consumables", "requisition process", and "emergency", and the associated business rules include "response within 2 hours" and "director's expedited approval", etc., and the data dimension descriptions include "urgency level", "estimated usage time", and "contact person and telephone number", etc.

[0064] Step S134: Perform business association matching between the core elements and the metadata information of the knowledge units. During the matching process, preliminary screening is first performed through business attribute tags, and then precise matching is performed through core data dimensions to generate a knowledge unit matching list.

[0065] The core elements are matched with the metadata information of the knowledge units through business association. First, initial screening is performed using business attribute tags. The business attributes related to "emergency application for high-value consumables" in the core elements are matched with the business attribute tags of the knowledge units: "high-value consumables," "application process," and "emergency," resulting in knowledge units such as "high-value consumables application process" and "emergency application processing guidelines." Then, precise matching is performed using core data dimensions. Core data dimensions such as "name of consumables applied for," "specifications," "quantity," "urgency level," and "operation time" are matched with the data dimension descriptions of the knowledge units, further filtering out knowledge units with higher matching degrees. For example, the data dimension description of the "emergency application processing guidelines" knowledge unit includes "urgency level" and "estimated usage time," showing a high matching degree with the core data dimensions. Finally, a knowledge unit matching list is generated, containing knowledge units such as "emergency application processing guidelines," "knowledge of commonly used high-value consumables in orthopedics," and "methods for checking high-value consumable inventory."

[0066] Step S135: Based on the overlap of business attributes, data dimension matching degree, and fit of related business rules between core elements and knowledge units, calculate the association strength value of each knowledge unit in the knowledge unit matching list. The higher the association strength value, the stronger the association between the knowledge unit and the search intent.

[0067] For each knowledge unit in the knowledge unit matching list, the association strength value is calculated from three aspects: business attribute overlap, data dimension matching, and the fit of associated business rules. Business attribute overlap is determined by comparing the number and rate of overlap between the core elements and the knowledge unit's business attribute tags; data dimension matching is determined by comparing the data types, formats, and value ranges described in the core data dimensions and the knowledge unit's data dimensions; and the fit of associated business rules is determined by analyzing the logical consistency and applicable scenario overlap between the associated business rules of the core elements and the associated business rules of the knowledge unit. Then, the values ​​of these three dimensions are weighted to obtain the association strength value for each knowledge unit. The higher the association strength value, the stronger the association between the knowledge unit and the search intent.

[0068] Step S1351: Set up the association strength value calculation system. The association strength value calculation system includes three calculation dimensions: business attribute overlap dimension, data dimension matching dimension, and business rule fit dimension. Each dimension is configured with specific calculation indicators.

[0069] A system for calculating association strength values ​​is established, defining three calculation dimensions and their specific metrics. The metrics for the business attribute overlap dimension include the number of overlapping business attribute tags and the overlap rate; the metrics for the data dimension matching dimension include the number of matching data types, the degree of data format matching, and the degree of matching data value ranges; the metrics for the business rule fit dimension include the degree of consistency in rule logic, the degree of overlap in applicable scenarios, and the degree of matching in rule importance. Each metric has a corresponding calculation method and weight to comprehensively calculate the values ​​of each dimension.

[0070] Step S1352: In the dimension of overlapping business attributes, calculate the number and rate of overlap between the business attribute tags of the core elements of the scenario adaptation parameters and the business attribute tags of the knowledge unit metadata, and calculate the degree of overlap of business attributes based on the number and rate of overlap.

[0071] In terms of overlapping business attributes, the business attribute tags in the core elements (such as "high-value consumables," "emergency application," and "orthopedic surgery") are compared with the business attribute tags in the metadata of the knowledge units. For example, the business attribute tags of the knowledge unit "emergency application processing specifications" are "high-value consumables," "application process," and "emergency," which overlap with the business attribute tags of the core elements by 3. Assuming that the core elements have a total of 4 business attribute tags, the overlap rate is 3 / 4 = 0.75. Then, the business attribute overlap degree is calculated based on the number of overlaps and the overlap rate. If the weight of the number of overlaps is set to 0.6 and the weight of the overlap rate is set to 0.4, the business attribute overlap degree = 3 × 0.6 + 0.75 × 0.4 = 1.8 + 0.3 = 2.1 (assuming a maximum score of 3 points).

[0072] Step S1353: In the data dimension matching dimension, extract the core data dimension from the core elements and the metadata dimension description of the knowledge unit, compare the matching of the data type, data format and data value range of the two, and generate the data dimension matching degree.

[0073] Extract core data dimensions from the core elements, such as "urgency level," "operation time," "consumable name," "specification," and "quantity," as well as metadata dimensions from the knowledge unit "Emergency Application Processing Specifications," such as "urgency level (divided into general, urgent, and critical)," "estimated usage time (format YYYY-MM-DDHH:MM)," "consumable name," "specification," and "quantity." Compare the data types: "urgency level" is categorical data, "operation time" and "estimated usage time" are date / time data, "consumable name" and "specification" are string data, and "quantity" is numeric; the data types match. Regarding data format, the date / time format is consistent. Regarding the range of data values, the range of values ​​for "urgency level" matches. Based on these matching criteria, set scores for data type matching, data format matching, and data value range matching, such as 0.9, 1.0, and 0.9 respectively. Then, calculate the average or weighted average to generate the data dimension matching score, such as (0.9+1.0+0.9) / 3=0.933.

[0074] Step S1354: In terms of business rule fit, compare the related business rules in the core elements with the business rules related to the knowledge units, analyze the logical consistency of the business rules and the overlap of applicable scenarios, and generate the fit of related business rules.

[0075] The core elements include related business rules such as "urgent application forms must be responded to within 2 hours" and "department head's review and signature." The knowledge unit "urgent application processing specifications" includes related business rules such as "urgent application response mechanism within 2 hours" and "director's expedited approval process." Comparing these business rules, "urgent application forms must be responded to within 2 hours" and "urgent application response mechanism within 2 hours" are logically consistent and have overlapping applicable scenarios; "department head's review and signature" and "director's expedited approval process" are logically consistent and have overlapping applicable scenarios. Analyzing the logical consistency and overlap of applicable scenarios of the business rules generates the fit score of related business rules. For example, with a logical consistency score of 0.95 and an applicable scenario overlap score of 0.9, the overall score is (0.95 + 0.9) / 2 = 0.925.

[0076] Step S1355: Perform a weighted calculation on the overlap of business attributes, the matching degree of data dimensions, and the fit of associated business rules to obtain the association strength value of each knowledge unit.

[0077] The weights for business attribute overlap, data dimension matching, and the fit of associated business rules are set to 0.4, 0.3, and 0.3, respectively. For the knowledge unit "Emergency Application Processing Specifications," the business attribute overlap is 2.1 (assuming a maximum score of 3, normalized to 0.7), the data dimension matching is 0.933, and the fit of associated business rules is 0.925. Therefore, the association strength value is 0.8374. Using a similar calculation method, the association strength value for each knowledge unit in the knowledge unit matching list is obtained.

[0078] Step S1356: Collect the scoring data of medical SPD business experts on historical knowledge unit matching cases, establish a correlation strength value correction model based on the scoring data, input the initially calculated correlation strength value into the correlation strength value correction model, and obtain the corrected final correlation strength value.

[0079] For example, step S13561: Construct a historical case database, which stores historical search requests in medical SPD business, corresponding scenario adaptation parameters, a list of matched knowledge units, preliminarily calculated association strength values, and scoring data of medical SPD business experts.

[0080] When building the historical case database, historical search requests from the past three years of medical SPD business are collected, such as "monthly requisition of general internal medicine consumables" and "emergency allocation of surgical instruments". Each search request corresponds to storage scenario adaptation parameters, such as the business process identifier "requisition-002" and the core data dimensions "requisition quantity and department code"; the matching knowledge unit list, such as "general consumables requisition process" and "monthly usage statistics standard"; the preliminary calculated association strength values, such as 0.78 and 0.65; and the scoring data of medical SPD business experts on the reasonableness of the matching of knowledge units in each case, with a scoring range of 0 to 10 points. For example, the score for "general consumables requisition process" is 9 points, and the score for "monthly usage statistics standard" is 8 points.

[0081] Step S13562: Preprocess the data in the historical case database, normalize the initially calculated association strength value, use the preprocessed initial association strength value as the input feature, use the expert score as the target value, and divide the data into training set, validation set and test set. The training set is used for model training, the validation set is used to adjust the model parameters, and the test set is used to evaluate the model performance.

[0082] The data in the historical case database is preprocessed by normalizing the initially calculated association strength values ​​(range 0 to 1) to make them comparable to expert scores (0 to 10) in numerical range. This can be achieved, for example, by using the formula "Normalized Association Strength Value = Initial Association Strength Value × 10". The dataset is then divided into training, validation, and test sets in a 7:2:1 ratio. The training set contains a large number of normalized association strength values ​​(input features) and corresponding expert scores (target values) from historical cases, used to train the association strength value correction model. The validation set is used to adjust the model's hyperparameters, such as the learning rate and regularization coefficient, during training. The test set is used to evaluate the predictive performance of the model after training.

[0083] Step S13563: Select the gradient boosting regression algorithm to construct the correlation strength value correction model. This gradient boosting regression algorithm is used to capture the nonlinear relationship between input features and target values.

[0084] The gradient boosting regression algorithm is chosen to construct a correlation strength value correction model. This algorithm iteratively generates multiple weak regression models (such as decision trees) and weights and combines the predictions of these weak models to capture the non-linear relationship between input features (normalized correlation strength values) and target values ​​(expert ratings). For example, for cases with high initial correlation strength values ​​but low expert ratings, the algorithm can learn this bias and correct it, possibly because although the knowledge unit matches the surface features of the retrieval intent, it lacks deep business logic connections.

[0085] Step S13564: Train the association strength value correction model using the training set data. During the training process, monitor the prediction error of the association strength value correction model using the validation set data. Adjust the hyperparameters of the association strength value correction model using cross-validation to minimize the prediction error of the association strength value correction model.

[0086] The model is trained using the training set data, with an initial learning rate of 0.1, a tree depth of 3, and 100 iterations. During training, the prediction error (e.g., mean squared error) is calculated using the validation set data every 10 iterations. If the prediction error does not decrease, the hyperparameters are adjusted using 5-fold cross-validation, such as reducing the learning rate to 0.05, increasing the tree depth to 5, and increasing the number of iterations to 200, until the prediction error on the validation set is minimized, at which point the model parameters reach their optimal state.

[0087] Step S13565: Use test set data to evaluate the performance of the trained association strength value correction model. The evaluation metrics include mean absolute error, root mean square error and coefficient of determination. Deploy the association strength value correction model that meets the evaluation criteria to the knowledge unit matching system.

[0088] The model performance is evaluated using test set data. The mean absolute error (MAE) is calculated, which is the average of the absolute differences between the model's predicted correction value and the expert score; the root mean square error (RMSE) is the square root of the mean square error; and the coefficient of determination (R²) represents the model's ability to explain data variability. If MAE is less than 0.5, RMSE is less than 0.8, and R² is greater than 0.85, the model is considered to have passed the evaluation. The model with the corrected association strength values ​​is then deployed to the knowledge unit matching system to correct the initially calculated association strength values.

[0089] Step S1357: Compare the final association strength value with the preset association strength threshold, and filter out knowledge units whose final association strength value is higher than the association strength threshold and retain them in the knowledge unit matching list.

[0090] A preset association strength threshold, such as 0.7, is set. The final association strength value of each knowledge unit is compared with this threshold. The final association strength value of the "Emergency Application Processing Specification" knowledge unit is 0.8374, which is higher than the threshold, so it is retained in the knowledge unit matching list. The final association strength value of the "High-Value Consumables Inventory Query Method" knowledge unit is 0.65, which is lower than the threshold, so it is removed from the list. This filtering process ensures that the knowledge units in the knowledge unit matching list have a strong relevance to the search intent.

[0091] Step S136: Construct the initial structure of the dynamic association network. The initial structure includes three layers: the first layer is the retrieval intent layer, which stores the core text fragments of the retrieval request and the retrieval intent feature vector; the second layer is the core element layer, which stores the core elements and weight ratios of the scenario adaptation parameters; and the third layer is the knowledge unit layer, which stores the knowledge units in the knowledge unit matching list that meet the association strength value and the association strength value.

[0092] The initial structure of the dynamic association network is divided into three layers. The first layer, the retrieval intent layer, stores the core text fragment of the retrieval request: "There are three surgeries in the orthopedic ward tomorrow morning, and we urgently need to apply for a batch of high-value consumables, including artificial joints and bone screws," along with the generated retrieval intent feature vector. The second layer, the core element layer, stores the core elements in the scenario adaptation parameters, such as business process identifiers "YGSC001-01" and "YGSC001-02," core data dimensions such as "name of consumables to be applied for," "specifications," "quantity," "urgency level," and "surgery time," as well as related business rules, and also stores the weight percentage of each core element. The third layer, the knowledge unit layer, stores knowledge units in the knowledge unit matching list that meet the association strength value criteria, such as "emergency application processing specifications" and "knowledge of commonly used high-value consumables in orthopedics," along with their corresponding association strength values.

[0093] Step S137: Establish association links between the three levels. The retrieval intent layer and the core element layer are associated through business feature words, and the core element layer and the knowledge unit layer are associated through business association matching relationships. Each association link is marked with the corresponding association basis and strength value.

[0094] A link is established between the three levels. The retrieval intent layer and the core element layer are linked through business feature words. For example, the core text fragment "urgent application for high-value consumables" in the retrieval intent layer is connected to the core data dimension "urgency level" and the business process identifier "YGSC001-01" in the core element layer through business feature words such as "high-value consumables," "application," and "urgent." The core element layer and the knowledge unit layer are linked through business association matching relationships. For example, the core element "urgency level" is connected to the knowledge unit "urgent application processing specifications," and the core element "orthopedic surgery" is connected to the knowledge unit "knowledge of commonly used high-value consumables in orthopedics." Each link is marked with the association basis, such as "business feature word matching" and "data dimension matching," as well as a strength value. This strength value is determined based on the weight ratio of the core element and the association strength value of the knowledge unit.

[0095] Step S138: Dynamically optimize the associated links in the initial structure, merge duplicate links pointing to the same knowledge unit, strengthen the connection stability of links with an association strength value greater than the set association strength value, and remove invalid links with an association strength value lower than the set standard.

[0096] Dynamically optimize the associated links in the initial structure. If multiple associated links point to the same knowledge unit, such as the retrieval intent layer connecting to multiple core elements in the core element layer through different business feature words, and these core elements are all connected to the same knowledge unit, merge these duplicate links to avoid network redundancy. For links with an association strength value greater than a set association strength value (e.g., 0.8), strengthen their connection stability by increasing the link weight or marking them with special identifiers; for invalid links with an association strength value lower than a set standard (e.g., 0.5), remove them from the network to improve the accuracy and efficiency of the dynamic association network.

[0097] Step S139: Add network dynamic update trigger conditions. When the business context of the retrieval request changes or the knowledge unit in the medical SPD vector knowledge base is updated, the reconstruction of the associated network is automatically triggered, and the node information and associated links of the dynamic associated network are updated synchronously, so that the content of the dynamic associated network corresponds to the current business requirements.

[0098] Add dynamic network update trigger conditions. When the business context of a retrieval request changes, such as adjustments to surgery time or changes in the required quantity of consumables, the dynamic network reconstruction will be automatically triggered. Similarly, when knowledge units in the medical SPD vector knowledge base are updated, such as changes to the high-value consumables requisition process or the addition of new orthopedic-specific high-value consumables knowledge units, network reconstruction will also be triggered. During reconstruction, the node information (e.g., adding or deleting knowledge unit nodes) and associated links (e.g., adjusting link strength values, adding or removing links) of the dynamic network will be updated synchronously based on the new business context or knowledge unit updates, ensuring that the content of the dynamic network always corresponds to current business needs.

[0099] For example, step S1391: Define the criteria for judging changes in the business context of the retrieval request. The circumstances of changes in the business context include changes in the type of business document currently operated by the user, adjustments to the associated medical departments, changes in the urgency of the business, and the addition of required data dimensions.

[0100] The criteria for judging changes in the business context of a search request are clearly defined. It is stipulated that when the type of the business document currently operated by the user changes from "high-value consumables requisition form" to "ordinary consumables requisition form", the associated medical department changes from "orthopedics" to "internal medicine", the business urgency level changes from "urgent" to "critical", or "consumables manufacturer" is added to the required data dimension, the business context is judged to have changed.

[0101] Step S1392: Set up a context monitoring module in the medical SPD business terminal. The context monitoring module collects user operation behavior data and business document status data in real time. It compares the collected operation behavior data and business document status data with the initial business context data. When the difference in the comparison result exceeds the preset context change threshold, it is determined that the business context has changed.

[0102] A context monitoring module is installed on the medical SPD (Service Provider Device) business terminal. This module collects user operation behavior data in real time, such as clicks, inputs, and modifications on the terminal, as well as business document status data, such as document completion progress and approval status. The collected data is compared with the initial business context data, and the difference value is calculated. A preset context change threshold is set. If the difference value reaches 20%, when the difference in the comparison result exceeds this threshold, it is determined that the business context has changed, triggering the reconstruction of the dynamic association network.

[0103] Step S1393: Define trigger events for updating knowledge units in the medical SPD vector knowledge base. The trigger events include the addition of new knowledge units, modification of existing knowledge unit content, updating of knowledge unit metadata, and adjustment of knowledge unit association relationships.

[0104] Define trigger events for updating knowledge units in the medical SPD vector knowledge base. These events are triggered when: a new knowledge unit is added to the knowledge base, such as adding a new knowledge unit called "Guidelines for the Use of New Orthopedic High-Value Consumables"; the content of an existing knowledge unit is modified, such as changing the response time in "Emergency Application Processing Specifications" from 2 hours to 1.5 hours; the metadata information of a knowledge unit is updated, such as adding "New Materials" to the business attribute tag of "Knowledge of Commonly Used High-Value Orthopedic Consumables"; or the relationship between knowledge units is adjusted, such as establishing a new relationship between the "High-Value Consumables Application Process" knowledge unit and the "Financial Reimbursement Rules" knowledge unit.

[0105] Step S1394: Set up an update monitoring engine in the medical SPD vector knowledge base. The monitoring engine listens to the operation logs of knowledge units in real time. When an operation that meets the trigger event is detected, the update identifier, update type and update time of the knowledge unit are recorded.

[0106] An update monitoring engine is set up in the medical SPD vector knowledge base. This engine listens to various operation logs of knowledge units in real time, including operations such as adding, modifying, deleting, and updating metadata. When an operation that matches the knowledge unit update trigger event is detected, the engine immediately records the knowledge unit's identifier, such as the identifier of the knowledge unit "Guide to the Use of New Orthopedic High-Value Consumables"; the update type, such as "add", "modify content", "update metadata", etc.; and the update time, accurate to the second.

[0107] Step S1395: Construct a network dynamic update decision module. This network dynamic update decision module receives change signals from the context monitoring module and update signals from the update monitoring engine, and determines whether the reconstruction of the dynamic association network needs to be triggered based on the signal type.

[0108] The network dynamic update decision module simultaneously receives business context change signals from the context monitoring module and knowledge unit update signals from the update monitoring engine. It determines whether to trigger a reconfiguration of the dynamically associated network based on the signal type. For example, when a significant business context change signal is received, or when an update signal for an important knowledge unit is received, the decision module determines that a reconfiguration needs to be triggered; when the signal's impact is minor, reconfiguration may not be triggered immediately, waiting for a certain amount of accumulated changes before proceeding.

[0109] Step S1396: Configure an update priority mechanism. When business context changes and knowledge unit updates occur simultaneously, prioritize responding to the refactoring request triggered by the business context change. If the overlap of the knowledge domains involved in the two reaches the set overlap threshold, then merge and trigger a refactoring.

[0110] A configuration update priority mechanism is implemented. When business context changes and knowledge unit updates occur simultaneously, the refactoring request triggered by the business context change is prioritized because it directly reflects the current user's immediate needs. If the overlap between the knowledge domain involved in the business context change and the knowledge domain involved in the knowledge unit update reaches a set overlap threshold, such as 80%, the two refactoring requests are merged into a single refactoring request to reduce the number of network refactoring operations and improve efficiency.

[0111] Step S1397: Activate the reconstruction process of the dynamic association network. During the reconstruction process, retain the core structure and effective association links of the original dynamic association network, and only adjust the parts affected by the changes. After the reconstruction is completed, compare the new dynamic association network with the original dynamic association network to generate a network update report. The network update report includes the updated nodes, the adjusted association links, and a description of the differences before and after the update.

[0112] A dynamic network reconstruction process is implemented, preserving the core structure of the original network, such as the three-tier division and valid interconnected links unaffected by changes. Adjustments are made only to parts affected by changes in business context or knowledge unit updates, such as adding or deleting knowledge unit nodes, adjusting the strength values ​​or connection relationships of interconnected links. After reconstruction, the new dynamic network is compared with the original network, generating a network update report. The report details the updated nodes (added, deleted, or modified knowledge unit nodes), adjusted interconnected links (added, deleted, or strength-adjusted links), and explanations of the differences in network structure and interconnected relationships before and after the update, facilitating subsequent tracking and management of network changes.

[0113] Step S140: Input the dynamic association network into the medical SPD business enhancement model to generate a knowledge enhancement set. The medical SPD business enhancement model is jointly trained with medical SPD business data and scenario cases, and has the ability to complete knowledge and deduce business logic.

[0114] The constructed and optimized dynamic association network is input into the large-scale medical SPD business enhancement model. This model is jointly trained based on a large amount of medical SPD business data and scenario cases. Through training, the model acquires knowledge completion and business logic deduction capabilities. For example, when a knowledge unit in the dynamic association network lacks certain details, the model can complete it based on existing knowledge and business logic; when it is necessary to deduce subsequent business process steps or potential problems and solutions based on existing knowledge units, the model can perform logical deduction. Through processing the dynamic association network, the model generates a knowledge enhancement set, which includes the completed knowledge units, business logic association descriptions, and other content.

[0115] Step S141: Read the hierarchical structure, node information and association link data of the dynamic association network, convert the hierarchical structure, node information and association link data of the dynamic association network into tensor data format that can be recognized by the medical SPD business enhancement big model, and obtain the converted tensor data. During the conversion process, the association relationship and strength information between nodes are preserved.

[0116] First, the hierarchical structure of the dynamic network is read, including the division into the retrieval intent layer, core element layer, and knowledge unit layer, and the number of nodes in each layer; node information, such as the identifier, attribute description, and weight ratio of each node; and link data, such as the connecting nodes, association basis, and strength values ​​of the links. Then, the above data is converted into a tensor data format recognizable by the medical SPD business enhancement model. During the conversion process, the node information and link data need to be mapped to elements in the tensor, while ensuring that the association relationships and strength information between nodes are accurately preserved through the dimensions and values ​​of the tensor, so that the model can correctly parse and process the information of the dynamic network.

[0117] Step S142: Input the transformed tensor data into the business understanding module of the medical SPD business enhancement model. This business understanding module contains a medical SPD business semantic encoding unit. Through semantic encoding, it captures the deep business relationships between retrieval intent, core elements and knowledge units, outputs a business relationship feature matrix, and passes it to the knowledge completion module of the medical SPD business enhancement model. The knowledge completion module calls the built-in medical SPD business knowledge graph, which stores medical SPD business entities, relationships and attribute information in the form of triples.

[0118] The transformed tensor data is input into the business understanding module of the medical SPD business enhancement model. The medical SPD business semantic encoding unit within the business understanding module processes the tensor data, using semantic encoding technology to convert the textual information of retrieval intent, core elements, and knowledge units into computer-understandable vector representations, capturing the deep business relationships between them. For example, the encoding unit can identify the causal relationship between "urgent application for high-value consumables" and "urgent approval process," and the usage relationship between "orthopedic surgery" and "artificial joints." After encoding, a business relationship feature matrix is ​​output, where the element values ​​reflect the strength and type of the relationship between different nodes. The business relationship feature matrix is ​​then passed to the knowledge completion module, which calls the built-in medical SPD business knowledge graph. This knowledge graph stores business entities (such as "high-value consumables," "artificial joints," and "department head"), relationships between entities (such as "belongs to," "requires approval," and "used for"), and entity attribute information (such as "specifications" and "approval authority") in the form of triples (entity-relationship-entity).

[0119] Step S143: Compare the business association feature matrix with the medical SPD business knowledge graph to identify missing information and logical gaps in the knowledge units of the dynamic association network, and supplement the missing business knowledge and repair the logical gaps based on the association relationships in the medical SPD business knowledge graph.

[0120] Step S1431: Read the business association feature matrix, extract the knowledge unit feature vector and the association relationship features between knowledge units contained therein, and convert the knowledge unit feature vector and the association relationship features between knowledge units into a knowledge graph query statement. The query statement contains the entity and association relationship type corresponding to the knowledge unit.

[0121] In the emergency requisition scenario for high-value orthopedic consumables, the business association feature matrix, after extraction, yields knowledge unit feature vectors, such as the feature vector for "Emergency Requisition Processing Specification," and inter-knowledge unit relationship features, such as the relationship between "Emergency Requisition Processing Specification" and "Approval Process." These features are then converted into knowledge graph query statements. For example, for the knowledge unit "Emergency Requisition Processing Specification," the query statement might be "MATCH(n:KnowledgeUnit{name:'Emergency Requisition Processing Specification'})-[:HAS_ATTRIBUTE]->(a:Attribute)RETURNn.name,a.name,a.value," where "KnowledgeUnit" is the entity type and "HAS_ATTRIBUTE" is the relationship type. This aims to obtain the attribute information of this knowledge unit from the medical SPD business knowledge graph.

[0122] Step S1432: Call the query interface of the medical SPD business knowledge graph, execute the query statement, obtain the entity, relationship and attribute information associated with the knowledge unit in the knowledge graph, and form a knowledge graph association information set.

[0123] The query interface of the medical SPD business knowledge graph is invoked, which supports standard graph query language. After executing the above query, the interface returns entities in the knowledge graph associated with "Emergency Application Processing Specifications", such as "Emergency Contact Person", "Approval Authority", and "Processing Time Limit"; relationships such as "Required" (the Emergency Application Processing Specifications require an emergency contact person) and "Included" (the Emergency Application Processing Specifications include a processing time limit); attribute information such as the emergency contact person's name "Director Zhang", contact number "138XXXX5678", approval authority "department director and above", and processing time limit "respond within 2 hours". This information together constitutes the set of associated information in the knowledge graph.

[0124] Step S1433: Compare the knowledge graph association information set with the knowledge unit information in the dynamic association network. The attribute information that exists in the medical SPD business knowledge graph but is missing in the knowledge unit is the missing information of the knowledge unit.

[0125] The information set associated with the knowledge graph was compared with the "Emergency Application Processing Specifications" knowledge unit information in the dynamic network. The original information of this knowledge unit in the dynamic network included processing steps and approval nodes, but did not mention the emergency contact person or specific contact number. However, the information set associated with the knowledge graph contained the attribute information "Emergency Contact Person: Director Zhang, Contact Number: 138XXXX5678". Therefore, it was determined that this knowledge unit was missing information, namely the emergency contact person and contact number.

[0126] Step S1434: Analyze the business logic relationships between knowledge units in the dynamic association network, and configure the logical relationship judgment criteria based on the medical SPD business process specifications. The logical relationship judgment criteria include the preset logical association types between knowledge units in different business links.

[0127] Analyze the business logic relationships between knowledge units in the dynamic network. For example, the "Submit Application Form" knowledge unit should be followed by the "Department Review" knowledge unit, and "Department Review" should be followed by the "Consumables Management Department Approval" knowledge unit. Based on the medical SPD business process specifications, configure logical relationship judgment criteria and preset logical association types including "Sequential Execution," "Parallel Execution," and "Conditional Branch." Among them, "Sequential Execution" requires that the previous knowledge unit must be completed before the next knowledge unit can be started, while "Parallel Execution" allows two knowledge units to be executed simultaneously.

[0128] Step S1435: Based on the logical relationship judgment criteria, check whether the association links between knowledge units in the dynamic association network are complete. If the knowledge units of adjacent business links lack the necessary logical association, or the association relationship does not conform to the medical SPD business process specification, then it is determined that there is a logical break in the dynamic association network.

[0129] Based on the logical relationship judgment criteria, examine the related links in the dynamic association network. For example, according to the business process specifications, there should be an intermediate link between the "department review" knowledge unit and the "warehouse issuance" knowledge unit, forming a sequential execution relationship of "department review → consumables management department approval → warehouse issuance". However, in the current dynamic association network, "department review" directly points to "warehouse issuance", lacking the association of the "consumables management department approval" knowledge unit, and the association relationship does not conform to the preset "sequential execution" type. Therefore, it is determined that there is a logical break.

[0130] Step S1436: Classify and organize the identified missing information, and divide it into basic attribute missing, related attribute missing and extended attribute missing according to attribute type. Different types of missing information correspond to different completion priorities.

[0131] The identified missing information is categorized and organized as follows: Basic attribute missing refers to the missing core attributes upon which the knowledge unit depends, such as the missing processing time limit attribute in the "Emergency Application Processing Specification"; Related attribute missing refers to the missing attributes required for the knowledge unit to connect with other entities, such as the missing emergency contact person attribute; Extended attribute missing refers to attributes that are auxiliary to business execution but not essential, such as historical processing cases. In this scenario, the emergency contact person and contact number are related attribute missing, while the processing time limit is a basic attribute missing. The completion priority is set as follows: basic attribute missing takes precedence over related attribute missing, and related attribute missing takes precedence over extended attribute missing.

[0132] Step S1437: Locate the business process where the logical break occurs, record the knowledge units involved and the types of missing logical connections, and generate an identification report of missing information and logical break. The identification report includes knowledge unit identifiers, missing information types, logical break locations, and repair suggestions.

[0133] The business process where the logical break occurs is identified as the "approval process," involving the knowledge units "departmental review" and "warehouse issuance." The missing logical association type is the "consumables management department approval" knowledge unit within the "sequential execution" relationship. An identification report is generated, with the knowledge units identified as "departmental review" (KU001) and "warehouse issuance" (KU003). Missing information types include missing basic attributes (processing time limit) and missing association attributes (emergency contact person) for the "emergency application processing specifications." The logical break location is described as "a missing approval node between KU001 and KU003." The suggested repair is to "supplement the 'consumables management department approval' knowledge unit (KU002) and establish a sequential execution association of KU001→KU002→KU003."

[0134] Step S144: Input the completed knowledge unit data into the logic deduction module of the model, and call the medical SPD business process specification library. The medical SPD business process specification library stores business process standards, which cover the entire process from material procurement to cost settlement. Each specification item includes process steps, operation requirements and logical relationship descriptions.

[0135] The completed knowledge unit data is input into the model's logical deduction module. The logical deduction module calls the Medical SPD Business Process Standard Library, which stores detailed business process standards, covering the entire process from material procurement planning, supplier selection, and purchase order issuance, to material receiving and acceptance, inventory management, consumable requisition and distribution, and finally, cost accounting and settlement. Each standard item includes specific process steps, such as "submitting a purchase request," "procurement department review," and "signing a contract with the supplier"; operational requirements, such as "the purchase request must specify the material name, specifications, quantity, and budget amount"; and logical relationship explanations, such as "a purchase order can only be issued after the purchase request has been approved."

[0136] Step S145: Extract the business process identifiers and operation descriptions from the completed knowledge units, match the business process identifiers and operation descriptions from the completed knowledge units with the specification entries in the business process specification library, and determine the process position and logical rules to be followed for each knowledge unit.

[0137] Extract the business process identifiers from the completed knowledge unit, such as "YGSC001-01" (submit requisition form) and "YGSC001-02" (department review), as well as operation description information, such as "fill out the high-value consumables requisition form" and "department head's signature review." Match the above information with the standard entries in the business process standard library. For example, the "submit requisition form" business process identifier and its corresponding operation description information match the "submit requisition form" process step in the "high-value consumables requisition process" standard entry in the standard library. This determines that the process position corresponding to this knowledge unit is the starting step of the high-value consumables requisition process and clarifies the logical rules to be followed, such as "the requisition form must be filled out completely without any missing items" and "the consumables inventory status must be confirmed before submission."

[0138] Step S146: Construct a knowledge unit logical relationship matrix. The rows and columns of the knowledge unit logical relationship matrix correspond to knowledge units. The matrix element values ​​represent the logical relationship type between the knowledge unit corresponding to the row and the knowledge unit corresponding to the column. The logical relationship type includes pre-relationship, post-relationship, parallel relationship, or no direct relationship.

[0139] Construct a logical relationship matrix for knowledge units, where rows and columns correspond to the completed knowledge units. For example, rows correspond to the "Submit Application Form" knowledge unit, and columns correspond to the "Department Review" knowledge unit. Matrix element values ​​represent the type of logical relationship between the two. Since "Department Review" must be performed after "Submit Application Form," they are prerequisites, and the element value is marked as "Prerequisite." The "Department Review" knowledge unit is also prerequisite to the "Consumables Management Department Approval" knowledge unit; the "Warehouse Outbound" knowledge unit is prerequisite to the "Delivery to Operating Room" knowledge unit; while the "Fill in Application Form" and "Confirm Inventory" knowledge units may be performed simultaneously, belonging to a parallel relationship, and the element value is marked as "Parallel"; knowledge units without direct business logic connections are marked as "No Direct Relationship."

[0140] Step S147: Verify the rationality of the knowledge unit logical relationship matrix according to the business process standard. If the logical relationship recorded in the knowledge unit logical relationship matrix does not match the process order specified in the specification, mark it as a logical anomaly and correct the knowledge unit relationship marked as a logical anomaly, and adjust the logical relationship type in the knowledge unit logical relationship matrix.

[0141] The rationality of the knowledge unit logical relationship matrix is ​​verified based on the business process standards in the business process specification library. For example, if the business process standard clearly stipulates that "consumables management department approval" should be carried out after "departmental review," and the relationship between "consumables management department approval" and "departmental review" in the knowledge unit logical relationship matrix is ​​incorrectly marked as "parallel," it is inconsistent with the standard process order and is marked as a logical anomaly. For knowledge unit relationships marked as logical anomalies, corrections are made according to the business process standard, adjusting the "parallel" relationship to a "preceding" relationship to ensure that the logical relationship types in the knowledge unit logical relationship matrix conform to the business process specification.

[0142] Step S148: Sort the knowledge units according to the chronological order of the medical SPD business process to generate a knowledge association sequence. Each knowledge unit in the knowledge association sequence is labeled with its preceding and succeeding knowledge units. The result is then output to the feature enhancement module of the medical SPD business enhancement model. This feature enhancement module uses an attention mechanism to focus on the features of knowledge units that are associated with the retrieval intent, obtains the focused knowledge unit features, and performs fusion processing on the focused knowledge unit features. The fused feature data is then organized according to the order of the business links to generate a knowledge enhancement feature vector. Each knowledge enhancement feature vector corresponds to a business knowledge module.

[0143] Following the sequence of the medical SPD (Supply, Processing, and Distribution) business process, knowledge units are ordered to generate a knowledge association sequence. For example, the knowledge association sequence could be: "Submit Application Form" (Precedence: None, Subsequent: Departmental Review) → "Departmental Review" (Precedence: Submit Application Form, Subsequent: Consumables Management Department Approval) → "Consumables Management Department Approval" (Precedence: Departmental Review, Subsequent: Warehouse Issuance) → "Warehouse Issuance" (Precedence: Consumables Management Department Approval, Subsequent: Delivery to Operating Room) → "Delivery to Operating Room" (Precedence: Warehouse Issuance, Subsequent: None). The knowledge association sequence is output to the feature enhancement module. This module uses an attention mechanism, assigning different attention weights to knowledge unit features based on the retrieval intent feature vector. It focuses on knowledge unit features highly relevant to the retrieval intent of "urgent application for high-value consumables," such as features related to "urgent application form completion guidelines" and "urgent approval process," which receive higher weights. The focused knowledge unit features are fused together, integrating the feature information of different knowledge units and organizing the fused feature data according to the business process sequence to generate knowledge-enhanced feature vectors. Each knowledge-enhanced feature vector corresponds to a business knowledge module, such as "emergency application process module" and "high-value consumables information module".

[0144] Step S149: Convert the knowledge-enhanced feature vector into a structured knowledge-enhanced set. The knowledge-enhanced set includes the completed knowledge unit, business logic association description, knowledge source identifier and association strength description. Each knowledge unit in the knowledge-enhanced set is accompanied by a corresponding business application scenario description.

[0145] The knowledge-enhanced feature vectors are transformed into structured knowledge-enhanced sets. These sets contain completed knowledge units, such as "Emergency Requisition Processing Guidelines (including emergency contact information)" and "List of Commonly Used High-Value Consumables in Orthopedics"; business logic association descriptions, such as "After submitting a requisition form, it must be reviewed by the department head before proceeding to the consumables management department's approval process"; knowledge source identifiers, such as the knowledge unit originating from "Medical SPD Business Process Specification Library" and "High-Value Consumables Management Manual V2.0"; and association strength descriptions, such as "High association strength with search intent" and "Medium association strength." Each knowledge unit also includes a corresponding business application scenario description, such as "Applicable to emergency shortages of high-value consumables in orthopedic surgery and other situations," making the knowledge-enhanced set clearer and more practical.

[0146] Step S150: Generate a medical SPD business execution plan based on the knowledge enhancement set, and push the business execution plan to the medical SPD business terminal. The business execution plan includes operation steps, related knowledge basis, and data support sources.

[0147] Medical SPD (Service Provider Development) business execution plans are generated based on knowledge-enhanced sets. In the emergency requisition scenario for high-value orthopedic consumables, the business execution plan details each step from submitting the requisition form to the delivery of the consumables to the operating room. Each step is annotated with the associated knowledge basis, such as "Submitting the requisition form must be done in accordance with the emergency requisition form filling specifications (from the knowledge unit 'Emergency Requisition Form Filling Specifications')," as well as the data support sources, such as "Current inventory data of high-value orthopedic consumables (from real-time data of the inventory management system)" and "Historical emergency requisition processing time records (from the business database)." The generated business execution plan is pushed to the medical SPD business terminal, allowing orthopedic ward staff to execute the emergency requisition operation for high-value consumables according to the plan.

[0148] Step S151: Parse the knowledge enhancement set, extract the knowledge units, business logic association descriptions and business application scenario descriptions, and generate a business execution element list. The business execution element list includes the operation object, operation action, operation basis and operation sequence requirements.

[0149] The knowledge enhancement set is analyzed, and knowledge units are extracted, such as "Emergency Requisition Form Filling Specifications," "Departmental Review Process," and "Warehouse Outbound Operation Guidelines." Business logic connections are explained, such as "The requisition form must be submitted for departmental review immediately after completion." Business application scenarios are described, such as "Applicable to emergency consumable requisitions within 24 hours before surgery." Based on this information, a list of business execution elements is generated. The operation objects in the list include "Emergency Requisition Form," "Department Head," "Consumables Management Department," and "Warehouse Manager." Operation actions include "Fill out," "Submit," "Review," "Approval," "Outbound," and "Delivery." Operational basis includes "Emergency Requisition Form Filling Specifications" and "Departmental Review Authority Regulations." Operational timing requirements include "Requisition form filling must be completed within 10 minutes" and "Departmental review must be completed within 30 minutes."

[0150] Step S152: Call the medical SPD business process template library. The medical SPD business process template library contains standard process templates for all business processes of medical SPD. Each standard process template corresponds to a set business scenario. The standard process template clearly defines the order and relationship of the operation steps.

[0151] The Medical SPD Business Process Template Library covers standard process templates for all aspects of medical SPD business, such as templates for routine requisition of high-value consumables, procurement of ordinary consumables, and inventory of consumables. Each standard process template corresponds to a specific business scenario; for example, there is a "High-Value Consumables Emergency Requisition Process Template" for the emergency requisition scenario. The standard process templates clearly define the sequence of operation steps, such as "fill out requisition form → submit for review → superior approval → warehouse issuance → delivery and receipt," as well as the relationships between steps. For example, "submit for review" is a subsequent step after "fill out requisition form" and a prerequisite step for "superior approval."

[0152] Step S153: Based on the business application scenario description in the knowledge enhancement set, match the corresponding standard process template, fill the elements in the business execution element list into the standard process template, and generate a draft of the business execution plan.

[0153] Based on the business application scenario description in the knowledge enhancement set, "applicable to emergency shortages of high-value consumables such as orthopedic surgery," the corresponding "Emergency Application Process Template for High-Value Consumables" is matched in the medical SPD business process template library. Elements such as the operation object, operation action, operation basis, and operation sequence requirements from the business execution element list are filled into the corresponding positions in the standard process template. For example, in the "Fill out application form" step, the operation object is filled with "Emergency Application Form," the operation action is "Fill out," the operation basis is "Emergency Application Form Filling Specifications," and the operation sequence requirement is "Complete within 10 minutes," generating a draft of the business execution plan.

[0154] Step S154: Optimize the initial draft of the business execution plan to obtain the optimized business execution plan.

[0155] The initial draft of the business execution plan underwent process optimization to improve efficiency and accuracy. The optimization process included checking the logical order of steps, the timing of operations, and the presence of redundant steps. For example, it was found that the operation of "contacting the warehouse to prepare consumables" could be performed in parallel between the "department review" and "consumables management department approval" steps to save time; this operation was added as a parallel step. The timing of the "warehouse outbound" and "delivery to the operating room" operations was adjusted so that delivery would begin immediately after outbound processing, reducing waiting time. Through these optimization measures, an optimized business execution plan was obtained.

[0156] Step S1541: Extract all operation steps in the initial draft of the business execution plan, and organize the operation object, operation action, expected input and expected output information contained in each operation step into a step information table. Each operation step contains operation object, operation action, expected input and expected output information.

[0157] Extract all operational steps from the initial draft of the business execution plan, such as "fill out an emergency requisition form," "submit for departmental review," "department head review," "submit for approval by the consumables management department," "consumables management department approval," "warehouse issuance," and "delivery to the operating room." Organize the operational objects, actions, expected inputs, and expected outputs for each step into a step information table. For example, the operational object of the "fill out an emergency requisition form" step is "emergency requisition form," the action is "fill out," the expected input is "surgical information, required consumables information, urgency level," and the expected output is "a completed emergency requisition form." Similarly, the operational object of the "department head review" step is "department head," the action is "review," the expected input is "a completed emergency requisition form," and the expected output is "approval / disapproval comments and signature."

[0158] Step S1542: Configure the operation step sequence relationship model. This operation step sequence relationship model is based on the sequential rules of the medical SPD business process. It has built-in preconditions and postconditions between different operation actions. The preconditions are the input content required to perform the operation action, and the postconditions are the output content generated after the operation action is completed.

[0159] The operational step sequence model is built upon the sequential rules of medical SPD (Service Process Development) business processes. It incorporates preconditions and postconditions for different operational actions. Preconditions specify the input required to perform the action; for example, the precondition for the "Department Head Review" action is "A completed emergency requisition form exists." Postconditions specify the output generated after the action is completed; for example, the postcondition for the "Department Head Review" action is "Generation of approval / disapproval comments and signatures." Through these preconditions and postconditions, the model can determine whether the temporal relationship between operational steps is reasonable.

[0160] Step S1543: Input the operation steps in the step information table into the operation step time sequence relationship model, analyze the preconditions and postconditions of each step, determine whether the logical relationship between the steps is reasonable, and generate the time sequence relationship analysis results.

[0161] The operation steps in the step information table are input into the operation step sequence relationship model. The model analyzes the preconditions and postconditions of each step one by one. For example, the precondition of the "Submit to Department for Review" step is "Complete Emergency Requisition Form," which corresponds to the postcondition "Complete Emergency Requisition Form" of the "Complete Emergency Requisition Form" step, showing a reasonable logical connection. Similarly, the precondition of the "Consumables Management Department Approval" step is "Emergency Requisition Form Approved by Department," which corresponds to the postcondition "Approval Opinion and Signature" of the "Department Head Review" step, also showing a reasonable logical connection. Through analysis, the sequence relationship analysis results are generated, identifying the logically reasonable parts and potentially problematic parts between the steps.

[0162] Step S1544: Based on the timing relationship analysis results, adjust the order of operation steps, arrange steps with pre- and post-step relationships in logical order, so that the expected output of the previous step corresponds to the expected input of the next step.

[0163] Based on the temporal relationship analysis, the order of operation steps was adjusted. For steps with clear pre- and post-step relationships, they were arranged logically to ensure that the expected output of the previous step could serve as the expected input of the next step. For example, the expected output of the "Fill out the emergency requisition form" step is "The completed emergency requisition form," so it should precede the "Submit to department for review" step, because the expected input of the "Submit to department for review" step is the requisition form; the expected output of the "Department head's review" step is "Approval comments and signature," so it should precede the "Submit to consumables management department for approval" step. This adjustment ensures the step order aligns with business logic.

[0164] Step S1545: Identify parallel operation steps. For operation steps that do not have a pre- or post-order relationship and belong to the same business process, set them to parallel execution mode and mark the parallel relationship and the start time node of execution in the business execution plan.

[0165] In the emergency requisition process for high-value consumables, the "department head review" and "contact warehouse manager to confirm inventory" steps do not have a prerequisite or successor relationship and both belong to the "requisition approval preparation" business stage. Therefore, they can be set to execute in parallel. In the business execution plan, these two steps should be marked as parallel and a start time node should be set, such as "while the department head begins review, immediately contact the warehouse manager to confirm inventory," to shorten the overall process execution time.

[0166] Step S1546: Supplement the connection description between steps, which clarifies the specific method of transferring the output content of the previous step to the next step, including the data transfer method, the transfer interface number, and the data verification method.

[0167] The supplementary instructions clarify the transitions between steps, explaining how the output of each step is passed to the next. For example, after completing the "Fill out the emergency requisition form" step, the output "Completed Emergency Requisition Form" is passed to the "Submit to Department for Review" step via the internal data transfer interface of the medical SPD system (interface number: SPD-INT-001), with the transmission method being automatic system push. After completing the "Department Head Review" step, the review result is confirmed through electronic signature and passed to the "Submit to Consumables Management Department for Approval" step via interface SPD-INT-002, with the data verification method being automatic system verification of the validity and integrity of the electronic signature.

[0168] Step S1547: Call the medical SPD business process optimization case library, which stores historical optimized business process solutions. Compare the adjusted business execution solution with similar solutions in the case library, extract the optimization measures from the similar solutions and apply them to the current business execution solution.

[0169] The system utilizes a medical SPD (Service Process Development) case library, which stores historical optimization solutions for various business processes, such as "Optimization of Emergency Requisition Process for High-Value Consumables in Internal Medicine" and "Fast Requisition Solution for Surgical Consumables." The adjusted business execution plan is compared with similar plans in the library to identify highly similar cases, such as "Optimization of Emergency Requisition of High-Value Consumables for Orthopedic Joint Replacement Surgery." Optimization measures from these similar plans are extracted, such as "Establishing a pre-defined list of high-value consumables for common orthopedic surgeries to reduce requisition form completion time" and "Opening a green channel for emergency orthopedic requisitions to shorten approval levels." These optimization measures are then applied to the current business execution plan.

[0170] Step S1548: Simulate the execution process of the business execution plan, and verify the rationality and logical coherence of the step sequence through the process simulation tool. If an input-output mismatch occurs during the simulation, return to the step adjustment stage to readjust the step sequence.

[0171] The execution process of a business execution plan is simulated using a process simulation tool. During the simulation, the inputs of each step are checked to ensure they meet the preconditions, and the outputs meet the input requirements of subsequent steps, verifying the rationality and logical coherence of the step sequence. For example, the simulation might reveal that the expected input for the "warehouse outbound" step requires "a document approved by the consumables management department," but the actual document delivered lacks the approver's signature, resulting in an input-output mismatch. In this case, the process returns to the step adjustment stage, re-examining the output requirements of the "consumables management department approval" step to ensure the output approval document contains complete signature information, and adjusting the relevant steps accordingly.

[0172] Step S155: Based on the current business status data fed back by the medical SPD business terminal, adjust the operation parameters in the optimized business execution plan, display the adjusted business execution plan in a structured manner according to the operation steps, form the final medical SPD business execution plan, and push the final business execution plan to the business terminal. Each operation step includes a step number, operation content, operation basis, data support, and expected output.

[0173] The medical SPD (Service Provider Development) terminal provides real-time feedback on current business status data, such as whether there have been changes to the surgical schedule in the orthopedic ward, whether the inventory of high-value consumables has been updated, and whether relevant approval personnel are on duty. Based on this data, the operational parameters in the optimized business execution plan are adjusted. For example, the original plan's "department review" step was expected to be completed in 30 minutes. Based on the current business status data, the department head is in the hospital and can handle it immediately, so the operational parameter is adjusted to "complete review within 20 minutes." The adjusted business execution plan is presented in a structured manner according to the operational steps. Each operational step includes a step number (e.g., S1, S2, S3, etc.), detailed operational content (e.g., "Log in to the medical SPD system, enter the high-value consumables emergency requisition module, select the orthopedic surgery preset list, and fill in the requisition quantity"), operational basis (e.g., "Emergency requisition form filling specification 3.2"), data support (e.g., "The current inventory of artificial joints is 5 sets, meeting the requisition requirement"), and expected output (e.g., "Generate an emergency requisition form with the number YJSQ-20231026-001"). The final business execution plan is pushed to the medical SPD business terminal.

[0174] For example, step S1551: Obtain current business status data through the data acquisition interface of the medical SPD business terminal. The obtained current business status data includes the real-time inventory quantity of medical supplies, inventory location, information on supplies in transit, summary of departmental requisition needs, supplier supply capacity, and current business processing progress.

[0175] Through the data acquisition interface of the medical SPD business terminal, current business status data is obtained from multiple data sources, including the hospital's inventory management system, material supply chain system, and business processing system. Real-time inventory quantities of medical supplies include the current inventory quantities of specific high-value consumables such as artificial joints and bone screws; inventory location information indicates which warehouse and shelf these consumables are stored on; in-transit information shows whether similar consumables are en route and their estimated arrival time; departmental requisition requests are summarized and statistically analyzed to determine the current requisition status of high-value consumables by various departments in the hospital, assessing any resource competition; supplier supply capacity information includes the current capacity and delivery cycle of major suppliers; and the current business processing progress displays the approval progress of requisition forms from other departments, assessing the current system workload.

[0176] Step S1552: Clean the acquired current business status data and extract the operation parameters from the optimized business execution plan. The operation parameters include the quantity of materials, operation time, execution entity, associated document number, and material delivery address. Associate the operation parameters such as the quantity of materials, operation time, execution entity, associated document number, and material delivery address in the initial draft of the business execution plan with the current business status data.

[0177] The acquired current business status data is cleaned to remove duplicate, erroneous, and invalid data, ensuring accuracy and consistency. For example, data with negative inventory quantities is removed, and typos in inventory location information are corrected. Then, operational parameters are extracted from the optimized business execution plan, such as material quantity (3 sets of artificial joints, 10 boxes of bone screws), operation time (estimated completion time for each step), executing entity (orthopedic ward nurse, department head, consumables management department approver), associated document number (surgical notification number: SS-20231026-001), and material delivery address (orthopedic operating room 1). These operational parameters are then correlated with the current business status data; for example, material quantity is correlated with real-time inventory quantity, operation time is correlated with current business processing progress, and the executing entity is correlated with personnel on-duty status.

[0178] Step S1553: Configure the operation parameter adjustment rule base, wherein each rule in the operation parameter adjustment rule base corresponds to a business state scenario and the corresponding parameter adjustment method.

[0179] The configuration includes a rule base for adjusting operation parameters. Each rule in the rule base corresponds to a specific business scenario and a corresponding parameter adjustment method. For example, one rule states: "When the real-time inventory quantity is greater than 120% of the requested quantity, the material quantity operation parameter remains unchanged; when the real-time inventory quantity is between 80% and 120% of the requested quantity, the material quantity operation parameter is adjusted to match the inventory quantity; when the real-time inventory quantity is less than 80% of the requested quantity, an emergency transfer process is initiated, and the material quantity is adjusted to the actual inventory quantity plus the transfer quantity." Another rule states: "When the executing entity is currently on duty and the workload is less than 50%, the operation time parameter is shortened by 20%; when the workload is between 50% and 80%, the operation time parameter remains unchanged; when the workload is greater than 80%, the operation time parameter is extended by 10%."

[0180] Step S1554: Match the current business status data with the rules in the operation parameter adjustment rule library, determine the matching adjustment rules and adjustment direction, and generate parameter adjustment suggestions. The parameter adjustment suggestions include the original parameter values, the adjusted parameter values, and the basis for adjustment.

[0181] The system matches current business status data with rules in the operational parameter adjustment rule base. For example, if the current real-time inventory of artificial joints is 4 sets, and the requested quantity is 3 sets, with the inventory exceeding 120% of the requested quantity (3 × 120% = 3.6), the rule "When the real-time inventory exceeds 120% of the requested quantity, the material quantity operational parameter remains unchanged" is matched. The adjustment direction is no adjustment; the original parameter value is 3 sets, and the adjusted parameter value remains 3 sets, based on sufficient real-time inventory. If the department head is currently on duty with a workload of 40% (below 50%), the rule "When the executing entity is currently on duty and the workload is below 50%, the operation time parameter is shortened by 20%" is matched. The original operation time parameter is 30 minutes, and the adjusted parameter value is 24 minutes (30 × 80%), based on the department head's relatively light workload. Based on these matching results, parameter adjustment suggestions are generated.

[0182] Step S1555: For operation parameters involving the quantity of materials, make comprehensive adjustments based on real-time inventory data, information on materials in transit, and departmental requisition requests; for operation parameters involving operation time, adjust them based on the current business processing progress and the standard processing time for each step; for operation parameters involving the executing entity, adjust them according to the current workload of each executing entity.

[0183] When comprehensively adjusting the operational parameters for material quantity, in addition to real-time inventory data, it is also necessary to consider information on materials in transit and departmental requisition requests. If the real-time inventory is insufficient, but materials in transit are expected to arrive within a few hours, and the departmental requisition request summary shows that other departments' needs for the materials are not urgent, the material quantity can be adjusted appropriately, and the materials can be issued together after the materials in transit arrive. When adjusting the operational parameters for operation time, the current business processing progress should be considered. If there are few requisitions awaiting approval in the system, the processing speed of each step may be accelerated, and the operation time parameter can be appropriately shortened; conversely, it may need to be extended. When adjusting the operational parameters for the executing entity, the personnel with lighter workloads should be selected as the executing entities based on the current workload of each entity to ensure that the operation steps can be completed on time.

[0184] Step S1556: Update the adjusted operation parameters to the business execution plan and generate parameter adjustment instructions. The parameter adjustment instructions record in detail the reasons for the adjustment of each parameter, the business status data on which it is based, and the adjustment process.

[0185] The adjusted operation parameters are updated in the business execution plan. For example, the operation time parameter for the "Department Review" step is updated from 30 minutes to 24 minutes, and the "Material Quantity" parameter is set to 3 sets. At the same time, parameter adjustment instructions are generated, recording in detail the reasons for the adjustment of each parameter, such as "Reason for adjusting the operation time of the Department Review step: The current workload of the department director is relatively light (workload 40%)"; the business status data on which it is based, such as "The workload data of the department director comes from the hospital human resources management system at 09:30 on 2023-10-26"; the adjustment process, such as "According to the rule in the rule base 'When the executing entity is currently on duty and the workload is less than 50%, the operation time parameter is shortened by 20%', 30 minutes × (1-20%) = 24 minutes".

[0186] To enable the healthcare SPD (Special Purpose Development) business enhancement model to possess knowledge completion and business logic deduction capabilities, model training is necessary. Training data includes structured data from the healthcare SPD business domain (such as business process specifications and knowledge graph triples), unstructured data (such as expert documents and case reports), and historical business execution data. During training, data preprocessing is required, including data cleaning, annotation, word segmentation, and vectorization. The model structure adopts a deep learning architecture, including modules for business understanding, knowledge completion, logic deduction, and feature enhancement. During training, appropriate hyperparameters are set, such as learning rate, batch size, and number of training epochs. The model parameters are continuously adjusted using the backpropagation algorithm to ensure the model can accurately perform knowledge completion and logic deduction, ultimately achieving the preset performance metrics (such as accuracy, recall, and F1 score).

[0187] First, a large amount of medical SPD (Special Purpose Development) business data was collected, including structured and unstructured data. The collected data underwent preprocessing, starting with data cleaning to remove noisy, duplicate, and erroneous data. Then, unstructured data was labeled, such as with business entities, relationships, and attributes. Text data was segmented and converted into word sequences. Finally, word embedding techniques were used to vectorize the word sequences, and structured data was also converted into a unified vector representation for input into the model for training.

[0188] Next, the architecture of the medical SPD business enhancement model was designed, employing a multi-layer neural network structure. The business understanding module uses a bidirectional recurrent neural network (BiLSTM) or Transformer architecture to capture textual semantic information and deep business relationships; the knowledge completion module combines knowledge graph embedding models (such as TransE and DistMult) to learn entity and relationship representations from the knowledge graph to achieve knowledge completion; the logical deduction module uses a graph neural network (GNN) to model the logical relationships between knowledge units and perform deductions; the feature enhancement module uses an attention mechanism to focus on important knowledge unit features. All modules are connected through fully connected layers or attention mechanisms to form a complete model architecture.

[0189] The preprocessed data is input into the model for training. An initial learning rate is set, such as 0.001, the batch size is 32, and the number of training epochs is 100. During training, the cross-entropy loss function is used to calculate the loss between the model's predicted values ​​and the true values, and the model parameters are updated using the Adam optimizer. After a certain number of training epochs, the model performance is evaluated on the validation set, and the learning rate (e.g., a learning rate decay strategy) and other hyperparameters are adjusted based on the validation results. When the model's performance on the validation set no longer improves, training is stopped, and the trained model parameters are saved.

[0190] The trained model was evaluated using a test set, with evaluation metrics including the accuracy of knowledge completion, the accuracy of logical deduction, and the quality of the business-related feature matrix. Based on the evaluation results, shortcomings of the model were analyzed, such as poor knowledge completion performance in certain business scenarios and errors in logical deduction within complex processes. To address these issues, model optimization was performed, such as increasing the amount of relevant scenario data in the training dataset, adjusting the weights of each module in the model, and improving the knowledge graph embedding method, until the model performance met the expected requirements.

[0191] In healthcare SPD (Special Purpose Development) operations, a large amount of privacy-sensitive data is involved, such as patient information, medical staff information, and medical supply prices. To protect the privacy of this data and prevent leakage, the following technical measures are adopted: Data anonymization: Sensitive information such as patient names, ID numbers, and contact information is anonymized, for example, by using virtual identifiers to replace real names and partially replacing characters in ID numbers; Access control: Role-based access control (RBAC) policies are used to assign different access permissions to different users, ensuring that only authorized users can access specific sensitive data; Data encryption: Sensitive data is encrypted during storage and transmission, using the AES encryption algorithm to encrypt sensitive fields in the database and the SSL / TLS protocol to encrypt the data transmission process; Audit logs: All access and operation behaviors to sensitive data are recorded, including the accessing user, access time, and operation content, for subsequent auditing and tracking; Differential privacy: When data is published or shared, a suitable amount of noise is added to prevent attackers from inferring sensitive individual information from the published data, thus protecting data privacy.

[0192] Based on the same inventive concept, please refer to Figure 2 This paper shows a schematic block diagram of the structure of an information enhancement retrieval system 100 based on a large model and vector knowledge base, which is provided in an embodiment of this application for performing the above-described information enhancement retrieval method based on a large model and vector knowledge base. The information enhancement retrieval system 100 based on a large model and vector knowledge base may include a communication unit 110, a machine-readable storage medium 120, and a processor 130.

[0193] In this embodiment, both the machine-readable storage medium 120 and the processor 130 are located in the information enhancement retrieval system 100 based on a large model and vector knowledge base, and are separately configured. Alternatively, the machine-readable storage medium 120 can also be integrated into the processor 130 and can communicate and interact with external systems through the communication unit 110. The machine-readable storage medium 120 is used to store machine-executable instructions for executing the scheme of this application, and the processor 130 is used to execute the machine-executable instructions stored in the machine-readable storage medium 120 to implement the information enhancement retrieval method based on a large model and vector knowledge base provided in the aforementioned method embodiments.

[0194] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A large model and vector knowledge base based information enhancement retrieval method, characterized in that, The method includes: The system receives a search request initiated by a user through a medical SPD business terminal, triggering a medical SPD business scenario perception model. The search request includes business operation requirements and associated medical SPD business context. The medical SPD business scenario perception model has built-in mapping rules between business processes and search intent. The medical SPD business scenario perception model outputs scenario adaptation parameters that match the retrieval request, and then calls the medical SPD vector knowledge base. The scenario adaptation parameters include business process identifiers, core data dimensions, and related business rules. The medical SPD vector knowledge base stores structured and unstructured knowledge units of the entire medical SPD business process. Using the scenario adaptation parameters as the core elements, a dynamic association network between retrieval intent and knowledge units is constructed in the medical SPD vector knowledge base. Each node in the dynamic association network corresponds to a knowledge unit or retrieval intent element, and the connections between nodes represent business logic associations. The dynamic association network is input into the medical SPD business enhancement model to generate a knowledge enhancement set. The medical SPD business enhancement model is jointly trained with medical SPD business data and scenario cases and has the ability to complete knowledge and deduce business logic. Based on the knowledge enhancement set, a medical SPD business execution plan is generated, and the business execution plan is pushed to the medical SPD business terminal. The business execution plan includes operation steps, related knowledge basis, and data support sources. The process of generating a medical SPD service execution plan based on the knowledge enhancement set includes: The knowledge enhancement set is parsed, and the knowledge units, business logic association descriptions, and business application scenario descriptions are extracted to generate a business execution element list. The business execution element list includes the operation object, operation action, operation basis, and operation sequence requirements. The medical SPD business process template library is invoked. This library contains standard process templates for all business processes in medical SPD. Each standard process template corresponds to a specific business scenario. The standard process template clearly defines the order and relationship of the operation steps. Based on the business application scenario descriptions in the knowledge enhancement set, match the corresponding standard process template, fill the elements in the business execution element list into the standard process template, and generate a draft of the business execution plan. The initial draft of the business execution plan was optimized to obtain the optimized business execution plan; Based on the current business status data fed back by the medical SPD business terminal, the operation parameters in the optimized business execution plan are adjusted, and the adjusted business execution plan is displayed in a structured manner according to the operation steps to form the final medical SPD business execution plan. The final business execution plan is then pushed to the business terminal. Each operation step includes a step number, operation content, operation basis, data support, and expected output. 2.The large model and vector knowledge base based information enhancement retrieval method according to claim 1, characterized in that, The process of receiving a search request initiated by a user through a medical SPD service terminal and triggering the medical SPD service scenario awareness model includes: The system receives user input information through the interactive interface of the medical SPD business terminal. The interactive interface supports text input, voice input, and business form import. It converts different forms of user input information into raw search text in a unified format. During the conversion process, the tone and pause characteristics of voice input and the field association relationship of imported forms are preserved. The original search text is cleaned for business semantics to obtain core text segments containing business requirements. The division of core text segments is based on sentence-end punctuation and semantic pauses. Extract medical SPD business feature words from the core text fragment. The business feature words cover medical supply names, business operation verbs, business process names, data requirement types, and responsible entity identifiers. Each business feature word corresponds to a unique business attribute code. The context association module of the medical SPD business scenario perception model is invoked to read the business permission information, historical operation records and currently associated medical SPD business document data of the currently logged-in user of the medical SPD business terminal, and generate a set of business context features. The business feature words are fused with the business context feature set to generate a search intent feature vector. During the fusion process, feature words associated with the current business document data are given higher weights. The weight allocation is determined based on the urgency and relevance of the business document. The scenario rule engine of the medical SPD business scenario perception model is invoked to store the matching rules between the search intent feature vector and the business scenario constructed by medical SPD business experts. Each matching rule includes feature vector matching conditions, scenario identifier and scenario description information. The search intent feature vector is input into the scene rule engine, and a scene matching result is generated through rule matching. The scene matching result includes the scene identifier of the successfully matched scene, the matching degree score, and the associated business process node. Based on the matching score, a main scenario identifier and an associated scenario identifier are selected. The main scenario identifier corresponds to the core business requirement of the retrieval request, and the associated scenario identifier corresponds to the extended business requirement related to the core requirement. The main scenario identifier and the associated scenario identifier are sent together to the parameter generation module of the medical SPD business scenario perception model to trigger the parameter generation module to start running. 3.The large model and vector knowledge base based information enhancement retrieval method according to claim 1, characterized in that, The construction of a dynamic association network between retrieval intent and knowledge units in the medical SPD vector knowledge base, using the scenario adaptation parameters as core elements, includes: The scenario adaptation parameters are analyzed, and the business process identifiers, core data dimensions and related business rules are extracted as the core elements of network construction. Business attribute descriptions and weight proportions in search intents are added to each core element. The knowledge organization engine of the medical SPD vector knowledge base is invoked. This knowledge organization engine divides the medical SPD business process into multiple knowledge domains. Each knowledge domain contains several knowledge units, and each knowledge unit corresponds to a specific business knowledge point. Based on the business process identifier, locate the corresponding target knowledge domain, and read the metadata information of all knowledge units in the target knowledge domain. The metadata information includes knowledge unit identifier, business attribute tag, associated business rules and data dimension description. The core elements are matched with the metadata information of the knowledge units for business association. During the matching process, the initial screening is performed by business attribute tags, and then the precise matching is performed by core data dimensions to generate a knowledge unit matching list. Based on the overlap of business attributes, data dimension matching degree, and fit of related business rules between core elements and knowledge units, the association strength value of each knowledge unit in the knowledge unit matching list is calculated. The higher the association strength value, the stronger the association between the knowledge unit and the search intent. The initial structure of the dynamic association network is constructed, which includes three layers: the first layer is the retrieval intent layer, which stores the core text fragments of the retrieval request and the retrieval intent feature vector; the second layer is the core element layer, which stores the core elements and weight ratios of the scenario adaptation parameters; and the third layer is the knowledge unit layer, which stores the knowledge units in the knowledge unit matching list that meet the association strength value and the association strength value. Establish association links between the three levels: the retrieval intent layer and the core element layer are associated through business feature words, and the core element layer and the knowledge unit layer are associated through business association matching relationships. Each association link is marked with the corresponding association basis and strength value. The associated links in the initial structure are dynamically optimized by merging duplicate links pointing to the same knowledge unit, strengthening the connection stability of links with an association strength value greater than a set association strength value, and removing invalid links with an association strength value lower than a set standard. Add network dynamic update trigger conditions. When the business context of the retrieval request changes or the knowledge unit in the medical SPD vector knowledge base is updated, the reconstruction of the associated network is automatically triggered, and the node information and associated links of the dynamic associated network are updated synchronously, so that the content of the dynamic associated network corresponds to the current business needs. 4.The method of claim 1, wherein, The step of inputting the dynamic association network into the medical SPD business enhancement model to generate a knowledge enhancement set includes: Read the hierarchical structure, node information and association link data of the dynamic association network, convert the hierarchical structure, node information and association link data of the dynamic association network into tensor data format that can be recognized by the medical SPD business enhancement big model, and obtain the converted tensor data. During the conversion process, the association relationship and strength information between nodes are preserved. The transformed tensor data is input into the business understanding module of the medical SPD business enhancement model. This business understanding module contains a medical SPD business semantic encoding unit. Through semantic encoding, it captures the deep business relationships between search intent, core elements and knowledge units, and outputs a business relationship feature matrix. This matrix is ​​then passed to the knowledge completion module of the medical SPD business enhancement model. The knowledge completion module calls the built-in medical SPD business knowledge graph, which stores medical SPD business entities, relationships and attribute information in the form of triples. The business association feature matrix is ​​compared with the medical SPD business knowledge graph to identify missing information and logical gaps in the knowledge units of the dynamic association network. Based on the association relationships in the medical SPD business knowledge graph, the missing business knowledge is supplemented and the logical gaps are repaired. The completed knowledge unit data is input into the model's logical deduction module, which calls the medical SPD business process specification library. This library stores business process standards that cover the entire process from material procurement to cost settlement. Each specification item includes process steps, operational requirements, and logical relationship descriptions. Extract the business process identifiers and operation descriptions from the completed knowledge units, match them with the specification entries in the business process specification library, and determine the process position and logical rules to be followed for each knowledge unit. Construct a knowledge unit logical relationship matrix, wherein the rows and columns of the knowledge unit logical relationship matrix correspond to knowledge units, and the matrix element values ​​represent the logical relationship type between the knowledge units corresponding to the rows and the knowledge units corresponding to the columns. The logical relationship type includes pre-relationship, post-relationship, parallel relationship, or no direct relationship. The rationality of the knowledge unit logical relationship matrix is ​​verified according to the business process standard. When the logical relationship recorded in the knowledge unit logical relationship matrix does not match the process order specified in the specification, it is marked as a logical anomaly. The knowledge unit relationships marked as logical anomalies are corrected and the logical relationship types in the knowledge unit logical relationship matrix are adjusted. The knowledge units are sorted according to the chronological order of the medical SPD business process to generate a knowledge association sequence. Each knowledge unit in the knowledge association sequence is labeled with its preceding and succeeding knowledge units. The result is then output to the feature enhancement module of the medical SPD business enhancement model. This feature enhancement module uses an attention mechanism to focus on the features of knowledge units that are associated with the retrieval intent, obtains the focused knowledge unit features, and performs fusion processing on the focused knowledge unit features. The fused feature data is then organized according to the order of the business process to generate a knowledge enhancement feature vector. Each knowledge enhancement feature vector corresponds to a business knowledge module. The knowledge-enhanced feature vector is converted into a structured knowledge-enhanced set, which includes the completed knowledge unit, business logic association description, knowledge source identifier and association strength description. Each knowledge unit in the knowledge-enhanced set is accompanied by a corresponding business application scenario description. 5.The large model and vector knowledge base based information enhancement retrieval method according to claim 2, characterized in that, The extraction of medical SPD business feature words from the core text fragment includes: Call the preset medical SPD business feature word dictionary, which contains commonly used business feature words in the medical SPD field and their corresponding business categories, expression variations and semantic weights. The business categories include material information, operation behavior, business process and data requirement. The core text segment is segmented into words, and the resulting word units are matched with the medical SPD business feature word dictionary. Successfully matched word units are selected as candidate business feature words, and the business category and semantic weight corresponding to each candidate business feature word are recorded. The candidate business feature words are deduplicated, and the deduplicated candidate business feature words are used as network nodes. Connections are established between nodes according to the semantic relationships in the core text fragments. The thickness of the connection represents the tightness of the semantic relationship, and a business feature word association network is constructed. The importance of nodes in the business feature word association network is analyzed, and the core feature words are determined by calculating the degree centrality of the nodes. The higher the degree centrality, the wider the association range of the feature word in the business feature word association network, and the more critical its role in expressing the search intent. The core feature words that meet the degree centrality standard are grouped according to business categories to form material information feature group, operational behavior feature group, business process feature group and data demand feature group. The feature words in each group are sorted according to semantic weight. A business attribute code is added to the feature words in each group. The business attribute code includes a business classification identifier, semantic weight level and associated business scenario code. The grouped feature words and the corresponding business attribute codes together constitute the medical SPD business feature word set. 6.The information retrieval method based on large model and vector knowledge base according to claim 3, characterized in that, The method of calculating the association strength value of each knowledge unit in the knowledge unit matching list based on the overlap of business attributes, data dimension matching degree, and fit of related business rules between core elements and knowledge units includes: Set up a correlation strength value calculation system, which includes three calculation dimensions: business attribute overlap dimension, data dimension matching dimension, and business rule fit dimension. Each dimension is configured with specific calculation indicators. In terms of overlapping business attributes, the number and rate of overlap between the business attribute tags of the core elements of the scenario adaptation parameters and the business attribute tags of the knowledge unit metadata are calculated, and the degree of overlap of business attributes is calculated based on the number and rate of overlap. In the data dimension matching dimension, the core data dimensions in the core elements are extracted and the metadata dimension descriptions of the knowledge units are compared. The matching of the data types, data formats and data value ranges of the two is then used to generate the data dimension matching degree. In terms of business rule fit, the associated business rules in the core elements are compared with the business rules associated with the knowledge units. The logical consistency of the business rules and the overlap of applicable scenarios are analyzed to generate the fit of the associated business rules. The overlap of business attributes, the matching degree of data dimensions, and the fit of related business rules are weighted and calculated to obtain the association strength value of each knowledge unit. Collect scoring data of historical knowledge unit matching cases from medical SPD business experts, establish a correlation strength value correction model based on the scoring data, input the initially calculated correlation strength value into the correlation strength value correction model, and obtain the corrected final correlation strength value. The final association strength value is compared with the preset association strength threshold, and knowledge units with a final association strength value higher than the association strength threshold are selected and retained in the knowledge unit matching list. 7.The large model and vector knowledge base based information enhancement retrieval method according to claim 4, characterized in that, The step of comparing the business association feature matrix with the medical SPD business knowledge graph to identify missing information and logical gaps in knowledge units in the dynamic association network includes: Read the business association feature matrix, extract the knowledge unit feature vector and the association relationship features between knowledge units contained therein, and convert the knowledge unit feature vector and the association relationship features between knowledge units into a knowledge graph query statement. The query statement contains the entity and association relationship type corresponding to the knowledge unit. Call the query interface of the medical SPD business knowledge graph, execute the query statement, and obtain the entity, relationship and attribute information associated with the knowledge unit in the knowledge graph to form a knowledge graph association information set; The knowledge graph association information set is compared with the knowledge unit information in the dynamic association network. The attribute information that exists in the medical SPD business knowledge graph but is missing in the knowledge unit is the missing information of the knowledge unit. Analyze the business logic relationships between knowledge units in the dynamic association network, configure the logical relationship judgment criteria based on the medical SPD business process specifications, and the logical relationship judgment criteria include the preset logical association types between knowledge units in different business links; Based on the aforementioned logical relationship judgment criteria, check whether the connection links between knowledge units in the dynamic connection network are complete. If the knowledge units of adjacent business links lack the necessary logical connection, or the connection relationship does not conform to the medical SPD business process specification, then it is determined that there is a logical break in the dynamic connection network. The identified missing information is classified and organized according to attribute type into basic attribute missing, related attribute missing and extended attribute missing, and different types of missing information correspond to different completion priorities. Locate the business process where the logical gap is located, record the knowledge units involved and the types of missing logical connections, and generate an identification report of missing information and logical gap. The identification report includes knowledge unit identifiers, missing information types, logical gap locations, and repair suggestions. 8.The large model and vector knowledge base based information enhancement retrieval method according to claim 1, characterized in that, The process optimization of the initial draft of the business execution plan to obtain the optimized business execution plan includes: Extract all operation steps from the initial draft of the business execution plan, and organize the operation object, operation action, expected input and expected output information contained in each operation step into a step information table. Each operation step contains operation object, operation action, expected input and expected output information. Configure an operation step sequence relationship model. This operation step sequence relationship model is based on the sequential order rules of the medical SPD business process. It has built-in preconditions and postconditions between different operation actions. The preconditions are the input content required to perform the operation action, and the postconditions are the output content generated after the operation action is completed. Input the operation steps from the step information table into the operation step time sequence relationship model, analyze the preconditions and postconditions of each step, determine whether the logical relationship between the steps is reasonable, and generate the time sequence relationship analysis results. Based on the results of the temporal relationship analysis, the order of operation steps is adjusted, and steps with a pre- or post-precedence relationship are arranged in logical order so that the expected output of the previous step corresponds to the expected input of the next step. Identify parallel operation steps. For operation steps that do not have a pre- or post-order relationship but belong to the same business process, set them to parallel execution mode and mark the parallel relationship and the start time node of execution in the business execution plan. The supplementary steps include a connection description that clarifies the specific method by which the output of the previous step is passed to the next step. This description includes the data transmission method, the transmission interface number, and the data verification method. The system calls upon the medical SPD business process optimization case library, which stores historical optimized business process solutions. The adjusted business execution solution is compared with similar solutions in the case library, and the optimization measures from the similar solutions are extracted and applied to the current business execution solution. The simulation process of the business execution plan is used to verify the rationality and logical coherence of the step sequence through process simulation tools. If an input-output mismatch occurs during the simulation, the process is returned to the step adjustment stage to readjust the step sequence.

9. A large model and vector knowledge base based information augmentation retrieval system, characterized by, include: processor; A machine-readable storage medium for storing machine-executable instructions of the processor; The processor is configured to execute the information enhancement retrieval method based on a large model and vector knowledge base as described in any one of claims 1 to 8 by executing the machine-executable instructions.

Images