Knowledge graph-based esophageal cancer postoperative rehabilitation education strategy construction method and system

Abstract

The application relates to the technical field of medical health information, and particularly discloses an esophageal cancer postoperative rehabilitation education strategy construction method and system based on a knowledge graph. A domain ontology model fusing a dietary tendency entity is constructed, a structured postoperative rehabilitation knowledge graph is created, and a patient rehabilitation state model is designed, wherein an evaluation and labeling process of individualized dietary tendency affected by genetic and environmental factors is particularly integrated. Through an education strategy matching engine, a corresponding matching rule is activated according to a patient multi-dimensional state instance containing a dietary tendency label, education content that conforms to a current rehabilitation stage and a health problem and is adapted to a specific dietary tendency is intelligently retrieved and filtered from the knowledge graph, and finally a highly personalized rehabilitation education strategy is integrated and generated. The application realizes a change from general education to accurate and individualized guidance, and significantly improves the pertinence and practicality of rehabilitation education.

Description

Technical Field

[0001] This invention relates to the field of medical and health information technology, and specifically discloses a method and system for constructing postoperative rehabilitation education strategies for esophageal cancer based on knowledge graphs. Background Technology

[0002] In the field of postoperative rehabilitation management for esophageal cancer, traditional patient health education models typically rely on standardized educational materials and uniform educational pathways. These models are often based on fixed timelines or common disease stages, providing patients with universal rehabilitation guidance. They struggle to adequately address the individualized needs of patients influenced by genetic background, living environment, cultural habits, and specific disease contraindications, especially in the crucial area of ​​diet and nutrition. While existing digital health education tools, such as mobile applications or information push systems, improve the convenience of information access, their underlying content organization is often fragmented, and their push logic is relatively simple. Essentially, they have not broken through the one-to-many standardized paradigm, failing to achieve a deep match and precise adaptation between educational strategies and the multidimensional, dynamic individual states of patients.

[0003] In recent years, knowledge graphs, as a technology for organizing and managing knowledge in complex domains, have gained widespread attention and application in the healthcare field. By integrating medical concepts, entities, and their relationships through structured semantic networks, they offer the possibility of building intelligent clinical decision support systems. Existing research largely focuses on using knowledge graphs to assist in diagnosis, medication recommendations, or medical question answering, while their application in patient rehabilitation education is still in the exploratory stage. Even where they are applied, existing methods tend to focus on associating educational content based on a single dimension such as disease or symptoms, failing to systematically incorporate highly individualized factors that influence rehabilitation outcomes, such as long-term dietary tendencies shaped by both genes and environment, into the knowledge system and reasoning logic. This results in significant limitations in the individual adaptability of the generated guidance recommendations.

[0004] Therefore, a significant gap exists in the current technological landscape: the lack of a technological solution capable of systematically integrating patients' structured medical knowledge, real-time dynamic rehabilitation status, and stable individual characteristics, particularly dietary preferences, to automatically construct highly personalized rehabilitation education strategies. This limits the progress of postoperative rehabilitation management from standardization to precision, and also affects patients' acceptance, adherence, and ultimate rehabilitation outcomes regarding health education content. An innovative method is urgently needed that can deeply encode individual differences into a knowledge system, enabling intelligent matching and dynamic generation of educational content based on this. Summary of the Invention

[0005] To achieve the above objectives, this application provides the following technical solution:

[0006] A method for constructing a knowledge graph-based rehabilitation education strategy for esophageal cancer surgery includes the following steps:

[0007] S1. Construct an ontology model for the domain of health education for postoperative rehabilitation of esophageal cancer, and determine the core entity types, attributes of each entity type, and semantic relationships between the core entity types in the ontology model.

[0008] S2. Based on the domain ontology model, construct a knowledge graph for postoperative rehabilitation of esophageal cancer, acquire and structure authoritative medical knowledge in the field of postoperative rehabilitation of esophageal cancer, generate knowledge triples with the core entity type as nodes and the semantic relationship as edges according to the entity type and semantic relationship defined by the domain ontology model, and store the knowledge triples in the graph database to form the knowledge graph for postoperative rehabilitation of esophageal cancer.

[0009] S3. Define a patient rehabilitation status model. Based on the patient status parameter entity type and dietary tendency entity type in the domain ontology model, construct a multi-dimensional status description framework. The framework determines the patient's individualized dietary tendency label through a preset evaluation process and maps the label and other rehabilitation information of the individual patient together into identifiable standardized status parameter instances and dietary tendency instances in the knowledge graph.

[0010] S4. Design and implement an education strategy matching engine. Based on the knowledge graph of postoperative rehabilitation of esophageal cancer, construct a strategy matching rule base. The strategy matching rule base contains logical rules for associating specific patient status, specific dietary preferences and appropriate education content.

[0011] S5, when a specific patient state instance containing dietary preference tags is input as defined by the patient rehabilitation state model, the education strategy matching engine traverses the strategy matching rule base, activates the rules that match the current patient state instance and dietary preference tags, retrieves the educational content entity nodes that are associated with and match the dietary preference tags from the knowledge graph, and integrates the retrieved nodes and their related paths to generate a personalized rehabilitation education strategy that incorporates dietary adaptability.

[0012] Furthermore, in S1, the specific process of constructing the domain ontology model includes:

[0013] Define the entity type for the rehabilitation phase, with attributes including at least the phase identifier, phase name, start time, and end time.

[0014] Define a health problem entity type, whose attributes must include at least a problem identifier, a problem description, and the recovery stage to which it belongs;

[0015] Define an educational content entity type whose attributes include at least a content identifier, a content title, a content carrier format, and a detailed description text; this entity type has an applicability preference attribute, which is used to associate one or more dietary preference entity instances.

[0016] Define a patient status parameter entity type, whose attributes include at least parameter identifier, parameter type, and parameter value range. The parameter type includes time-series parameters, symptom-based parameters, and behavioral compliance parameters.

[0017] Define dietary preference entity types, whose attributes include at least preference identifiers and preference descriptions. These entity types are used to characterize food preferences or taboo patterns that are determined by genetic factors, environmental and cultural factors, or disease contraindications and need to be differentiated in postoperative nutritional support.

[0018] Define the semantic relationships between the core entity types, which include at least the following: the rehabilitation stage includes health problems, health problems correspond to educational content, patient status parameters indicate the rehabilitation stage, patient status parameters trigger attention to specific health problems, and dietary tendencies affect the selection of specific educational content.

[0019] Furthermore, the specific process of defining the patient rehabilitation status model in S3 includes:

[0020] S31. Determine the set of composite state parameters: Select key physiological and behavioral parameters for strategy matching from the patient state parameter entity types of the domain ontology model, and select the tendency dimensions to be evaluated from the dietary tendency entity types, together forming the set of composite state parameters.

[0021] S32. Design a dietary preference assessment process: For the dietary preference dimension in the composite state parameter set, design a structured assessment process, define mutually exclusive basic dietary preference categories, design corresponding assessment questions or observations for each category, collect patient feedback or observation data through an interactive interface, and classify the data into one or more specific basic dietary preference categories according to preset mapping rules to generate a personalized dietary preference label set for the patient.

[0022] S33. Constructing a state instantiation and tendency fusion process: Establish a data processing interface to receive individual patient physiological data from an external system and dietary tendency data obtained through the assessment process; the data processing interface extracts and formats the individual patient physiological data according to the requirements of the composite state parameter set, and converts the individual dietary tendency tag set into a data format that conforms to the knowledge graph dietary tendency entity instance specification; the formatted physiological data and dietary tendency data are merged to form a complete original state dataset;

[0023] S34. Apply the state assessment and mapping rules to define the assessment rules for mapping the physiological parameters or parameter combinations in the composite state parameter set to standardized state labels. Transform the physiological data part in the original state dataset into a set composed of several standardized state labels, and directly use the dietary tendency data part as the tendency label set. The standardized state label set and the tendency label set together constitute the current patient state instance.

[0024] S35. Implement knowledge graph mapping for composite state instances, and associate and bind each standardized state label and each tendency label in the current patient state instance with the corresponding patient state parameter related nodes and dietary tendency entity nodes in the esophageal cancer postoperative rehabilitation knowledge graph, so that the composite state instances output by the patient rehabilitation state model can be semantically understood and processed by the knowledge graph.

[0025] Furthermore, the specific process of designing and implementing the education strategy matching engine in S4 includes:

[0026] S41. Construct a strategy matching rule base for fusion tendency: Based on the entity relationships and medical logic in the knowledge graph of esophageal cancer postoperative rehabilitation, write condition-action rules to form the strategy matching rule base. The condition part of each rule consists of a logical expression about the standardized state label in the patient state instance, and may selectively include a logical expression about the dietary tendency label. The action part of the rule defines the type of target educational content entity node to be queried from the knowledge graph when the condition is met, the graph traversal path to be followed, and the filtering conditions for the applicable tendency attributes of the content node.

[0027] S42. Implement graph traversal for rule triggering and tendency perception. After receiving the current patient state instance, the education strategy matching engine performs matching calculations on the standardized state label set and tendency label set in the knowledge graph with the condition parts of all rules in the strategy matching rule base. For rules whose conditions are fully met, the engine activates the rule. When the engine executes the action part of the rule, it takes the node in the knowledge graph corresponding to the state label or tendency label that meets the conditions as the starting point and traverses along the specified semantic relationship edge. During the traversal or after locating the candidate education content node, it checks whether the applicable tendency attribute of the candidate content node is compatible with the tendency label in the patient state instance according to the filtering conditions defined in the action part, and only retains the compatible nodes as valid search results.

[0028] S43. Perform educational content aggregation, tendency adaptation, and strategy generation, and collect all valid educational content entity nodes located by the triggered rules; for each valid node, backtrack along the knowledge graph path to extract its context information, which includes its directly associated health problems, associated rehabilitation stages, and applicable tendency attribute values. Organize the extracted educational content and its context information according to a preset template. In the generated personalized rehabilitation education strategy document, clearly list the recommended educational content, the corresponding health problem background, the expected rehabilitation goals, and annotate the content that adapts to the patient's specific dietary tendencies.

[0029] S44. Define the strategy output and execution interface, and output the generated personalized rehabilitation education strategy document to the designated application terminal through a standardized data exchange format; the application terminal presents the educational content to the patient or medical staff in a set presentation format based on the structured information of the strategy document, and can selectively display or sort the content according to the applicable preference label.

[0030] Furthermore, the specific preference categories defined for the dietary preference entity type include:

[0031] Dairy tolerance tendency, cereal preference tendency, dietary fiber acceptance tendency, food texture sensitivity tendency, and specific nutrient supplementation tendency;

[0032] The applicable tendency attribute value of the educational content entity node is used to indicate which dietary tendency category the recommended diet or nutritional advice is compatible with or specifically designed for.

[0033] Furthermore, the dietary preference assessment process designed in S32 is specifically implemented by providing a digital assessment interface, which sequentially presents groups of questions related to different basic dietary preference categories;

[0034] The design of the question set covers inquiries about genetically related symptoms, investigations into past dietary habits, and confirmation of cultural and religious dietary restrictions; based on the patient's answers to the question set, the system automatically applies the preset mapping rules to generate the personalized dietary tendency label set.

[0035] Furthermore, the education strategy matching engine integrates a timeline scheduling module, which maintains a standard rehabilitation timeline bound to the rehabilitation stage entity;

[0036] When the time-series parameters in the patient recovery status model indicate that the patient has entered a certain preset time node, the timeline scheduling module automatically generates a benchmark rule that is strongly correlated with the time node.

[0037] When performing matching, the education strategy matching engine applies such benchmark rules to obtain a set of universal educational content candidates for that time point. Based on the dietary preference tags in the patient status instance, it filters and adapts this set of candidates, integrates the adapted content with personalized content triggered by other statuses, and together they constitute the final education strategy.

[0038] Furthermore, the policy matching rule base constructed in S4 is designed according to the principles of hierarchical and preference-specific adaptation, specifically including:

[0039] The design of the first-level general rules is based solely on the core physiological state labels in the patient rehabilitation state model and does not include restrictions on dietary preferences. The educational content nodes retrieved in the action part typically do not have specific applicable preference attributes or have general preference attributes, and are used to provide universal rehabilitation guidance.

[0040] The design includes a two-level preference screening rule, where the condition part includes the determination of one or more dietary preference tags; and the action part, when retrieving educational content nodes, requires that the applicable preference attribute value of the node must match the preference tag determined in the condition.

[0041] Design a three-level composite trigger rule, the condition part of which is a complex logical expression, and simultaneously associates a specific combination of physiological state labels with a specific combination of dietary tendency labels.

[0042] When performing matching, the education strategy matching engine processes the first-level, second-level, and third-level rules in parallel; for the triggered second-level and third-level rules, the engine strictly performs content filtering based on preference tags.

[0043] The generated educational strategy integrates all effective educational content that matches the triggering rules and passes the preference filtering, sorting and organizing it according to rule level and content priority during integration.

[0044] Furthermore, the method also includes S6, providing a visual tendency association and strategy editing tool:

[0045] A graphical user interface is provided, which can display the associated subgraphs between the dietary tendency entity nodes, health problem nodes, and educational content nodes in the knowledge graph of esophageal cancer postoperative rehabilitation.

[0046] The graphical user interface provides dedicated controls that allow users to: a) graphically bind or unbind one or more dietary preference nodes to educational content nodes in the knowledge graph, thereby modifying their applicable preference attributes; b) intuitively add or modify dietary preference-related conditional logic for rules in the strategy matching rule base by connecting dietary preference nodes with conditional expressions in the rule editing panel.

[0047] The tool translates the user's graphical operations into commands to modify the attribute values ​​of the knowledge graph or to update the logical rules base, and saves them synchronously.

[0048] The generation process of the personalized rehabilitation education strategy supports the invocation of strategy templates based on propensity tags:

[0049] Multiple strategy templates are predefined for typical rehabilitation scenarios and specific dietary preference combinations; each strategy template is associated with a specific patient physiological state pattern and dietary preference tags, and encapsulates a knowledge graph query path and content organization logic containing preference filtering conditions.

[0050] When the education strategy matching engine identifies that the current patient state instance matches the physiological state pattern associated with a predefined strategy template, and its dietary tendency label set completely matches or supersets the tendency labels associated with the template, the strategy template is invoked first.

[0051] When a strategy template is invoked, the engine instantiates and populates the template based on the specific parameter values ​​of the current state instance, and directly executes the query logic encapsulated within the template to generate a strategy draft.

[0052] If no perfectly matching strategy template is found, the engine will instead execute a general rule matching process based on the strategy matching rule base to generate a strategy.

[0053] According to a second aspect of the present invention, the present invention claims protection for a knowledge graph-based system for constructing postoperative rehabilitation education strategies for esophageal cancer, comprising:

[0054] One or more processors;

[0055] A memory that stores one or more programs, which, when executed by one or more processors, enable the processors to implement the method for constructing a knowledge graph-based esophageal cancer postoperative rehabilitation education strategy.

[0056] This invention relates to the field of medical and health information technology, specifically disclosing a method and system for constructing postoperative rehabilitation education strategies for esophageal cancer based on knowledge graphs. By constructing a domain ontology model that integrates dietary preference entities, a structured postoperative rehabilitation knowledge graph is created. Based on this, a patient rehabilitation status model is designed, which specifically integrates the assessment and labeling process for individualized dietary preferences influenced by both genetic and environmental factors. The core lies in using an education strategy matching engine. Based on multi-dimensional patient status instances including dietary preference labels, corresponding matching rules are activated to intelligently retrieve and filter educational content from the knowledge graph that is both consistent with the current rehabilitation stage and health issues and suitable for the patient's specific dietary preferences. This is then integrated to generate a highly personalized rehabilitation education strategy. This invention achieves a shift from generalized education to precise and individualized guidance, significantly improving the targeting and practicality of rehabilitation education. Attached Figure Description

[0057] Figure 1 A flowchart illustrating the process of constructing a knowledge graph-based postoperative rehabilitation education strategy for esophageal cancer, as claimed in an embodiment of the present invention.

[0058] Figure 2 The second flowchart is a method for constructing a knowledge graph-based rehabilitation education strategy for esophageal cancer surgery, as claimed in an embodiment of the present invention.

[0059] Figure 3 The third flowchart is a method for constructing a knowledge graph-based rehabilitation education strategy for esophageal cancer surgery, as claimed in an embodiment of the present invention. Detailed Implementation

[0060] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0061] The terms "first," "second," and "third" in this application are for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first," "second," or "third" may explicitly or implicitly include at least one of those features. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified. All directional indications in the embodiments of this application, such as up, down, left, right, front, back, etc., are only used to explain the relative positional relationships and movements between components in a specific orientation as shown in the accompanying drawings. If the specific orientation changes, the directional indications will change accordingly. Furthermore, the terms "including" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0062] References to embodiments herein mean that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0063] According to a first embodiment of the present invention, the present invention claims protection for a method for constructing a knowledge graph-based postoperative rehabilitation education strategy for esophageal cancer, referring to... Figure 1 This includes the following steps:

[0064] S1. Construct an ontology model for the domain of health education for postoperative rehabilitation of esophageal cancer, and determine the core entity types, attributes of each entity type, and semantic relationships between the core entity types in the domain ontology model.

[0065] S2. Based on the domain ontology model, construct a knowledge graph for postoperative rehabilitation of esophageal cancer, acquire and structure authoritative medical knowledge in the field of postoperative rehabilitation of esophageal cancer, generate knowledge triples with core entity types as nodes and semantic relationships as edges according to the entity types and semantic relationships defined by the domain ontology model, and store the knowledge triples in the graph database to form a knowledge graph for postoperative rehabilitation of esophageal cancer.

[0066] S3. Define the patient rehabilitation status model. Based on the patient status parameter entity type and dietary tendency entity type in the domain ontology model, construct a multi-dimensional status description framework. The framework determines the patient's individualized dietary tendency label through a preset evaluation process and maps this label and other rehabilitation information of the individual patient into identifiable standardized status parameter instances and dietary tendency instances in the knowledge graph.

[0067] S4. Design and implement an education strategy matching engine. Based on the knowledge graph of esophageal cancer postoperative rehabilitation, construct a strategy matching rule base. The strategy matching rule base contains logical rules for associating specific patient status, specific dietary preferences and appropriate education content.

[0068] S5, when a specific patient state instance containing dietary preference labels is input as defined by the patient rehabilitation state model, the education strategy matching engine traverses the strategy matching rule base, activates the rules that match the current patient state instance and dietary preference labels, retrieves the educational content entity nodes that are associated with and match the dietary preference labels from the knowledge graph, and integrates the retrieved nodes and their related paths to generate a personalized rehabilitation education strategy that incorporates dietary adaptability.

[0069] In this embodiment, an ontology model for the field of health education in esophageal cancer postoperative rehabilitation is constructed. This step aims to establish a formal conceptual framework for this field. First, by systematically reviewing clinical pathways, medical guidelines, and expert consensus on esophageal cancer postoperative rehabilitation, core defining concepts in this field are abstracted and summarized. Then, an entity type is defined for each identified core concept type, and the descriptive characteristics, i.e., attributes, that each entity type must possess are clearly specified. Finally, based on medical logic and clinical practice, the inherent, semantically meaningful connections between these different entity types are clearly defined. By completing the above work, a rigorously structured domain ontology model is formed, providing unified semantic specifications and constraints for the subsequent structured organization of knowledge.

[0070] Based on a domain ontology model, a knowledge graph for post-esophageal cancer rehabilitation is constructed. This step transforms authoritative, unstructured medical knowledge into a computable, structured knowledge network. Specifically, knowledge texts related to post-esophageal cancer rehabilitation are systematically collected from peer-reviewed medical literature, clinical guidelines, textbooks, and publicly available information from authoritative medical institutions. Based on the domain ontology model established in step S1, the collected knowledge is parsed and labeled manually or semi-automatically: identifying entities mentioned in the text that correspond to entity types in the ontology, extracting the specific attribute values ​​of these entities, and determining whether the relationships stated between entities correspond to semantic relationships defined in the ontology. The results of each identification and judgment are organized into standardized knowledge triples in the form of entity-relation-entity or entity-attribute-value. Subsequently, a graph database is selected as the storage infrastructure, and all generated knowledge triples are imported into it. In this graph database, each entity becomes a node, each relation becomes a directed edge connecting two nodes, and each attribute becomes a key-value pair belonging to its entity node. In this way, a massive amount of discrete knowledge points are connected into a large-scale, interconnected semantic network, namely the knowledge graph of esophageal cancer postoperative rehabilitation.

[0071] Defining the Patient Rehabilitation State Model: The goal of this step is to create a mechanism that can transform the complex and ever-changing individual circumstances of patients in the real world into standardized inputs that a knowledge graph system can accurately recognize. First, the model requires a multi-dimensional state description framework. The dimensions of this framework are directly derived from the patient state parameters and dietary preference entity types defined in the domain ontology model. For the dietary preference dimension, the model pre-defines a structured assessment process. This process includes a series of logically coherent steps aimed at identifying and confirming food preferences or contraindications that significantly influence postoperative nutritional support plans, formed by the patient's genetic background, long-term living environment, or specific disease contraindications, through questioning, observation, or analysis of specific data, and ultimately assigning one or more standardized dietary preference labels. For other rehabilitation information such as postoperative time, symptom presentation, and behavioral records, the model maps them to corresponding standardized state parameter instances according to predefined transformation rules. Finally, the model combines and encapsulates the generated dietary preference labels with other state parameter instances to form a complete, machine-readable patient rehabilitation state instance.

[0072] Designing and implementing an educational strategy matching engine: The core of this step is building an intelligent matching-retrieval system. The system is based on a strategy matching rule base, where each rule is a clearly defined condition-action instruction. The condition part of a rule is a set of logical judgment statements based on various tags in a patient's recovery state instance, including state parameter tags and dietary tendency tags. The action part of the rule specifies how to perform a navigable search within the knowledge graph constructed in step S2 to find matching educational content when the condition is met. These search paths essentially traverse the knowledge graph from one or more nodes related to the current state, along specific semantic relationship edges, ultimately reaching the node representing the educational content type.

[0073] When a specific instance of a patient's recovery status, tagged with dietary preferences, is input into the education strategy matching engine, the engine begins its operation. It compares and performs logical operations on each tag in the input instance against the conditional parts of all rules in the strategy matching rule base. Rules whose conditions are fully met are activated. For each activated rule, the engine strictly follows the instructions in its action section, initiating a traversal search within the knowledge graph. During the search process, or after obtaining preliminary candidate educational content nodes, the engine specifically verifies whether these nodes are tagged as compatible with the current patient's dietary preferences. Only content nodes that pass this compatibility check are retained as valid results. Finally, the engine aggregates all valid educational content nodes retrieved by the activated rules and, according to certain strategies such as categorizing by associated health issues or sorting by urgency, integrates and packages these nodes with their contextual information in the knowledge graph—for example, which health issue the content addresses and which recovery stage it belongs to—ultimately outputting a well-structured, personalized rehabilitation education strategy plan that fully considers the patient's dietary characteristics.

[0074] Furthermore, in S1, the specific process of constructing the domain ontology model includes:

[0075] Define the entity type for the rehabilitation phase, with attributes including at least the phase identifier, phase name, start time, and end time.

[0076] Define a health problem entity type, whose attributes must include at least a problem identifier, a problem description, and the recovery stage to which it belongs;

[0077] Define an educational content entity type whose attributes include at least a content identifier, a content title, a content carrier format, and a detailed description text; this entity type has an applicability preference attribute, which is used to associate one or more dietary preference entity instances.

[0078] Define the patient status parameter entity type, whose attributes include at least parameter identifier, parameter type and parameter value range. The parameter type includes time-series parameters, symptom-type parameters and behavioral compliance parameters.

[0079] Define dietary preference entity types, whose attributes include at least preference identifiers and preference descriptions. These entity types are used to characterize food preferences or taboo patterns that are determined by genetic factors, environmental and cultural factors, or disease contraindications and need to be differentiated in postoperative nutritional support.

[0080] Define the semantic relationships between core entity types. The semantic relationships should include at least the following: the rehabilitation stage includes health problems, health problems correspond to educational content, patient status parameters indicate the rehabilitation stage, patient status parameters trigger attention to specific health problems, and dietary tendencies affect the selection of specific educational content.

[0081] In this embodiment, a domain conceptualization analysis was conducted. A multidisciplinary expert group, including thoracic surgeons, clinical nutritionists, rehabilitation therapists, and senior nurses, was organized to break down and analyze the entire process of esophageal cancer surgery, from hospitalization to long-term home recovery, through multiple thematic discussions. The discussions focused on clarifying the key stages involved in the recovery process, the core health issues that patients may face, the guidance that medical staff need to provide, the indicators that can be used to assess the patient's condition, and the personalized dietary factors that may affect the effectiveness of nutritional interventions.

[0082] Secondly, based on the results of conceptual analysis, each core entity type and its attributes are precisely defined.

[0083] Define the rehabilitation phase entity type: This entity is used to characterize time intervals with different medical care focuses. Its phase identifier attribute is a unique code; the phase name is a colloquial description of the phase, such as preoperative preparation; the start and end times define the boundaries of the phase on the standard rehabilitation timeline, such as day 0 to day 7 postoperatively.

[0084] Define the health problem entity type: This entity represents the physiological or psychological condition of a patient that requires attention and treatment during a specific rehabilitation phase. Its problem identifier is a unique code; the problem description is a specific explanation of the problem, such as the prevention and identification of anastomotic leakage; the rehabilitation phase attribute clarifies the period in which the problem primarily occurs by referencing the rehabilitation phase entity's identifier.

[0085] Define the educational content entity type: This entity is the carrier of specific health guidance information. Its content identifier is a unique code; the content title is a brief summary of the content; the content carrier format describes how the information is presented, such as graphic manuals, animated videos, or audio explanations; the detailed explanatory text is the core body of the guidance content. In addition, a special "applicability preference" attribute is added. This attribute value is not free text, but a link to a dietary preference entity instance, used to indicate which specific dietary preference pattern this educational content, especially its dietary recommendations, is most suitable for or compatible with.

[0086] Define the patient status parameter entity type: This entity is used to standardize the description of an individual patient's objective condition at a given moment. Its parameter identifier is a unique code; the parameter type is used for classification, such as time-series parameters like postoperative days, symptom-based parameters like pain score and cough frequency, and behavioral compliance parameters like daily steps and frequency of nutritional supplement consumption; the parameter value range specifies the reasonable numerical interval or enumerated value for the parameter.

[0087] Define dietary preference entity types: This is an innovative entity designed to formally describe individual differences influencing nutritional support. Its preference identifier is uniquely coded; the preference description is a detailed explanation of that preference, such as lactose intolerance preference: indicating that patients, due to a lack of lactase, may experience symptoms such as bloating and diarrhea after digesting dairy products, and should avoid or choose lactose-free products. Instantiation of this entity represents stable preferences or limitations shaped by genetic, environmental, cultural, or pathological factors.

[0088] Finally, the semantic relationships between entities are precisely defined. This is not a simple connection, but rather giving the connection a clear clinical meaning. For example: the relationship between rehabilitation stage and health problems indicates that a certain rehabilitation stage usually involves a series of specific health problems; the relationship between health problems and educational content indicates that there are one or more sets of standardized educational materials for a specific health problem; the relationship between patient status parameters and rehabilitation stage indicates that by analyzing certain status parameters of a patient, such as the number of days after surgery, the current major rehabilitation stage can be inferred; the relationship between patient status parameters and attention to specific health problems indicates that when a certain status parameter shows an abnormal value, such as an elevated body temperature, it should be immediately associated with a potential health problem, such as an infection; and the newly added relationship between dietary preferences and the selection of specific educational content is one of the core logics of this method. It establishes a direct link between the patient's personal dietary characteristics and the applicability of educational content, meaning that when selecting educational content for patients with a certain dietary preference, content that matches their dietary preference attribute should be given priority.

[0089] Furthermore, referring to Figure 2 The specific process of defining the patient rehabilitation status model in S3 includes:

[0090] S31. Select key physiological and behavioral parameters for strategy matching from the patient state parameter entity types in the domain ontology model, and select the tendency dimensions to be evaluated from the dietary tendency entity types, together forming a composite state parameter set.

[0091] S32. For the dietary preference dimension in the composite state parameter set, design a structured assessment process, define mutually exclusive basic dietary preference categories, design corresponding assessment questions or observations for each category, collect patient feedback or observation data through an interactive interface, and classify the data according to the preset mapping rules. For specific basic dietary preference categories, generate a set of individualized dietary preference labels for patients.

[0092] S33. Establish a data processing interface to receive individual patient physiological data from external systems and dietary preference data obtained through the assessment process. The data processing interface extracts and formats individual patient physiological data according to the requirements of the composite state parameter set, and converts the individualized dietary preference tag set into a data format that conforms to the dietary preference entity instance specification of the knowledge graph. The formatted physiological data and dietary preference data are merged to form a complete original state dataset.

[0093] S34. Apply the status assessment and mapping rules to define the assessment rules for mapping physiological parameters or parameter combinations in the composite status parameter set to standardized status labels. Transform the physiological data part in the complete original status dataset into a set composed of several standardized status labels, and directly use the dietary tendency data part as the tendency label set. The standardized status label set and the tendency label set together constitute the current patient status instance.

[0094] S35. Implement knowledge graph mapping for composite state instances. Associate and bind each standardized state label and each tendency label in the current patient state instance with the corresponding patient state parameter related nodes and dietary tendency entity nodes in the esophageal cancer postoperative rehabilitation knowledge graph, so that the composite state instances output by the patient rehabilitation state model can be semantically understood and processed by the knowledge graph.

[0095] In this embodiment, determining the composite state parameter set is the blueprint planning phase of model building. The model builder needs to review all parameters under the defined patient state parameter entity types and select the key parameters most discriminative for developing rehabilitation education strategies, such as postoperative days, current pain level NRS score, presence of swallowing choking, and oral food intake in the past 24 hours. Simultaneously, from the dietary preference entity type, all preference dimensions that need to be included in the individual assessment are selected, such as dairy tolerance and food texture preference. These two sets of selection results are combined to form a complete set of composite state parameters to be collected and evaluated. This set serves as the basis for all subsequent data processing.

[0096] Designing a dietary preference assessment process: This is a crucial preliminary step in achieving individualized dietary adaptation. The process is designed as a structured, step-by-step interactive procedure. First, based on medical and nutritional knowledge, a set of non-overlapping basic preference categories covering major dietary influencing factors are defined, such as lactose intolerance, gluten sensitivity, low-fiber tolerance, preference for liquid / semi-liquid diets, and high protein requirements. For each basic category, a set of targeted assessment questions or observation indicators are carefully designed. For example, for the lactose intolerance category, questions might include whether you often experience bloating or diarrhea after drinking milk, and whether you have been diagnosed with lactase deficiency. Then, a digital assessment interface, such as a web form or mobile questionnaire, is developed to present these questions to the patient or their caregiver in an organized manner. After the patient's interactive answers are captured by the system, the process's built-in preset mapping rules begin to work. These rules are explicit IF-THEN logic, such as IF 'bloating after drinking milk' answered 'yes' AND 'diagnosed with lactase deficiency' answered 'yes' OR 'uncertain', THEN assign the label 'lactose intolerance tendency'. The system automatically executes these rules, converting the patient's original response into a set of individualized dietary preference labels consisting of standardized preference labels such as lactose intolerance preference and low fiber preference.

[0097] The process of establishing state instantiation and tendency fusion is as follows: This step establishes a pipeline for data aggregation and formatting. The model provides a standardized data processing interface that can receive data packets from different sources. One data packet contains the patient's physiological and behavioral data, which may come from hospital information systems, wearable devices, or manually entered logs by the patient. The other data packet comes from the individualized dietary tendency label set output from step S32. The interface has an internal parser that can accurately extract the raw values ​​of the required physiological parameters from the first data packet based on the composite state parameter set determined in S31. For example, it can parse "postoperative day 5" from the medical record system and "pain score = 3" from the pain rating table, and format them into a predefined structure such as key-value pairs. At the same time, the interface transforms the label set in the second data packet into a data format that conforms to the dietary tendency entity instance specification in the knowledge graph, usually mapping each label to a unique identifier of the corresponding tendency entity. Finally, the interface merges the two formatted data into a unified and complete raw state dataset.

[0098] Application of State Assessment and Mapping Rules: This step is responsible for translating the raw data into semantic symbols that the knowledge graph system can understand. For the physiological data portion of the raw state dataset, the model has pre-defined assessment rules for each parameter or common parameter combinations. For example, for the pain score parameter, the rule might be: IF pain score >= 4, THEN map to the state label 'moderate to severe pain'; IF pain score 1-3, THEN map to 'mild pain'. The system automatically invokes these rules to process each physiological data record in the dataset, outputting a set of semantic state labels such as postoperative acute phase, mild pain, and insufficient nutritional intake. For the dietary tendency data portion of the raw state dataset, since it is already a standardized label output from S32, it does not need to be mapped again and can be directly used as the tendency label set. At this point, the raw data has been transformed into two clear label sets, and their union constitutes a patient recovery state instance representing the patient's current condition.

[0099] Implementing knowledge graph mapping for composite state instances: This is the handshake between the state model and the knowledge graph, ensuring that the abstract labels in the state instance can find specific corresponding points in the knowledge graph. After receiving a patient's recovery state instance, the system iterates through each label. For each standardized state label, the system searches the knowledge graph for related nodes or attribute values ​​of patient state parameters with the same or similar semantics. For example, the label "moderate to severe pain" is associated with a specific entity node in the knowledge graph that describes postoperative pain and its attribute "pain level = moderate". For each tendency label, the system searches the knowledge graph for entity nodes in the dietary tendency whose identifiers match exactly. By establishing such pointers or associations to knowledge graph nodes for each label in the state instance, the state instance is anchored in the vast semantic network of the knowledge graph, and all subsequent reasoning and retrieval by the engine are based on these anchors.

[0100] Furthermore, referring to Figure 3 The specific process of designing and implementing the education strategy matching engine in S4 includes:

[0101] S41. Construct a strategy matching rule base for fusion tendency. Based on the entity relationships and medical logic in the knowledge graph of esophageal cancer postoperative rehabilitation, write condition-action rules to form a strategy matching rule base. The condition part of each rule consists of a logical expression about the standardized state label in the patient state instance, and may optionally include a logical expression about the dietary tendency label. The action part of the rule defines the type of target educational content entity node to be queried from the knowledge graph when the condition is met, the graph traversal path to be followed, and the filtering conditions for the applicable tendency attributes for the content node.

[0102] S42. Implement graph traversal for rule triggering and tendency perception. After receiving the current patient state instance, the education strategy matching engine matches and calculates the condition parts of the standardized state label set and tendency label set in the knowledge graph with all rules in the strategy matching rule base. For rules whose conditions are fully met, the engine activates the rule. When the engine executes the action part of the rule, it takes the node in the knowledge graph corresponding to the state label or tendency label that meets the conditions as the starting point and traverses along the specified semantic relationship edge. During the traversal or after locating the candidate education content node, it checks whether the applicable tendency attribute of the candidate content node is compatible with the tendency label in the patient state instance according to the filtering conditions defined in the action part. Only compatible nodes are retained as valid search results.

[0103] S43. Perform educational content aggregation, tendency adaptation and strategy generation. Collect all effective educational content entity nodes located by the triggered rules. For each effective node, backtrack along the knowledge graph path to extract its context information. The context information includes its directly related health problems, related rehabilitation stages and its applicable tendency attribute values. Organize the extracted educational content and its context information according to the preset template. In the generated personalized rehabilitation education strategy document, clearly list the recommended educational content, the corresponding health problem background, the expected rehabilitation goals, and mark the content that adapts to the patient's specific dietary tendencies.

[0104] S44. Define the strategy output and execution interface. Output the generated personalized rehabilitation education strategy document to the designated application terminal through a standardized data exchange format. The application terminal presents the educational content to patients or medical staff in a set presentation format based on the structured information of the strategy document. It can also selectively display or sort the content according to the applicable preference label.

[0105] In this embodiment, a rule base for matching strategies based on fusion tendencies is constructed: this is the brain-building process of the engine. The compilation of the rule base is a core task of knowledge engineering, requiring collaboration between clinical experts and knowledge engineers. Each rule is independent and clearly defined. The condition part of a rule is a logical expression whose variables are various labels that may appear in the patient state instance. The expression can be simple, such as the postoperative day label == postoperative day 1; it can also contain complex logic, such as the recovery stage label == acute hospitalization AND the symptom label includes fever. The key innovation is that the condition part can optionally embed judgments about dietary tendency labels, such as AND the dietary tendency label includes lactose intolerance tendency. The action part of the rule specifically specifies the retrieval path: it first specifies the type of target educational content entity node to search for; then, it describes which node(s) in the knowledge graph to start from—nodes that are typically associated with the labels mentioned in the condition—and which semantic relationship(s) to traverse the graph along. For example, the action part might be: starting from the 'infection risk' health issue node associated with the 'fever' symptom label, search for all 'educational content' nodes along the 'correspondence' relationship. Simultaneously, filtering criteria are set: only nodes with an 'applicability tendency' attribute value of 'general' or containing 'lactose intolerance tendency' are retained. A large collection of such rules constitutes a strategy matching rule base covering various rehabilitation scenarios.

[0106] The engine implements a real-time thinking and search process for rule triggering and preference perception through graph traversal. When the engine receives a specific patient recovery status instance, it loads all the tags in that instance, including status tags and dietary preference tags, into its working memory. Next, the engine initiates a rule matching loop: it traverses each rule in the rule base, substituting the current instance's tag set into the rule's conditional logic expression for calculation. If the calculation result is true, the rule is activated and placed in the execution queue. For each activated rule in the queue, the engine begins executing its action part. It first locates the starting nodes for traversal in the knowledge graph based on the action description. Then, like navigating a maze, the engine jumps from one node to another along the specified relational edges. For example, starting from the anastomotic fistula health problem node, it finds three educational content nodes A, B, and C along the corresponding relation. At this point, the preference perception mechanism is activated: the engine checks the applicable preference attribute values ​​of these three candidate nodes. Assuming the patient has a low fiber preference, node A's attribute is general, node B's attribute is low fiber preference, and node C's attribute is high fiber requirement. Based on the compatibility logic that may be set in the action section, such as whether the applicable tendency includes patient preference or is general, node C will be filtered out, and only nodes A and B will be retained as valid search results. This process ensures that the retrieved content is not only relevant but also tailored to the patient's specific dietary needs.

[0107] The process of aggregating, adapting, and generating educational content, and then reporting it, is the engine's process of organizing and reporting. The engine may obtain multiple valid educational content nodes from various rules. This step first collects these nodes. Next, to generate a contextualized and easily understandable strategy report, the engine performs a backtracking operation on each valid node: that is, iterates backward from the content node to find the directly connected health problem nodes and higher-level rehabilitation stage nodes, thereby extracting the contextual information of the content—that is, this content aims to solve health problem XX and is suitable for rehabilitation stage YY. Then, all this information—the educational content itself, its context, and its adapted dietary tendencies—is fed into a pre-defined document organization template. This template defines the structure of the strategy document; for example, it first groups by rehabilitation stage, then lists specific educational content by health problem under each stage, and marks each piece of content with a specific icon or text indicating its adapted dietary tendency, such as "[Adapted for lactose intolerance]". Finally, the engine outputs a structured, personalized rehabilitation education strategy document that not only lists what to do but also clearly explains why it addresses the corresponding health problem and who the adapted dietary tendency is specifically optimized for.

[0108] Define the strategy output and execution interface: This is the delivery process for the engine to interact with the external world. The engine encapsulates the generated strategy document into a standardized data exchange format, such as JSON or XML. The schema of this format clearly defines the data fields corresponding to each part of the document, such as group titles, content items, and preference labels. The engine pushes the encapsulated data to the designated application terminal through a well-defined application programming interface (API) or message queue. This terminal can be a patient-side mobile application, a nurse workstation management interface, or a web portal. After receiving the data, the application terminal's internal rendering engine transforms the structured data into visual page elements for display according to a pre-designed user interface style. For example, rehabilitation stage groups are rendered as tabs, educational content items are rendered as graphic cards, and a special colored badge is added to the cards based on the preference label field in the data. The terminal can also sort the cards according to preference labels, placing content that matches the current user's preference at the top or providing filtering functions. Through this interface, the intelligently generated strategy ultimately reaches and serves patients and medical staff.

[0109] Furthermore, the specific categories of dietary preference entity types include:

[0110] Dairy tolerance tendency, cereal preference tendency, dietary fiber acceptance tendency, food texture sensitivity tendency, and specific nutrient supplementation tendency;

[0111] The applicable tendency attribute value of the educational content entity node is used to indicate which dietary tendency category the recommended diet or nutritional advice is compatible with or specifically designed for.

[0112] In this embodiment, the dietary preference entity type is specifically defined as a set of preference categories that have been refined from clinical nutrition and the characteristics of esophageal cancer postoperative rehabilitation, possessing clear distinguishability and management significance. These categories are not arbitrarily listed, but rather designed to address common postoperative nutritional challenges and individual differences:

[0113] Dairy tolerance propensity categories specifically differentiate patients' ability to digest lactose. This is crucial for patients who require high-quality protein and calcium supplementation in the early postoperative period, but who cannot tolerate regular dairy products due to lactase deficiency. This propensity directly influences the choice between regular milk, yogurt, lactose-free formula, or plant-based protein drinks.

[0114] Grain preference categories primarily focus on patients' acceptance and specific requirements regarding staple foods such as wheat and rice. For example, whether there is gluten intolerance requiring a gluten-free diet, or a preference for specific grains like millet or oats due to regional cultural factors. This relates to recommendations for staple food types and recipe development.

[0115] Dietary fiber acceptability propensity category is used to characterize a patient's gut response to dietary fiber. During the early postoperative bowel function recovery phase, some patients may require a low-fiber diet to reduce irritation; while during the long-term recovery period, a high-fiber intake may be encouraged to prevent constipation. This propensity determines the types of fruits and vegetables that should be chosen, and the degree of processing, such as whether to peel or chop them.

[0116] Food texture sensitivity categories address potential swallowing function changes, anastomotic stenosis, or psychological fear that patients may experience after esophageal surgery. It distinguishes whether a patient can only tolerate full liquids, semi-liquids / mucosa, or can gradually transition to soft foods or even a regular diet. This is the core basis for determining cooking recommendations based on the physical form of the food.

[0117] Specific nutrient supplementation preference categories are used to identify patients' prominent needs for certain nutrients due to preoperative malnutrition, postoperative hypermetabolism, or specific complications. These may include high protein requirements, high energy requirements, or the need to restrict sodium intake. This guides the selection of nutritional supplements and the focus of dietary planning.

[0118] Accordingly, in the knowledge graph, each specific educational content entity node, especially those involving dietary advice, will have its applicability preference attribute value assigned one or more specific tags from the aforementioned categories. For example, an educational content item about a sample diet for the first week after surgery might have its applicability preference attribute set to [Food texture sensitivity preference: Full liquid / Semi-liquid]. Another item about how to prevent constipation through diet might have its applicability preference set to [Dietary fiber acceptance preference: High fiber] and [Food texture sensitivity preference: Soft food]. This attribute value acts as a precise filter tag, enabling the strategy matching engine to perform accurate content selection.

[0119] Furthermore, the dietary preference assessment process designed in S32 is specifically implemented by providing a digital assessment interface, which sequentially presents groups of questions related to different basic dietary preference categories;

[0120] The question set design covers inquiries about genetically related symptoms, investigations into past dietary habits, and confirmation of cultural and religious dietary restrictions; based on the patient's answers to the question set, the system automatically applies preset mapping rules to generate a personalized set of dietary preference labels.

[0121] In this embodiment, the interface is designed with user-friendliness and logical clarity in mind. It is typically not a simple questionnaire list, but rather a guided or step-by-step interactive interface. The interface first explains the purpose of the assessment to the user, then unfolds different assessment modules sequentially. Each module focuses on a basic dietary preference category. For example, when entering the dairy tolerance preference module, the interface displays a set of related questions. These questions include not only direct self-reports such as whether you feel unwell after drinking milk, but also potentially more objective inquiries to help identify symptoms, such as whether your discomfort is bloating, diarrhea, or both, and how long after consumption it usually occurs, to improve the accuracy of the assessment. The design logic of the question set is comprehensive: it covers inquiries about genetically related symptoms, such as whether there is a family history of similar intolerances; in-depth investigations into past dietary habits, such as how many times you have consumed dairy products on average per week in the past year, and what type of dairy products you usually choose; and confirmation of cultural and religious dietary restrictions, such as whether you have any animal-derived products that you abstain from due to religious beliefs. Patients answer questions by clicking radio buttons, checking boxes, or entering numbers. All answers are submitted in real time or ultimately submitted to the backend system. The backend system has pre-defined mapping rules, which are decision logics approved by experts. The system automatically matches the patient's answer sequence with these rules. For example, a rule might specify that if "discomfort after drinking milk" is answered "yes," and the "main symptom" includes "diarrhea," and the "onset time" is within "2 hours," then it is mapped to "high probability of lactose intolerance tendency." Based on the answers from all modules and the rule matching results, the system generates a final, structured, individualized dietary tendency label set, which is directly stored and retrieved as part of the patient's feature model.

[0122] Furthermore, the education strategy matching engine integrates a timeline scheduling module, which maintains a standard rehabilitation timeline bound to the rehabilitation stage entity;

[0123] When the time-series parameters in the patient's recovery status model indicate that the patient has entered a certain preset time node, the timeline scheduling module automatically generates a baseline rule that is strongly correlated with the time node.

[0124] When performing matching, the education strategy matching engine applies such benchmark rules to obtain a set of universal educational content candidates for that time point. Based on the dietary preference tags in the patient status instance, it filters and adapts this set of candidates, integrates the adapted content with personalized content triggered by other statuses, and together they form the final education strategy.

[0125] In this embodiment, the timeline scheduling module integrated into the education strategy matching engine has the core function of ensuring the continuity and basic coverage of rehabilitation education. This module maintains a standard rehabilitation timeline, which is essentially a time axis connecting the start / end times of various rehabilitation stage entities and their attributes defined in the domain ontology. Based on this, it marks more granular key time nodes within each stage, such as postoperative day 1, the day of chest tube removal, one day before discharge, and the first month postoperative check-up. Each time node is pre-associated with one or more basic, universal health education topics, such as the importance of bed activities and breathing exercises on postoperative day 1. When the time-series parameter in the patient's rehabilitation status model, most typically the postoperative days, is updated, indicating that the patient has entered or reached a certain preset time node, the timeline scheduling module responds immediately. It does not directly manipulate the knowledge graph but automatically generates one or more corresponding baseline rules based on its internal time-topic mapping table. The condition parts of these rules are very simple, usually only bound to the time node identifier, such as IF current time node == postoperative day 3. The action component instructs the engine to retrieve basic educational content strongly related to the topic of that node from the knowledge graph. During matching, these time-triggered baseline rules are evaluated and activated along with other rules triggered by symptoms, behaviors, etc. After obtaining the candidate set of universal educational content retrieved from these baseline rules, the engine does not directly add it to the final strategy. Instead, it first performs a preference-adaptation process: using the patient's current dietary preference tags to filter the candidate set, eliminating content whose applicable preference attributes clearly conflict with the patient's preferences. For example, the universal content triggered by dietary guidance in the first week post-surgery might include yogurt, but if the patient's tag indicates lactose intolerance, this content will be filtered or replaced with a more suitable version. Finally, the timeline content, after preference filtering and adaptation, is merged, deduplicated, and logically integrated with more personalized educational content triggered by other states, such as specific guidance for the patient's current pain symptoms, to form a final educational strategy that both follows standard rehabilitation procedures and deeply integrates the patient's individual real-time status and stable dietary preferences.

[0126] Furthermore, the policy matching rule base built in S4 follows the principles of hierarchical and preference-specific adaptation in its rule design, specifically including:

[0127] The design of the first-level general rules is based solely on the core physiological state labels in the patient's rehabilitation state model and does not include restrictions on dietary preferences. The educational content nodes retrieved in the action part typically do not have specific applicable preference attributes or have general preference attributes, and are used to provide universal rehabilitation guidance.

[0128] The design includes a two-level preference screening rule, where the condition part includes the determination of one or more dietary preference tags; and the action part, when retrieving educational content nodes, requires that the applicable preference attribute value of the node must match the preference tag determined in the condition.

[0129] Design a three-level composite trigger rule, the condition part of which is a complex logical expression, and simultaneously associates a specific combination of physiological state labels with a specific combination of dietary tendency labels.

[0130] When performing matching, the education strategy matching engine processes first-level, second-level, and third-level rules in parallel; for triggered second-level and third-level rules, the engine strictly enforces content filtering based on preference tags.

[0131] The generated educational strategy integrates all effective educational content that matches the triggering rules and passes the preference filtering, sorting and organizing it according to rule level and content priority during integration.

[0132] In this embodiment, the constructed strategy matching rule base follows a hierarchical and preference-specific adaptation principle in its rule design, representing a systematic rule organization method to improve the accuracy and efficiency of strategy matching. This method divides rules into three logical levels based on the complexity and specificity of their conditions, and clarifies the interaction between each level and dietary preferences.

[0133] First-level general rules form the foundational layer of the strategy library. Their conditional components rely solely on the patient's most core and common physiological state labels, such as the recovery phase label "acute hospitalization" or the key symptom label "fever." These rules do not include any judgment regarding dietary preferences. The action components retrieve educational content nodes, typically general, broadly applicable guidance on the topic, with their applicability attribute often set to general or empty. For example, a first-level rule might trigger a general educational article on the importance of preventing lung infections during acute hospitalization.

[0134] The secondary preference screening rules are the main layer for implementing individualized dietary adaptation. Its condition section, in addition to including certain physiological state labels, explicitly adds a judgment on one or more dietary preference labels, such as AND dietary preference label includes 'low fiber preference'. Its action section, during retrieval, strictly requires that the applicable preference attribute value of the found educational content nodes must include the preference label judged in the condition or be general. For example, when a patient is in a dietary transition phase and has a low fiber preference, a secondary rule will specifically search for content tagged with the applicable preference: low fiber, specifically guidance on how to choose low-fiber fruits and vegetables, rather than general dietary transition advice.

[0135] The three-tiered composite triggering rules, used to handle complex, multi-factor intertwined clinical scenarios, reside at the top of the strategy library. Their conditional part consists of complex Boolean logic expressions, simultaneously associating specific physiological state labels (e.g., anastomotic stenosis symptoms AND insufficient nutrient intake) with specific dietary preference labels (e.g., food texture sensitivity preference, liquid-only diet AND high-protein requirement preference). The graph traversal path defined in the action part is also more complex, potentially involving multiple nodes, multiple relationship jumps, and ultimately locating educational content nodes that can provide comprehensive, integrated solutions, such as a complete nutritional liquid diet and feeding plan specifically designed for patients with anastomotic stenosis requiring a high-protein liquid diet.

[0136] The education strategy matching engine does not simply execute matching sequentially, but can process rules at all three levels in parallel. It first performs rapid matching calculations on all rules, regardless of level. For triggered second- and third-level rules, the engine strictly enforces the bias-label-based filtering instructions in the subsequent content retrieval stage to ensure strong content relevance. Finally, the engine collects all effective educational content generated by triggering rules, including first-, second-, and third-level rules, that has passed the filtering requirements of its respective level. The final step before generating the final strategy is to integrate all this effective content. Integration is not a simple accumulation, but rather based on a set of meta-rules: content matched by higher-level third-level rules typically has higher contextual relevance and priority; simultaneously, the content itself may contain metadata such as urgency and recommendation order. Based on the rule level and this content metadata, the engine sorts and organizes the content in the final output education strategy document, forming a well-structured, logically coherent, and personalized rehabilitation education plan.

[0137] Furthermore, the method also includes S6, providing visualization of tendency associations and strategy editing tools:

[0138] Provides a graphical user interface that can display the interconnected subgraphs between dietary tendency entity nodes, health problem nodes, and educational content nodes in the knowledge graph of esophageal cancer postoperative rehabilitation.

[0139] The graphical user interface provides dedicated controls that allow users to: a) graphically bind or unbind one or more dietary preference nodes to educational content nodes in the knowledge graph, thereby modifying their applicable preference attributes; b) intuitively add or modify dietary preference-related conditional logic for rules in the strategy matching rule base by connecting dietary preference nodes with conditional expressions in the rule editing panel.

[0140] The tool translates the user's graphical operations into commands to modify the attribute values ​​of the knowledge graph or to update the rule base, and saves them synchronously.

[0141] The process of generating personalized rehabilitation education strategies supports the invocation of strategy templates based on propensity tags.

[0142] Multiple strategy templates are predefined for typical rehabilitation scenarios and specific dietary preference combinations; each strategy template is associated with a specific patient physiological state pattern and dietary preference tags, and encapsulates a knowledge graph query path and content organization logic containing preference filtering conditions.

[0143] When the education strategy matching engine identifies that the current patient status instance matches the physiological status pattern associated with a predefined strategy template, and its dietary tendency label set completely matches or supersets the tendency labels associated with the template, the strategy template is invoked first.

[0144] When a strategy template is invoked, the engine instantiates and populates the template based on the specific parameter values ​​of the current state instance, and directly executes the query logic encapsulated within the template to generate a strategy draft.

[0145] If no perfectly matching policy template is found, the engine will instead execute a general rule matching process based on the policy matching rule base to generate a policy.

[0146] In this embodiment, the generation process of personalized rehabilitation education strategies also supports a more efficient strategy template invocation mechanism. This mechanism requires experts to construct a series of strategy templates in advance. Each template has a unique name and is explicitly associated with a typical rehabilitation scenario, such as postoperative discharge instructions, and a set of specific dietary preference tags, such as lactose intolerance and food texture sensitivity (semi-liquid diet). The template encapsulates a fixed, pre-optimized knowledge graph query path and content organization logic, which includes built-in filtering conditions for associated dietary preferences. When the education strategy matching engine is running, it quickly compares the current patient state instance with all defined strategy templates. If the current patient's physiological state pattern highly matches the scenario associated with a certain template, and the current patient's dietary preference tag set completely includes all preference tags associated with that template (i.e., superset matching), the engine will prioritize invoking that template. The invocation process is as follows: the engine fills the template's variable placeholders with specific parameter values ​​from the current patient state instance, such as the patient's name and the specific number of days post-surgery, and then directly executes the validated, efficient query logic encapsulated within the template, thereby quickly generating a strategy draft. If the patient's predisposition labels do not perfectly match the template, or if there is no matching scenario template, the engine reverts to the conventional process of generating a strategy by matching each rule based on a general rule base. This template-first mechanism greatly improves the efficiency of strategy generation for common and typical situations.

[0147] According to a second aspect of the present invention, the present invention claims protection for a knowledge graph-based system for constructing postoperative rehabilitation education strategies for esophageal cancer, comprising:

[0148] One or more processors;

[0149] A memory that stores one or more programs, which, when executed by one or more processors, enable one or more processors to implement a knowledge graph-based method for constructing postoperative rehabilitation education strategies for esophageal cancer.

[0150] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between apparatuses or units, and may be electrical, mechanical, or other forms.

[0151] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units described above can be implemented in hardware or as software functional units. The above are merely embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made based on the description and drawings of this application, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

[0152] The specific embodiments of the invention have been described in detail above, but they are only examples, and this application is not limited to the specific embodiments described above. For those skilled in the art, any equivalent modifications or substitutions to the invention are also within the scope of this application. Therefore, all equivalent changes, modifications, and improvements made without departing from the spirit and principles of this application should be covered within the scope of this application.

Claims

1. A method for constructing a postoperative rehabilitation education strategy for esophageal cancer based on knowledge graphs, characterized in that, Includes the following steps: S1. Construct an ontology model for the domain of health education for postoperative rehabilitation of esophageal cancer, and determine the core entity types, attributes of each entity type, and semantic relationships between the core entity types in the ontology model. S2. Based on the domain ontology model, construct a knowledge graph for postoperative rehabilitation of esophageal cancer, acquire and structure authoritative medical knowledge in the field of postoperative rehabilitation of esophageal cancer, generate knowledge triples with the core entity type as nodes and the semantic relationship as edges according to the entity type and semantic relationship defined by the domain ontology model, and store the knowledge triples in the graph database to form the knowledge graph for postoperative rehabilitation of esophageal cancer. S3. Define a patient rehabilitation status model. Based on the patient status parameter entity type and dietary tendency entity type in the domain ontology model, construct a multi-dimensional status description framework. The framework determines the patient's individualized dietary tendency label through a preset evaluation process and maps the label and other rehabilitation information of the individual patient together into identifiable standardized status parameter instances and dietary tendency instances in the knowledge graph. S4. Design and implement an education strategy matching engine. Based on the knowledge graph of postoperative rehabilitation of esophageal cancer, construct a strategy matching rule base. The strategy matching rule base contains logical rules for associating specific patient status, specific dietary preferences and appropriate education content. S5, when a specific patient state instance containing dietary preference tags is input as defined by the patient rehabilitation state model, the education strategy matching engine traverses the strategy matching rule base, activates the rules that match the current patient state instance and dietary preference tags, retrieves the educational content entity nodes that are associated with and match the dietary preference tags from the knowledge graph, and integrates the retrieved nodes and their related paths to generate a personalized rehabilitation education strategy that incorporates dietary adaptability. The specific process of designing and implementing the education strategy matching engine in S4 includes: S41. Construct a strategy matching rule base for fusion tendency: Based on the entity relationships and medical logic in the knowledge graph of esophageal cancer postoperative rehabilitation, write condition-action rules to form the strategy matching rule base. The condition part of each rule consists of a logical expression about the standardized state label in the patient state instance, and may selectively include a logical expression about the dietary tendency label. The action part of the rule defines the type of target educational content entity node to be queried from the knowledge graph when the condition is met, the graph traversal path to be followed, and the filtering conditions for the applicable tendency attributes of the content node. S42. Implement graph traversal for rule triggering and tendency perception. After receiving the current patient state instance, the education strategy matching engine performs matching calculations on the standardized state label set and tendency label set in the knowledge graph with the condition parts of all rules in the strategy matching rule base. For rules whose conditions are fully met, the engine activates the rule. When the engine executes the action part of the rule, it takes the node in the knowledge graph corresponding to the state label or tendency label that meets the conditions as the starting point and traverses along the specified semantic relationship edge. During the traversal or after locating the candidate education content node, it checks whether the applicable tendency attribute of the candidate content node is compatible with the tendency label in the patient state instance according to the filtering conditions defined in the action part, and only retains the compatible nodes as valid search results. S43. Perform educational content aggregation, tendency adaptation, and strategy generation, and collect all valid educational content entity nodes located by the triggered rules; for each valid node, backtrack along the knowledge graph path to extract its context information, which includes its directly associated health problems, associated rehabilitation stages, and applicable tendency attribute values. Organize the extracted educational content and its context information according to a preset template. In the generated personalized rehabilitation education strategy document, clearly list the recommended educational content, the corresponding health problem background, the expected rehabilitation goals, and annotate the content that adapts to the patient's specific dietary tendencies. S44. Define the strategy output and execution interface, and output the generated personalized rehabilitation education strategy document to the designated application terminal through a standardized data exchange format; the application terminal presents the educational content to the patient or medical staff in a set presentation format based on the structured information of the strategy document, and can selectively display or sort the content according to the applicable preference label.

2. The method according to claim 1, characterized in that, In S1, the specific process of constructing the domain ontology model includes: Define the entity type for the rehabilitation phase, with attributes including at least the phase identifier, phase name, start time, and end time. Define a health problem entity type, whose attributes must include at least a problem identifier, a problem description, and the recovery stage to which it belongs; Define an educational content entity type whose attributes include at least a content identifier, a content title, a content carrier format, and a detailed description text; this entity type has an applicability preference attribute, which is used to associate one or more dietary preference entity instances. Define a patient status parameter entity type, whose attributes include at least parameter identifier, parameter type, and parameter value range. The parameter type includes time-series parameters, symptom-based parameters, and behavioral compliance parameters. Define dietary preference entity types, whose attributes include at least preference identifiers and preference descriptions. These entity types are used to characterize food preferences or taboo patterns that are determined by genetic factors, environmental and cultural factors, or disease contraindications and need to be differentiated in postoperative nutritional support. Define the semantic relationships between the core entity types, which include at least the following: the rehabilitation stage includes health problems, health problems correspond to educational content, patient status parameters indicate the rehabilitation stage, patient status parameters trigger attention to specific health problems, and dietary tendencies affect the selection of specific educational content.

3. The method according to claim 1, characterized in that, The specific process of defining the patient rehabilitation status model in S3 includes: S31. Determine the set of composite state parameters: Select key physiological and behavioral parameters for strategy matching from the patient state parameter entity types of the domain ontology model, and select the tendency dimensions to be evaluated from the dietary tendency entity types, together forming the set of composite state parameters. S32. Design a dietary preference assessment process: For the dietary preference dimension in the composite state parameter set, design a structured assessment process, define mutually exclusive basic dietary preference categories, design corresponding assessment questions or observations for each category, collect patient feedback or observation data through an interactive interface, and classify the data into one or more specific basic dietary preference categories according to preset mapping rules to generate a personalized dietary preference label set for the patient. S33. Constructing a state instantiation and tendency fusion process: Establish a data processing interface to receive individual patient physiological data from an external system and dietary tendency data obtained through the assessment process; the data processing interface extracts and formats the individual patient physiological data according to the requirements of the composite state parameter set, and converts the individual dietary tendency tag set into a data format that conforms to the knowledge graph dietary tendency entity instance specification; the formatted physiological data and dietary tendency data are merged to form a complete original state dataset; S34. Apply the state assessment and mapping rules to define the assessment rules for mapping the physiological parameters or parameter combinations in the composite state parameter set to standardized state labels. Transform the physiological data part in the original state dataset into a set composed of several standardized state labels, and directly use the dietary tendency data part as the tendency label set. The standardized state label set and the tendency label set together constitute the current patient state instance. S35. Implement knowledge graph mapping for composite state instances, and associate and bind each standardized state label and each tendency label in the current patient state instance with the corresponding patient state parameter related nodes and dietary tendency entity nodes in the esophageal cancer postoperative rehabilitation knowledge graph, so that the composite state instances output by the patient rehabilitation state model can be semantically understood and processed by the knowledge graph.

4. The method according to claim 2, characterized in that, The specific preference categories defined for the dietary preference entity type include: Dairy tolerance tendency, cereal preference tendency, dietary fiber acceptance tendency, food texture sensitivity tendency, and specific nutrient supplementation tendency; The applicable tendency attribute value of the educational content entity node is used to indicate which dietary tendency category the recommended diet or nutritional advice is compatible with or is specifically designed for.

5. The method according to claim 3, characterized in that, The dietary preference assessment process designed in S32 is specifically implemented by providing a digital assessment interface, which sequentially presents groups of questions related to different basic dietary preference categories. The design of the question set covers inquiries about genetically related symptoms, investigations into past dietary habits, and confirmation of cultural and religious dietary restrictions; based on the patient's answers to the question set, the system automatically applies the preset mapping rules to generate the personalized dietary tendency label set.

6. The method according to claim 1, characterized in that, The education strategy matching engine integrates a timeline scheduling module, which maintains a standard rehabilitation timeline bound to the rehabilitation stage entity. When the time-series parameters in the patient recovery status model indicate that the patient has entered a certain preset time node, the timeline scheduling module automatically generates a benchmark rule that is strongly correlated with the time node. When performing matching, the education strategy matching engine applies such benchmark rules to obtain a set of universal educational content candidates for that time point. Based on the dietary preference tags in the patient status instance, it filters and adapts this set of candidates, integrates the adapted content with personalized content triggered by other statuses, and together they constitute the final education strategy.

7. The method according to claim 1, characterized in that, The policy matching rule base constructed in S4 follows the principles of hierarchical and preference-specific adaptation in its rule design, specifically including: The design of the first-level general rules is based solely on the core physiological state labels in the patient rehabilitation state model and does not include restrictions on dietary preferences. The educational content nodes retrieved in the action part typically do not have specific applicable preference attributes or have general preference attributes, and are used to provide universal rehabilitation guidance. The design includes a two-level preference screening rule, where the condition part includes the determination of one or more dietary preference tags; and the action part requires that when retrieving educational content nodes, the applicable preference attribute value of the node must match the preference tag determined in the condition. Design a three-level composite trigger rule, the condition part of which is a complex logical expression, and simultaneously associates a specific combination of physiological state labels with a specific combination of dietary tendency labels. When performing matching, the education strategy matching engine processes the first-level, second-level, and third-level rules in parallel; for the triggered second-level and third-level rules, the engine strictly performs content filtering based on preference tags. The generated educational strategy integrates all effective educational content that matches the triggering rules and passes the preference filtering, sorting and organizing it according to rule level and content priority during integration.

8. The method according to claim 1, characterized in that, The method also includes S6, providing visual tendency association and strategy editing tools: A graphical user interface is provided, which can display the associated subgraphs between the dietary tendency entity nodes, health problem nodes, and educational content nodes in the knowledge graph of esophageal cancer postoperative rehabilitation. The graphical user interface provides dedicated controls that allow users to: a) graphically bind or unbind one or more dietary preference nodes to educational content nodes in the knowledge graph, thereby modifying their applicable preference attributes; b) intuitively add or modify dietary preference-related conditional logic for rules in the strategy matching rule base by connecting dietary preference nodes with conditional expressions in the rule editing panel. The tool translates the user's graphical operations into commands to modify the attribute values ​​of the knowledge graph or to update the logical rules base, and saves them synchronously. The generation process of the personalized rehabilitation education strategy supports the invocation of strategy templates based on propensity tags: Multiple strategy templates are predefined for typical rehabilitation scenarios and specific dietary preference combinations; each strategy template is associated with a specific patient physiological state pattern and dietary preference label, and encapsulates a knowledge graph query path and content organization logic containing preference filtering conditions; When the education strategy matching engine identifies that the current patient state instance matches the physiological state pattern associated with a predefined strategy template, and its dietary tendency label set completely matches or supersets the tendency labels associated with the template, the strategy template is invoked first. When a strategy template is invoked, the engine instantiates and populates the template based on the specific parameter values ​​of the current state instance, and directly executes the query logic encapsulated within the template to generate a strategy draft. If no perfectly matching strategy template is found, the engine will instead execute a general rule matching process based on the strategy matching rule base to generate a strategy.

9. A knowledge graph-based system for constructing postoperative rehabilitation education strategies for esophageal cancer, characterized in that, include: One or more processors; A memory having stored one or more programs, which, when executed by one or more processors, cause the one or more processors to implement a method for constructing a knowledge graph-based postoperative rehabilitation education strategy for esophageal cancer, as described in any one of claims 1 to 8.

Images