Intelligent data management and business collaboration method for medical beauty industry ERP system

Abstract

This application discloses a method for intelligent data management and business collaboration in an ERP system for the medical aesthetics industry. The method includes: S1: Constructing an intelligent integration layer for multi-source heterogeneous data in the medical aesthetics field; S2: Designing a consumables intelligent association and early warning management module based on "service paths"; S3: Realizing dynamic business collaboration based on knowledge graphs and role permissions, automatically generating and synchronizing tasks according to role configuration context information, and supporting collaboration among multiple roles under a unified view; S4: Providing intelligent decision support based on graph analysis and machine learning. This application, by constructing an ontology model for the medical aesthetics field and utilizing multi-path knowledge extraction technology, transforms scattered and heterogeneous operational data into a unified knowledge graph with semantic associations. This fundamentally breaks down data silos, providing a high-quality, associative, and reasonable data foundation for the entire business chain, and achieving deep integration and unified governance of multi-source heterogeneous data in the medical aesthetics field.

Description

Technical Field

[0001] This application belongs to the field of data management technology, specifically involving intelligent data management and business collaboration methods for ERP systems in the medical aesthetics industry. Background Technology

[0002] With the rapid development of the medical aesthetics industry, the operation of institutions has become increasingly complex, and traditional management models and general-purpose ERP systems have gradually revealed many shortcomings in dealing with the industry's unique challenges.

[0003] At the data management level, medical aesthetic institutions have long business chains and highly dispersed data sources, encompassing multiple systems such as Customer Relationship Management (CRM), Electronic Medical Records (EMR), inventory, finance, and social media. Furthermore, the data formats are heterogeneous and semantically isolated, forming "data silos" that are difficult to integrate. This prevents institutions from building a unified customer overview and business knowledge system, limiting the depth of data value mining. Regarding consumable management, especially high-value and implantable consumables, there is a general problem of inefficient management and disconnect from clinical service processes. Consumable usage often relies on personal experience and lacks standardized linkage with specific medical aesthetic projects, leading to difficulties in traceability, unreasonable inventory, high compliance risks, and potential waste or safety hazards. Furthermore, in terms of business collaboration, information transmission among multiple roles such as consultants, doctors, nurses, warehouse managers, and finance relies on traditional methods, resulting in fragmented processes, opaque task status, and low collaboration efficiency. Moreover, a large amount of tacit knowledge existing in the diagnosis and treatment process cannot be effectively accumulated and reused. At the decision support level, management finds it difficult to conduct accurate customer value analysis, consumable demand forecasting, and service portfolio optimization based on fragmented and static data. Decisions rely heavily on experience and lack forward-looking, data-driven intelligent support.

[0004] Therefore, there is an urgent need for an ERP system management method that can deeply integrate with the business characteristics of the medical aesthetics industry to achieve intelligent data integration, refined process control, efficient and smooth collaboration, and scientific and accurate decision-making. Summary of the Invention

[0005] This application provides a method for intelligent data management and business collaboration in ERP systems for the medical aesthetics industry, aiming to solve the problems of highly dispersed data sources, difficulty in traceability, unreasonable inventory, high compliance risks, and potential waste or security risks in existing technologies.

[0006] A method for intelligent data management and business collaboration in an ERP system for the medical aesthetics industry, the method comprising:

[0007] S1: Construct an intelligent integration layer for multi-source heterogeneous data in the medical aesthetics field, and use knowledge graph technology to perform semantic modeling, standardized extraction and unified storage of operational data in the medical aesthetics industry;

[0008] S2: Design a consumables intelligent association and early warning management module based on "service path" to perform association management, real-time early warning and closed-loop traceability of consumables according to the standardized service path of medical aesthetics projects;

[0009] S3: Enables dynamic business collaboration based on knowledge graphs and role permissions, automatically generates and synchronizes tasks according to role configuration context information, and supports collaboration among multiple roles under a unified view;

[0010] S4: Provides intelligent decision support based on graph analysis and machine learning, offering decision-making basis through customer profiling, demand forecasting, and operational analysis.

[0011] Optionally, S1 specifically includes:

[0012] S1.1: Construct an ontology model for the medical aesthetics field, defining entity classes, attributes, and their semantic relationships in the medical aesthetics field;

[0013] S1.2: Through multi-path knowledge extraction and transformation, multi-source heterogeneous data is transformed into standard triples or attribute graph data that conform to the ontology model;

[0014] S1.3: A hybrid storage architecture with multiple graph storage engines is used to store knowledge data, and a consistent query interface is provided through a unified data access layer.

[0015] Optionally, in S1.2, multi-path knowledge extraction includes:

[0016] Structured data is extracted using pattern mapping.

[0017] Extraction of semi-structured data is achieved by combining a rule engine with natural language processing.

[0018] Information extraction based on deep learning models is used for unstructured data.

[0019] Optionally, S2 specifically includes:

[0020] S2.1: Build and maintain a standard "service path" library for medical aesthetic projects, with each path associated with a "standard consumables set" for each service stage;

[0021] S2.2: Instantiate the service path when a service order is created and automatically associate it with the expected consumables list;

[0022] S2.3: Real-time monitoring and multi-layered intelligent early warning of consumable usage through a rules engine;

[0023] S2.4: Achieve closed-loop traceability of consumable consumption and feed back actual consumption data to the service path database.

[0024] Optionally, S2.3 also includes a self-learning and dynamic optimization mechanism for early warning rules, including: constructing an early warning event-result analysis sample library;

[0025] Risk prediction model based on supervised learning training;

[0026] The parameters and logic of the early warning rules are dynamically adjusted based on the model evaluation results.

[0027] Optionally, the baseline usage of the "standard consumable set" in S2.1 is generated through a personalized dynamic optimization algorithm, including: constructing a feature vector of usage influencing factors;

[0028] The usage prediction model trained by the application calculates the recommended usage based on the algorithm.

[0029] The final dynamic baseline usage is generated by verifying the safety and rationality rule base.

[0030] Optionally, S3 specifically includes:

[0031] S3.1: Configure permissions and context templates based on an open role model;

[0032] S3.2: Query the knowledge graph in real time based on the context template, aggregate and push relevant information;

[0033] S3.3: Automatically generate tasks based on service path status and synchronize them to the collaboration dashboard;

[0034] S3.4: Supports semantic annotation during business processes to enable incremental construction of knowledge graphs.

[0035] Optionally, S4 specifically includes: constructing a 360° customer profile based on graph embedding and community discovery and making personalized project recommendations;

[0036] By integrating time series analysis and graph correlation, accurate prediction of consumable demand can be achieved;

[0037] Multidimensional analysis of project association networks, physician operational efficiency, and high-margin service combinations based on graph algorithms and clustering.

[0038] Optionally, the personalized project recommendation includes: extracting a customer subgraph and performing graph embedding to obtain a customer vector;

[0039] Discover customer communities and analyze their common characteristics through clustering;

[0040] A list of recommended items and reasons for recommendation are generated based on customer vectors, community features, and graph reasoning.

[0041] Optionally, the accurate prediction of consumable demand includes: constructing a prediction feature set that integrates time series and map correlation features;

[0042] Demand forecasting is performed using an integrated forecasting model.

[0043] Procurement recommendations are generated based on forecast results, and inventory strategies are dynamically adjusted.

[0044] Compared with the prior art, this application has at least the following beneficial effects:

[0045] This application constructs an ontology model in the medical aesthetics field and utilizes multi-path knowledge extraction technology to transform scattered and heterogeneous operational data into a unified knowledge graph with semantic associations. This fundamentally breaks down data silos and provides a high-quality, associative, and reasonable data foundation for the entire business chain, achieving deep integration and unified governance of multi-source heterogeneous data in the medical aesthetics field.

[0046] This application designs a consumable association mechanism based on "service path" to deeply bind consumable use with standardized clinical service processes. Combined with multi-layer intelligent early warning (including self-learning optimization) and closed-loop traceability, it achieves precise control over the entire process from planning, preparation, use to feedback, reduces consumable waste and compliance risks, and improves the level of refinement, intelligence and compliance of consumable management. Attached Figure Description

[0047] Figure 1 This is a flowchart illustrating an embodiment of the intelligent data management and business collaboration method for an ERP system in the medical aesthetics industry provided in this application. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments.

[0049] The intelligent data management and business collaboration method for the medical aesthetics industry ERP system provided in this application includes the following steps:

[0050] S1: Construct an intelligent integration layer for multi-source heterogeneous data in the medical aesthetics field;

[0051] Specifically, in order to address the problem of fragmented data sources, inconsistent formats, and severe semantic isolation in the medical aesthetics industry, which makes it impossible to form a unified customer view and business knowledge system, this application constructs a multi-source heterogeneous data intelligent integration layer in the medical aesthetics field to realize the data foundation for intelligent management and collaboration. Its core lies in using knowledge graph technology to perform semantic modeling, standardized extraction, and unified storage of massive and heterogeneous operational data.

[0052] S1 specifically includes the following steps:

[0053] S1.1: Construction of an Ontology Model for the Medical Aesthetics Field. Based on the unique business logic and professional knowledge of the medical aesthetics industry, a structured domain ontology model is constructed. This model formally defines the core concepts (i.e., entity classes), attributes, and semantic relationships between them in the medical aesthetics field. Core entity classes include at least: clients, medical aesthetic projects (such as mandibular angle reduction surgery, Thermage, and mesotherapy), project plans (including specific procedures and parameters), pharmaceuticals and consumables (subdivided into fillers, implants, phototherapy devices, dressings, etc., and labeled with high-value and implantation status), service personnel (doctors, consultants, nurses, associated with their qualifications and expertise), service stages (consultation, in-person examination, design, treatment, nursing, and follow-up), physiological indicators, and aesthetic parameters. Relationships between entities are defined through rich business semantic associations such as client-acceptance-project, project-includes-consumables, consumables-used-in-service stages, doctor-specializes-project, and project-affects-aesthetic parameters. This ontology model serves as a globally unified data schema, providing semantic specifications for the subsequent integration, association, and reasoning of all data.

[0054] S1.2: Multi-path knowledge extraction and transformation. For the multi-source heterogeneous data that actually exists in medical aesthetic institutions, three parallel knowledge extraction paths are designed and implemented to transform it into standard triples (subject-relation-object) or attribute graph data that conform to the ontology model defined in S1.1, as follows;

[0055] For structured data extraction from CRM systems, financial software, and warehouse management systems (WMS) using structured databases (such as MySQL and SQL Server), a schema mapping-based technique is employed. By configuring mapping tools like D2RQ, tables and fields from relational databases are directly mapped to entity classes and attributes in the ontology model, enabling automatic and batch conversion of data from customer information tables, inventory tables, and order tables into the model.

[0056] For semi-structured data extraction, documents with fixed formats but flexible content, such as electronic medical records (EMR), surgical records, and informed consent forms, employ a combination of rule engines and natural language processing (NLP). First, predefined templates and rules (such as regular expressions and keyword matching) are used to locate key field areas in the document (e.g., patient complaints, diagnostic conclusions, and lists of consumables used). Then, named entity recognition (NER) technology based on dictionary or conditional random fields (CRF) is used to extract specific entities (e.g., drug name "Sodium Hyaluronate for Injection," project name "Rhinoplasty") and their attributes (e.g., specifications, dosage) from these areas.

[0057] For unstructured data extraction, a deep learning-based end-to-end information extraction model is used for purely unstructured text such as customer service dialogue records, social media reviews, doctors' handwritten notes, and postoperative follow-up recordings (which need to be converted to text). Specifically, pre-trained language models (such as BERT and RoBERTa) can be fine-tuned to achieve joint extraction of entities and relations.

[0058] S1.3: Multi-engine shared knowledge base storage and management. To address the differentiated needs of data querying, analysis, and reasoning in different business scenarios, this invention adopts a hybrid storage architecture that supports multiple graph storage engines. The standardized knowledge data obtained after processing in S1.2 is not stored in a single database, but is dynamically or configurably stored in the most suitable storage engine according to data characteristics and access patterns.

[0059] Attribute graph databases (such as Neo4j) are used to store business network data that requires frequent complex path queries and real-time relationship exploration, such as customer social recommendation networks and inter-project performance collaboration relationships. Their efficient graph traversal capabilities are well-suited for dynamic business collaboration scenarios.

[0060] RDF triplet libraries (such as GraphDB, gStore): are used to store core knowledge data that needs to strictly follow ontology semantics, perform logical reasoning and complex SPARQL queries, such as complete customer-project-consumable relationships. Their powerful semantic reasoning capabilities can be used to discover implicit knowledge.

[0061] Relational databases or time-series databases are used to store data that requires high transaction consistency or intensive access in a time series, such as specific transaction logs, consumable inventory change logs, and customer vital signs time-series data.

[0062] The system shields upper-layer applications from the differences between underlying multi-engines through a unified data access and virtualization layer, providing a consistent query interface. This layer maintains a unified global data identifier (URI) and metadata index, ensuring that data can be uniquely identified and queried across engines through semantic associations in the knowledge graph, regardless of where the data is physically stored.

[0063] S2: Design a consumables intelligent association and early warning management module based on "service path";

[0064] Specifically, to address issues such as lax management, disconnect from clinical services, difficulty in traceability, and easy compliance risks in the medical aesthetics industry, especially in the use of high-value and implantable consumables, a consumables intelligent association and early warning management module based on "service path" was designed and implemented.

[0065] S2 specifically includes the following steps:

[0066] S2.1: Construction and Maintenance of the Medical Aesthetics Project "Service Path" Library. First, define a standard "service path" for each standardized medical aesthetics project performed within the institution (such as "breast augmentation," "rhinoplasty," and "picosecond laser treatment"). Each service path is a structured process template, clearly decomposing the sequence of key service stages that the project must undergo from start to finish (e.g., "client assessment and treatment plan design," "preoperative preparation and informed consent," "anesthesia and surgical procedure," "postoperative care," and "recovery care and follow-up"). More importantly, at each service stage node, associate a predefined "standard consumables set." This consumables set lists the recommended or mandatory drugs, materials, devices, their specifications, brands (if specified), and baseline dosage or units of use for that stage. This "service path" library is maintained as a configurable and scalable knowledge base, developed and regularly reviewed and updated by the institution's Medical Quality Management Committee in conjunction with product instructions, clinical guidelines, expert consensus, and the institution's own regulations.

[0067] S2.2: Intelligent association between service instantiation and consumables. When a specific medical aesthetic service order is created for a particular customer, the system automatically instantiates the corresponding standard "service path" based on the selected items, generating a "service execution path" belonging to that customer-order. During this process, the system automatically pre-associates the "standard consumable set" associated with each stage of the path to this order, forming a preliminary "estimated consumable list." Physicians or therapists can make fine-tuning adjustments to this list based on the customer's individual circumstances (such as allergy history, anatomical characteristics), such as adding alternative consumables or adjusting the estimated usage. The system records this adjustment as a deviation reason. This "estimated consumable list" then becomes the benchmark for consumable management for this service and is linked in real-time with the inventory management system to pre-allocate or check the availability of consumables within the list.

[0068] S2.3: Multi-level intelligent early warning and compliance intervention. The system has a built-in configurable rule engine that monitors and issues early warnings for consumable usage in real time throughout the entire service execution process.

[0069] Inventory and expiration date warnings: During the service preparation phase, if the inventory of any item in the "Expected Consumables List" is lower than the safety threshold or is close to its expiration date, the system will automatically send a warning to the warehouse management personnel and service preparation personnel.

[0070] The system provides compliance alerts for service delivery procedures. When physicians prescribe or nurses prepare consumables for actual use, if the type or brand significantly deviates from the "standard consumables set" set in the "service delivery procedure" (e.g., using a high-value implant from an unapproved alternative brand), or if the dosage exceeds the baseline range by a certain percentage without reasonable notes, the system will trigger different levels of alerts based on the severity of the rule. For example, a yellow alert prompts the operator to confirm, while a red alert requires online approval from a senior physician or department manager before execution can continue.

[0071] Mandatory verification at key points: Before the start of stages such as "surgery implementation", the system can require nurses to scan the barcode / QR code of consumables to verify the "expected consumables list". Only after the system confirms that the comparison is correct can the stage be marked as ready, ensuring that "the items match the list".

[0072] Meanwhile, in order to enable the early warning system to adapt to the complexity of clinical practice, reduce the over-reliance on fixed rules, and improve the accuracy and practicality of the early warning system, this invention introduces a machine learning-based self-learning and dynamic optimization mechanism for early warning rules on the basis of a configurable rule engine, so as to provide a self-learning and dynamic optimization mechanism for early warning rules.

[0073] Specifically, the implementation of the self-learning and dynamic optimization mechanism for early warning rules includes the following steps:

[0074] First, a sample library for early warning events and their results is constructed. The system continuously records all early warning events triggered by deviations from the standard consumable set in the "service path," forming a structured "deviation event" record. Each record contains rich feature information: the type of deviation (e.g., brand substitution, specification change, dosage increase / decrease), specific items and consumable categories, relevant customer characteristics (age, skin type, etc.), the executing physician, the magnitude of the deviation (e.g., the percentage by which the dosage exceeds the baseline), and the reason for the deviation provided by the operator. Crucially, the system associates subsequent processing flows and results as tags with each event: including the approval result of the early warning (automatic approval, manual approval, rejection), the approver's level, and subsequent customer follow-up feedback and effectiveness evaluation (e.g., sentiment and keywords extracted from the evaluation text using natural language processing technology, or standardized follow-up questionnaire scores).

[0075] A risk prediction model is trained using supervised learning. The aforementioned "deviation event" records are used as training samples, and "whether high-level intervention is needed" is used as the prediction target (e.g., defining events that are "rejected" or associated with "negative feedback" as high risk, and events that are "automatically approved" and associated with "positive or neutral feedback" as low risk). A classification model (such as logistic regression, support vector machine, or ensemble learning model) is trained. This model learns the mapping relationship from combinations of event features to true risk outcomes. Through attribution analysis of model features (e.g., using SHAP values), the historical risk probability of different deviation scenarios (e.g., "Physician A used brand C as a substitute in project B") can be quantitatively assessed.

[0076] The system dynamically optimizes the parameters and logic of the early warning rules. Periodically (e.g., monthly) or after accumulating sufficient new samples, it runs the aforementioned risk prediction model to evaluate historical and simulated events. Based on the evaluation results, and automatically or with administrator confirmation, the static rule base is optimized as follows: 1) Risk reclassification: For deviation patterns predicted as high-risk by the model but with a low current warning level, the system suggests increasing the warning level (e.g., upgrading from alert to requiring approval); for common deviation patterns consistently predicted as low-risk by the model, the system suggests lowering the warning level or adding them to a "whitelist" to avoid warnings when specific conditions are met. 2) Adaptive threshold adjustment: For usage deviation warnings, the system can dynamically calculate and suggest a more reasonable threshold range based on historical data distribution and risk correlation, rather than a fixed percentage. 3) Rule derivation: When certain feature combinations are found to be strongly correlated with high risk (e.g., "brand substitution occurs when a certain consumable is used in a specific sensitive area"), a new, more refined specific early warning rule can be suggested.

[0077] S2.4: Closed-loop traceability and data feedback for consumable consumption. During service execution, all consumables actually used must be registered in real time by scanning their unique identification code (UDI or internal code). The system automatically records the "actual consumption details" of the consumables and accurately binds them to the "estimated consumable list," the specific "service stage," and the "executive personnel." After the service is completed, the system automatically generates a "closed-loop consumable consumption report," clearly showing the difference between the plan and actual usage. For high-value implantable consumables, full traceability is achieved with "one item, one code, one customer." This "actual consumption data," as valuable business data, is fed back to the "service path" database in S201 for regular analysis and optimization of the "baseline usage," enabling continuous iteration of the standard path.

[0078] Furthermore, in order to achieve precise management of consumable usage and avoid waste or shortage caused by fixed usage standards, this invention introduces a set of personalized dynamic optimization algorithms in the definition of "baseline usage" of "standard consumable set" in the service path. The core of this algorithm is that the baseline usage is not a static value or range, but a dynamic suggested value calculated in real time by the algorithm model based on multi-dimensional input parameters.

[0079] Specifically, the implementation of the personalized dynamic optimization algorithm for the baseline usage of consumables includes the following steps:

[0080] First, a feature vector of usage influencing factors is constructed. The system extracts multi-source features related to the current service order from the knowledge graph and constructs an input feature vector. These features mainly include two categories: 1) Customer individual characteristics, including objective physiological parameters obtained from customer files and this assessment (such as weight, BMI, and treatment area surface area for body shaping projects, obtained through standardized measurement tools or analysis algorithms based on uploaded images; and subdivided unit capacity and skin thickness assessment for facial filling projects), as well as relevant aesthetic parameters. 2) Physician operation characteristics, extracted through statistical learning by analyzing the actual usage records of consumables by the attending physician in similar projects in the historical database, such as the physician's average usage per unit area preference, commonly used brand conversion coefficient, and the dispersion of historical usage (reflecting its stability).

[0081] The calculation is performed using a trained usage prediction model. This invention employs an offline-trained, online-service machine learning model (such as Gradient Boosting Decision Tree (GBDT) or Random Forest) as the core prediction engine. The model's training data comes from historically completed service case data with "good" results. Its features are the aforementioned customer and physician characteristics, and its labels are the actual, verified, and reasonable consumption of consumables in the project. The model establishes a prediction function by learning the complex nonlinear relationship between "features" and "usage" from a large number of high-quality cases.

[0082] Finally, personalized dosage recommendations are generated and integrated with rule-based review. When a new service order is created, the system inputs a real-time feature vector into the prediction model, which outputs a basic "algorithm-recommended dosage." This recommended value then undergoes verification and fine-tuning through a "safety and rationality rule base" pre-defined by medical experts. For example, the rule base might specify an absolute safety upper limit for a filler injected at a single point in the nose. If the algorithm-recommended value exceeds this upper limit, the system will adopt the upper limit as the final "dynamic baseline dosage" and indicate the reason. The final generated "dynamic baseline dosage" will serve as an intelligent and personalized benchmark for consumable association, early warning comparison, and inventory pre-positioning at each stage of the service path for this order.

[0083] S3: Enables dynamic business collaboration based on knowledge graphs and role-based access control;

[0084] Specifically, S3 includes the following steps:

[0085] S3.1: Based on an open role model, the system incorporates a flexible open role management model, defining different role types for consultants, doctors, nurses, warehouse staff, and financial personnel. Its key feature is that permission allocation extends beyond access control of functional modules, reaching a finer granular level to the visibility, editability, and operability of entities and relationships within the knowledge graph. For example, a consultant role can view and edit basic customer information and consultation record entities, but cannot view specific treatment consumable details; a warehouse staff role can see consumable inventory and associated order entities, but cannot view customer medical record details. Simultaneously, a "context template" is configured for each role's workbench, defining which relevant entity information the system should automatically associate and extract from the knowledge graph for sidebar display when a user focuses on a core business object (such as a specific customer or order).

[0086] S3.2: Context-Aware Intelligent Information Aggregation and Push. When a user (e.g., a physician) enters a specific work scenario (e.g., developing a treatment plan for a client), the system queries the knowledge graph in real time based on the "context template" defined for that role and scenario in S301. The execution process is as follows: The system uses the current focus business object (the client) as the starting point for the query, traversing and aggregating information along the preset high-value relationship paths in the graph. For example, it automatically retrieves and presents: 1) the client's historical consumption items and their effect evaluations (client-acceptance -> item relationship); 2) the marked allergy history or contraindication entities (client-have -> health status relationship); 3) the consultation record entities created by the consultant in this service process and the extracted key needs (e.g., "desire to improve nasal bridge height, preferring to use implants"). This information is not simply listed, but intelligently sorted and summarized based on semantic relevance and timeliness, dynamically generating a "client panorama view" in the sidebar or panel of the collaborative workbench, providing users with immediate and comprehensive data support for decision-making.

[0087] S3.3: Process-driven automated task generation and state synchronization. The system transforms the "service path" defined in S2 into an executable task flow engine. Its key feature is that task generation is automatically triggered by state transition events of the service path and deeply bound to entities in the knowledge graph. Specifically, when a physician confirms a treatment plan in the system (i.e., pushes the service path into the "plan confirmation" stage) and completes the consumables list association, this operation is captured as an event by the task flow engine. The engine automatically instantiates and distributes the following tasks according to predefined rules, and each task contains a link to a relevant knowledge graph entity, as follows:

[0088] Generate a "Consumables Preparation Task" and assign it to the warehouse manager role. The task content includes a list of consumables entities that need to be stocked, their respective orders, and customer information links. The task target status is "Pending Collection" or "Delivered to Designated Treatment Room".

[0089] Generate a "preoperative preparation task" and assign it to the nurse role. The task content includes customer information, project information, and requirements for the instruments and venues to be prepared.

[0090] When "Pre-settlement Bill Generation" is triggered, the system automatically generates an estimated cost detail based on the project pricing, consumables list, and discount rules, pushes it to the finance module, and creates a pre-settlement bill entity to be confirmed, which is then associated with the order.

[0091] All generated tasks are centrally displayed on a unified "collaborative dashboard," where different roles can only see and operate the tasks assigned to them. Any changes to the task status (such as "pending," "in progress," or "completed") are synchronized to the dashboard in real time and can be notified to relevant roles via event notification mechanisms (such as in-app messages and application push notifications). The task completion status also updates the stage status of the service path in reverse.

[0092] S3.4: Incremental Construction of Knowledge Graph under Distributed Collaboration. To transform tacit knowledge scattered across various roles' business processes into reusable structured knowledge, the system provides an embedded knowledge annotation and contribution tool. Its key feature is that it allows authorized roles to directly semantically annotate text information within the business context and submit it to the knowledge graph's pending review queue. The specific process includes:

[0093] For example, a doctor might write in the postoperative record, "This client has thin nasal skin, so extra sculpting was performed when a certain brand of implant was selected." The doctor can select "thin nasal skin" and "certain brand of implant," and through the right-click menu or toolbar, label them as "client characteristic" and "consumables" entities respectively, establishing a semantic relationship of "client characteristic - impact - consumables handling method."

[0094] Submission and review: The tagged content, along with its context and submitter information, is entered into the system's knowledge review queue. It is then reviewed by a designated knowledge administrator or senior physician.

[0095] After the integration and inclusion process is approved, the system calls the knowledge fusion algorithm to integrate the new triples with the existing knowledge graph (e.g., to determine whether "thin nasal skin" is equivalent to the existing "skin thickness" entity). Finally, the knowledge graph is updated incrementally to continuously enrich the professional knowledge of the domain.

[0096] S4: Provides intelligent decision support based on knowledge graph analysis and machine learning. To transform aggregated, multi-dimensional, and interconnected data into deep insights that guide operations and development, it has built an intelligent decision support engine based on knowledge graph analysis and machine learning. By mining complex relationships and patterns between data, it provides forward-looking and quantifiable decision-making basis in three aspects: in-depth customer value mining, precise resource allocation, and business structure optimization. Specifically, it includes:

[0097] Based on graph embedding and community discovery, a 360° customer profile and personalized intelligent recommendation are generated using the topological structure of the knowledge graph. This is followed by in-depth analysis and recommendations. The specific steps are as follows:

[0098] Customer graph vectorization (graph embedding) takes all "customer" entities in the knowledge graph as the core, extracts their 2-3 degree relationship subgraphs (including entities such as customers, projects, consumables, effect tags, etc., and the relationships between them), and applies graph embedding algorithms (such as TransE, Node2Vec, or GraphSAGE) to map each customer and its associated complex network structure into a low-dimensional dense vector. This vector comprehensively encodes multi-dimensional information such as the customer's consumption history, project preferences, consumable brand preferences, and effect feedback, forming a deep "customer profile."

[0099] Customer community discovery and pattern mining involves clustering analysis of all customer vectors (e.g., using K-means or DBSCAN algorithms) to automatically identify customer communities with similar consumption patterns and characteristics. For example, it may identify communities such as "enthusiastic about anti-aging phototherapy projects and preferring high-end brands" or "focused on facial fillers and pursuing natural results," and analyze the common feature paths of each community on the graph.

[0100] Personalized project and solution recommendations: When serving a specific client, the system first calculates their client vector and identifies their community and vector neighbors. By analyzing the projects commonly accepted by clients in that community and neighbors (based on the graph's client-subsequent acceptance-project relationship), and combining graph reasoning (such as complementary effects between projects and synergistic use of consumables), a personalized list of recommended projects is generated. Furthermore, by incorporating the client's current skin analysis data (added as an entity to the graph), specific project parameters or combined treatment plans can be recommended to form a "skincare plan." The rationale for the recommendation is displayed using interpretable AI technology (e.g., "Among clients with similar conditions to yours, 85% chose project B within 3 months of completing project A to consolidate the effects").

[0101] The method of combining time series analysis and knowledge graph association to accurately predict the demand for consumables enables proactive and precise management of consumable demand. The specific steps are as follows:

[0102] A spatiotemporal correlation prediction feature set is constructed, and for each type of consumable, the system builds a time series of its historical consumption. At the same time, strongly correlated predictive factors are extracted from the knowledge graph to form a feature vector, including: the future scheduled project dates and quantities (mapped to the demand plan of each consumable through the project-include-consumable relationship), seasonality index (learned from historical data), marketing activity plans, and the scheduling of associated physicians;

[0103] The training and execution of the integrated prediction model employs a fusion model for prediction. For example, a seasonal autoregressive integral moving average (SARIMA) model or a long short-term memory network (LSTM) is used to capture the basic trends and cycles of the time series. At the same time, a gradient boosting tree model (such as XGBoost) is used with the above-mentioned spectral correlation features as input to predict the deviation or increment of the trend. The final prediction result is a weighted sum or combination of the two outputs, and the prediction interval is also output.

[0104] The system generates procurement recommendations and inventory strategies. Based on forecasted demand, current inventory, safety stock threshold, procurement lead time, and supplier minimum order quantity, it automatically generates suggested procurement plans using inventory optimization models (such as variants of the economic order quantity model). It can also simulate costs and service levels under different procurement strategies. The forecast results are simultaneously fed back to the S2 inventory warning module to dynamically adjust safety stock parameters.

[0105] Based on graph algorithms and clustering, this multi-dimensional analysis of projects and operations provides managers with analytical tools to gain insights into business structure and optimize resource allocation, as detailed below:

[0106] Project Relationship Network and Value Analysis: Graph analysis algorithms are applied to the "project" entity subgraph of the knowledge graph. For example, community detection algorithms are used to identify "project packages" frequently consumed by the same customer groups; centrality algorithms (such as PageRank) are used to identify key projects that occupy pivotal positions in the consumption network. Simultaneously, indicators such as gross profit margin and customer satisfaction correlation are calculated for each project to create a "project value matrix," identifying star projects, cash flow projects, and projects requiring optimization.

[0107] The system analyzes physician operational efficiency and standardization, centering on the "physician" entity and aggregating relevant data. By comparing and analyzing different physicians' service path compliance, average time spent, consumable consumption efficiency (actual usage / dynamic baseline usage), post-operative customer evaluations, and repurchase rates when performing similar projects, it generates multi-dimensional physician performance profiles. The system can automatically identify best practice patterns and also detect abnormal operational patterns (such as a physician consistently consuming high amounts of consumables without a corresponding increase in customer satisfaction), providing a reference for quality management.

[0108] High-margin service portfolio mining: Utilizing association rule mining algorithms (such as Apriori or FP-Growth), analyze the frequency and profit contribution of combinations of items and specific consumable brands in historical orders. Identify service portfolios that frequently co-occur and whose overall profit margin is significantly higher than the sum of the profits from individual sales, providing direct data support for designing bundled packages and developing cross-selling strategies.

[0109] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A method for intelligent data management and business collaboration in an ERP system for the medical aesthetics industry, characterized in that: The method includes: S1: Construct an intelligent integration layer for multi-source heterogeneous data in the medical aesthetics field, and use knowledge graph technology to perform semantic modeling, standardized extraction and unified storage of operational data in the medical aesthetics industry; S2: Design a consumables intelligent association and early warning management module based on "service path" to perform association management, real-time early warning and closed-loop traceability of consumables according to the standardized service path of medical aesthetics projects; S3: Enables dynamic business collaboration based on knowledge graphs and role permissions, automatically generates and synchronizes tasks according to role configuration context information, and supports collaboration among multiple roles under a unified view; S4: Provides intelligent decision support based on graph analysis and machine learning, offering decision-making basis through customer profiling, demand forecasting, and operational analysis.

2. The intelligent data management and business collaboration method for the medical aesthetics industry ERP system according to claim 1, characterized in that, S1 specifically includes: S1.1: Construct an ontology model for the medical aesthetics field, defining entity classes, attributes, and their semantic relationships in the medical aesthetics field; S1.2: Through multi-path knowledge extraction and transformation, multi-source heterogeneous data is transformed into standard triples or attribute graph data that conform to the ontology model; S1.3: A hybrid storage architecture with multiple graph storage engines is used to store knowledge data, and a consistent query interface is provided through a unified data access layer.

3. The intelligent data management and business collaboration method for the medical aesthetics industry ERP system according to claim 2, characterized in that, In S1.2, multi-path knowledge extraction includes: Structured data is extracted using pattern mapping. Extraction of semi-structured data is achieved by combining a rule engine with natural language processing. Information extraction based on deep learning models is used for unstructured data.

4. The intelligent data management and business collaboration method for the medical aesthetics industry ERP system according to claim 1, characterized in that, S2 specifically includes: S2.1: Build and maintain a standard "service path" library for medical aesthetic projects, with each path associated with a "standard consumables set" for each service stage; S2.2: Instantiate the service path when a service order is created and automatically associate it with the expected consumables list; S2.3: Real-time monitoring and multi-layered intelligent early warning of consumable usage through a rules engine; S2.4: Achieve closed-loop traceability of consumable consumption and feed back actual consumption data to the service path database.

5. The intelligent data management and business collaboration method for the medical aesthetics industry ERP system according to claim 4, characterized in that, S2.3 also includes a self-learning and dynamic optimization mechanism for early warning rules, including: constructing an early warning event-result analysis sample library; Risk prediction model based on supervised learning training; The parameters and logic of the early warning rules are dynamically adjusted based on the model evaluation results.

6. The intelligent data management and business collaboration method for the medical aesthetics industry ERP system according to claim 4, characterized in that, The baseline usage of the "standard consumables set" in S2.1 is generated through a personalized dynamic optimization algorithm, including: constructing a feature vector of usage influencing factors; The usage prediction model trained by the application calculates the recommended usage based on the algorithm. The final dynamic baseline usage is generated by verifying the safety and rationality rule base.

7. The intelligent data management and business collaboration method for the medical aesthetics industry ERP system according to claim 1, characterized in that, S3 specifically includes: S3.1: Configure permissions and context templates based on an open role model; S3.2: Query the knowledge graph in real time based on the context template, aggregate and push relevant information; S3.3: Automatically generate tasks based on service path status and synchronize them to the collaboration dashboard; S3.4: Supports semantic annotation during business processes to enable incremental construction of knowledge graphs.

8. The intelligent data management and business collaboration method for the medical aesthetics industry ERP system according to claim 1, characterized in that, S4 specifically includes: building a 360° customer profile based on graph embedding and community discovery and making personalized project recommendations; By integrating time series analysis and graph correlation, accurate prediction of consumable demand can be achieved; Multidimensional analysis of project association networks, physician operational efficiency, and high-margin service combinations based on graph algorithms and clustering.

9. The intelligent data management and business collaboration method for an ERP system in the medical aesthetics industry according to claim 8, characterized in that, The personalized project recommendation includes: extracting a customer subgraph and performing graph embedding to obtain a customer vector; Discover customer communities and analyze their common characteristics through clustering; A list of recommended items and reasons for recommendation are generated based on customer vectors, community features, and graph reasoning.

10. The intelligent data management and business collaboration method for an ERP system in the medical aesthetics industry according to claim 8, characterized in that, The accurate prediction of consumable demand includes: constructing a prediction feature set that integrates time series and map correlation features; Demand forecasting is performed using an integrated forecasting model. Procurement recommendations are generated based on forecast results, and inventory strategies are dynamically adjusted.

Images