Dementia intervention scheme recommendation method and system based on knowledge graph

Abstract

The invention discloses a dementia intervention scheme recommendation method and system based on a knowledge graph, and relates to the technical field of intelligent medical treatment and artificial intelligence. And constructing a dementia intervention knowledge graph containing patients, symptoms, intervention measures and environmental elements. And performing intelligent semantic reasoning on the short-term state, the long-term behavior mode and the nursing environment of the patient by fusing multi-level reasoning and situational reasoning, generating a candidate intervention scheme matched with the current state of the patient, and outputting a personalized recommendation result. In the intervention implementation process, the knowledge graph association relation is dynamically updated based on patient behavior feedback and symptom changes, and continuous optimization and self-adaptive adjustment of an intervention scheme are achieved. According to the method, the individuation level, long-term adaptability and safety of dementia intervention recommendation are improved, and the method is suitable for clinical and long-term nursing scenes.

Description

Technical Field

[0001] This specification relates to the fields of smart healthcare and artificial intelligence technology. More specifically, this application relates to a method and system for recommending intervention programs for dementia based on knowledge graphs. Background Technology

[0002] Dementia is a chronic neurodegenerative disease characterized by progressive cognitive decline, often accompanied by behavioral and psychiatric symptoms and reduced daily living abilities. Intervention for dementia is a long-term, phased process with significant individual differences. Current dementia intervention programs largely rely on the experience and judgment of medical staff and caregivers, making it difficult to continuously and accurately adjust them dynamically according to changes in the patient's condition. While some decision support systems incorporate rule engines or machine learning methods, they are typically based on static features or fixed models, failing to systematically characterize the complex relationships between patient characteristics, symptoms, interventions, and environmental factors, and lacking effective modeling of long-term feedback and risk changes.

[0003] Knowledge graph technology provides a structured means for expressing medical knowledge with multiple entities and relationships. However, existing medical recommendation methods based on knowledge graphs mostly use the graph as a static knowledge base, with a single reasoning method, which is difficult to adapt to the actual needs of strong context dependence and continuous evolution in the process of dementia intervention.

[0004] Therefore, there is an urgent need for a dementia intervention recommendation method that can integrate multidimensional patient characteristics, contextual information, and behavioral feedback, and support dynamic reasoning and adaptive evolution, in order to improve the personalization level and long-term effectiveness of intervention programs. Summary of the Invention

[0005] The summary section introduces a series of simplified concepts, which will be further explained in detail in the detailed description section. This summary section is not intended to limit the key and essential technical features of the claimed technical solution, nor is it intended to determine the scope of protection of the claimed technical solution.

[0006] Firstly, this application proposes a knowledge graph-based method for recommending dementia intervention programs, including: Obtain individual characteristic data of dementia patients, including cognitive function assessment results, behavioral and psychiatric symptoms, living ability status, and interest and preference information; Based on the aforementioned individual characteristic data, a knowledge graph for dementia intervention is constructed and updated. This knowledge graph includes patient nodes, symptom nodes, intervention measure nodes, environmental element nodes, and their interrelationships. Based on the aforementioned knowledge graph, intelligent semantic reasoning is performed on the patient's current state to determine a set of candidate intervention plans that match the patient's current state. The aforementioned intelligent semantic reasoning is generated based on multi-level reasoning and contextual reasoning. Personalized intervention recommendations for the aforementioned dementia patients were generated from the set of candidate intervention options.

[0007] In one feasible implementation, the above-mentioned construction and updating of the dementia intervention knowledge graph based on the aforementioned individual characteristic data includes: The cognitive ability, behavioral symptoms, daily living skills, and interest preferences of the patients were extracted from the individual characteristic data mentioned above, and these data were mapped to patient nodes and symptom nodes in the dementia intervention knowledge graph mentioned above. Based on the characteristics of the patient nodes and symptom nodes mentioned above, and using the reasoning mechanism of the knowledge graph, the relationship between the intervention measure nodes and the environmental element nodes is automatically established, and the node weights are dynamically adjusted to construct a knowledge graph for dementia intervention. Based on the aforementioned individual characteristic data and patient behavioral feedback, the node relationships in the dementia intervention knowledge graph are updated regularly.

[0008] In one feasible implementation, based on the characteristics of the patient nodes and symptom nodes, and using a knowledge graph reasoning mechanism, the relationship between intervention measure nodes and environmental element nodes is automatically established, and the node weights are dynamically adjusted to construct a dementia intervention knowledge graph, including: Based on the cognitive ability level, behavioral and mental symptom type, and living ability status corresponding to the above patient nodes, a set of candidate intervention measures nodes associated with the above patient nodes is determined. Based on the historical intervention effect correlation between the above symptom nodes and the above intervention measure nodes, calculate the adaptation weight of the above intervention measure nodes for different symptom types; Based on the care environment constraints represented by the above environmental element nodes, the feasibility of the above intervention measures nodes is screened. Based on the above adaptation weights and the above feasibility screening results, the above-mentioned intervention measure nodes and the above-mentioned environmental element nodes are established, and the above-mentioned adaptation weights are written into the above-mentioned association relationship to form a weighted relationship structure that can be used for subsequent semantic reasoning.

[0009] In one feasible implementation, the node relationships in the dementia intervention knowledge graph are periodically updated based on the aforementioned individual characteristic data and patient behavioral feedback, including: During the implementation of the intervention program, behavioral response data, symptom change data, and adverse event marker data of the patients were collected during the execution of the intervention measures. Based on the above behavioral response data and the above symptom change data, the correlation between the above intervention measure nodes and the above symptom nodes is evaluated to generate the corresponding intervention effect evaluation value. Based on the above intervention effect evaluation values, the correlation weights between the above intervention measure nodes and the above symptom nodes are adjusted; The adjusted association weights are written back into the dementia intervention knowledge graph to update the reasoning basis of the knowledge graph, so that the subsequently generated set of candidate intervention programs can reflect the dynamic changes in the patient's condition.

[0010] In one feasible implementation, the above-mentioned semantic reasoning based on the knowledge graph to determine a set of candidate intervention schemes matching the patient's current state includes: By conducting multi-level inference on short-term characteristics such as patients’ cognitive function, behavioral symptoms and living abilities, multi-level inference results can be obtained. By combining patients' long-term behavioral patterns, historical feedback data, and other situational reasoning, the system analyzes real-time changes in patients' emotional state, environmental factors, and social interactions to generate situational reasoning results. Based on the above multi-level reasoning results and the above scenario reasoning results, the matching degree between the patient's current state and multiple intervention measure nodes is calculated to form a set of candidate intervention plans.

[0011] In one feasible implementation, the above-mentioned multi-level reasoning is performed on short-term characteristics such as the patient's cognitive function, behavioral symptoms, and daily living abilities to obtain multi-level reasoning results, including: Based on the cognitive function assessment results of the above patients, a first-level reasoning feature representing the patient's current cognitive level is generated; Based on the above-mentioned behavioral and psychological symptom characteristics of patients, a second-level inference feature is generated to characterize the patient's behavioral risk and symptom intensity. Based on the patients' living ability status, a third-level inference feature representing the degree of functional limitation of the patients is generated. The first-level, second-level, and third-level inference features are hierarchically fused to form the multi-level inference results used to characterize the patient's short-term comprehensive state. In one feasible implementation, the above-mentioned analysis of real-time changes in the patient's emotional state, environmental factors, and social interactions, combined with the patient's long-term behavioral patterns, historical feedback data, and other contextual reasoning, generates contextual reasoning results, including: Based on the historical behavioral feedback data of the patients mentioned above, the behavioral response patterns of the patients under different intervention scenarios were extracted to form long-term behavioral pattern characteristics. Based on the above environmental element nodes, the constraints and stimulating factors of the patient's current care environment are obtained to form environmental context characteristics. Based on the changes in patients' emotional state and social interactions, emotional and social characteristics that characterize patients' psychological and social adaptation status are generated. By jointly modeling the aforementioned long-term behavioral pattern characteristics, environmental context characteristics, and emotional and social characteristics, the aforementioned situational reasoning results are generated to characterize the patient's current external and psychological situation.

[0012] In one feasible implementation, based on the above-mentioned multi-level reasoning results and the above-mentioned scenario reasoning results, the matching degree between the patient's current state and multiple intervention measure nodes is calculated to form a set of candidate intervention plans, including: Based on the above multi-level reasoning results, the ability matching degree between the patient's short-term ability status and the functional requirements of the above intervention measures nodes is calculated. Based on the above scenario reasoning results, the scenario fit between the patient's current situation and the implementation conditions of the above intervention measures nodes is calculated; When there is a conflict between the above ability matching degree and the above situational adaptability degree, the above ability matching degree is adjusted punitively based on the preset conflict suppression rules; When the above capability matching degree is consistent with the above situational adaptability degree, the two are synergistically enhanced to generate the above comprehensive matching degree. Based on the comprehensive matching degree, the nodes of the above-mentioned intervention measures are screened and sorted to form the above-mentioned set of candidate intervention schemes.

[0013] In one feasible implementation, the node relationships in the dementia intervention knowledge graph are periodically updated based on the aforementioned individual characteristic data and patient behavioral feedback, including: During the intervention program implementation period, time-series analysis was conducted on the patient's behavioral response trends, symptom evolution trajectory, and frequency of adverse events during the implementation of the above intervention measures to generate feedback evolution characteristics that characterize the stability of the intervention and changes in risk. Based on the above feedback evolution characteristics, the correlation between the above intervention measure nodes and the above symptom nodes is classified into three types: stable enhancement type, fluctuation adaptation type and risk suppression type. For the aforementioned stable and enhanced associations, the association weights are incrementally updated to strengthen their priority in subsequent semantic reasoning. To address the aforementioned fluctuating adaptive correlations, a time decay or contextual condition constraint mechanism is introduced to conditionally adjust the correlation weights. For the aforementioned risk-suppressing associations, their association weights are reduced or frozen, and they are marked as risk-constrained associations in the aforementioned dementia intervention knowledge graph; The above-described classification and adjusted relationships are written back into the above-described dementia intervention knowledge graph, so that the knowledge graph forms a dynamic reasoning structure with risk perception ability and long-term self-evolution characteristics.

[0014] Secondly, this application proposes a knowledge graph-based dementia intervention program recommendation system, including: The acquisition unit is used to acquire individual characteristic data of dementia patients, including cognitive function assessment results, behavioral and mental symptom characteristics, living ability status, and interest and preference information. The construction and updating unit is used to construct and update the dementia intervention knowledge graph based on the above-mentioned individual characteristic data. The knowledge graph includes patient nodes, symptom nodes, intervention measure nodes, environmental element nodes and their interrelationships. The determination unit is used to perform intelligent semantic reasoning on the patient's current state based on the above knowledge graph, and to determine a set of candidate intervention plans that match the patient's current state. The intelligent semantic reasoning is generated based on multi-level reasoning and contextual reasoning. The generation unit is used to generate personalized intervention recommendations for the aforementioned dementia patients from the set of candidate intervention options.

[0015] In summary, compared to existing methods that primarily rely on scale thresholds or human experience for intervention selection, this invention introduces a dementia intervention knowledge graph containing patient nodes, symptom nodes, intervention measure nodes, and environmental element nodes. This model unifies the complex relationships between patient ability status, symptom presentation, intervention methods, and care environment in a structured semantic form, enabling intervention decisions to move beyond single indicators or static rules and instead allow for combined reasoning across multiple dimensions. This significantly improves the precision and rationality of intervention matching. Furthermore, this invention incorporates an intelligent semantic reasoning mechanism combining multi-level reasoning and contextual reasoning in the recommendation process, simultaneously constraining and filtering intervention plans from both the patient's short-term ability status and the surrounding context. Compared to existing technologies that recommend interventions solely based on patient ability or symptoms, this invention effectively avoids the problems of "suitable ability but unfeasible context" or "allowable context but mismatched ability," making the recommendations more operable and safer in practice, and reducing the risk of agitation, rejection, or adverse events. This invention continuously collects behavioral feedback, adverse event markers, and symptom change information during intervention implementation and writes this information back into a knowledge graph to dynamically update node relationships, enabling the knowledge graph to self-correct as the patient's state evolves. Compared to traditional static recommendation models, this invention can progressively optimize the priority and adaptation rules of intervention programs during long-term care, improving the stability and sustainability of intervention effects and preventing recommendation failure due to changes in patient state. This invention introduces a screening strategy prioritizing interest preferences and participation when generating personalized intervention recommendations, making the recommended programs not only reasonable from a medical and care perspective but also more advantageous in terms of patient subjective acceptance and participation enthusiasm, thus helping to improve adherence to non-pharmacological interventions and long-term implementation effects. Through a knowledge graph-driven dynamic semantic reasoning and feedback closed-loop mechanism, this invention achieves adaptive recommendations for dementia intervention programs in multiple scenarios and stages, achieving significant improvements in personalization level, implementation feasibility, safety, and long-term evolutionary capacity compared to background technologies, demonstrating good practical application value.

[0016] The knowledge graph-based dementia intervention recommendation method proposed in this application, along with other advantages, objectives, and features of this application, will be partly apparent from the following description and partly understood by those skilled in the art through research and practice of this application. Attached Figure Description

[0017] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit this specification. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1A flowchart illustrating a knowledge graph-based method for recommending intervention programs for dementia, provided in an embodiment of this application; Figure 2 This is a structural diagram of a knowledge graph-based dementia intervention program recommendation system provided in an embodiment of this application. Detailed Implementation

[0018] The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus. The technical solutions of the embodiments of this application will now be clearly and completely described in conjunction with the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them.

[0019] Please see Figure 1 This is a flowchart illustrating a knowledge graph-based method for recommending intervention programs for dementia, as provided in an embodiment of this application. Specifically, it may include: S110. Obtain individual characteristic data of dementia patients, wherein the aforementioned individual characteristic data includes cognitive function assessment results, behavioral and mental symptom characteristics, living ability status, and interest and preference information; S120. Construct and update a knowledge graph for dementia intervention based on the above individual characteristic data, wherein the knowledge graph includes patient nodes, symptom nodes, intervention measure nodes, environmental element nodes and their interrelationships. S130. Based on the above knowledge graph, perform intelligent semantic reasoning on the patient's current state to determine a set of candidate intervention schemes that match the patient's current state. The above intelligent semantic reasoning is generated based on multi-level reasoning and contextual reasoning. S140. Generate personalized intervention recommendations for the above-mentioned dementia patients from the above set of candidate intervention options.

[0020] For example, in step S110, individual characteristic data of dementia patients are first acquired. This individual characteristic data is not only used to characterize the patient's basic state but also serves as a unified input for subsequent knowledge graph modeling and reasoning decision-making. Specifically, the cognitive function assessment results are used to characterize the patient's current cognitive level and ability retention, and can be quantified using scales such as the MMSE. The behavioral and psychiatric symptom characteristics are used to characterize whether the patient exhibits BPSD-related manifestations such as agitation, hallucinations, and nocturnal behavioral disturbances, and their severity. The living abilities status is used to characterize the patient's independence or dependence in daily activities such as eating, dressing, toileting, and walking. The interest and preference information is used to characterize the patient's lifelong roles, habitual activities, and stimulating cues for active participation. This interest and preference information can be obtained from caregiver interviews, past activity records, or observations of the patient's daily behavior. Through the unified collection of these multidimensional features, the patient's state can be both clinically quantified and matched to activities, thus providing a calculable and traceable basis for subsequent personalized interventions.

[0021] In step S120, a dementia intervention knowledge graph is constructed and updated based on the aforementioned individual characteristic data. This knowledge graph is used to express and associate four core elements—patient, symptoms, intervention, and environment—in a structured form. Specifically, the patient node carries information such as patient identity, stage and classification, and ability profile. The symptom node expresses the type of cognitive deficit, BPSD category, functional limitations, and their changing trends. The intervention node expresses available non-pharmacological intervention activities, communication prompting strategies, task breakdown methods, difficulty adjustment rules, and risk avoidance points. The environmental element node expresses constraints such as differences between hospital wards, home, and community settings, caregiver capabilities and available time, the degree of spatial layout modification, and accessible items and potential stimuli. By establishing relationships between nodes, the knowledge graph can explicitly represent knowledge such as "a certain type of symptom is more suitable for a certain type of activity in a certain environment," "a certain type of activity requires a certain level of ability," and "the correlation between a certain type of environmental stimulus and a certain type of behavioral risk." This allows subsequent reasoning to move beyond relying on a single scale threshold and instead make combined judgments within a structured semantic space. Meanwhile, by updating the mechanism, newly collected behavioral feedback, adverse event markings, and symptom changes are continuously written back, enabling the knowledge graph to dynamically adjust the relationship strength and adaptation rules as the patient's state evolves, thus avoiding the failure of static rules in different patients or different scenarios.

[0022] In step S130, intelligent semantic reasoning is performed on the patient's current state based on the aforementioned knowledge graph to determine a set of candidate intervention programs that match the patient's state. This intelligent semantic reasoning is generated based on multi-level reasoning and contextual reasoning. The multi-level reasoning is used to form a hierarchical judgment chain on the ability side, starting from the patient's short-term ability status. For example, it first determines the level of cognitive retention and attention maintenance based on the cognitive function assessment results, then identifies the current main problem domain and its risk level based on the behavioral and mental symptom characteristics, and then determines the number of steps, prompting methods, and assistance intensity that the patient can bear in performing a task based on their living ability status, thereby forming an ability profile corresponding to the activity difficulty, task breakdown granularity, and communication strategies. The contextual reasoning is used to form a constraint and opportunity judgment chain on the "environment side," starting from the patient's situation. For example, it identifies limitations on space, safety, accessibility of items, noise and light stimulation, and resources available to caregivers in the environmental element nodes, and combines this with real-time changes in the patient's emotional state and social interaction to determine whether a certain type of activity has a situational risk of triggering agitation or rejection, or whether it has situational conditions that increase participation. Based on the combined results of the multi-level reasoning and the contextual reasoning, the above reasoning process filters intervention nodes from the knowledge graph that are consistent with the patient's symptoms and ability profile and meet environmental constraints, and then organizes them into several candidate intervention plans. Furthermore, these candidate intervention plans can simultaneously include activity content, task breakdown methods, difficulty levels, communication prompts, environmental modification suggestions, and safety protection points, so that the output plan not only "recommends what to do," but also "explains how to do it and under what conditions."

[0023] In step S140, personalized intervention recommendations for the aforementioned dementia patients are generated from the set of candidate intervention programs. This generation process is not a simple random selection, but rather a selection process that prioritizes participation based on patient interests and preferences. Simultaneously, it re-evaluates the stability and risks of the programs using past feedback data, prioritizing intervention programs that are more likely to be accepted by the patient and can be sustainably implemented. For example, when a patient has clear interests (such as tidying up, listening to music from a specific era, or simple crafts) and the current environment provides the necessary items and space, activities aligned with those interests can be prioritized. The activity tasks are then broken down and the intensity of prompts, guidance, and assistance is adjusted according to the patient's cognitive level and living abilities. When the patient is at risk of nighttime agitation or wandering, environmental stimulation reduction strategies and safety precautions can be provided in the recommendations. Furthermore, when caregivers have limited resources, low-cost, highly operable, and reusable activities and communication strategies are prioritized. Through the above methods, this embodiment realizes adaptive recommendation of intervention plans for multiple scenarios of "hospital-community-home", so that the recommendation results can simultaneously meet the requirements of individualization, feasibility and safety, and provide a structured closed-loop foundation for subsequent continuous feedback and updates.

[0024] In summary, compared to existing methods that primarily rely on scale thresholds or human experience for intervention selection, this invention introduces a dementia intervention knowledge graph containing patient nodes, symptom nodes, intervention measure nodes, and environmental element nodes. This model unifies the complex relationships between patient ability status, symptom presentation, intervention methods, and care environment in a structured semantic form, enabling intervention decisions to move beyond single indicators or static rules and instead allow for combined reasoning across multiple dimensions. This significantly improves the precision and rationality of intervention matching. Furthermore, this invention incorporates an intelligent semantic reasoning mechanism combining multi-level reasoning and contextual reasoning in the recommendation process, simultaneously constraining and filtering intervention plans from both the patient's short-term ability status and the surrounding context. Compared to existing technologies that recommend interventions solely based on patient ability or symptoms, this invention effectively avoids the problems of "suitable ability but unfeasible context" or "allowable context but mismatched ability," making the recommendations more operable and safer in practice, and reducing the risk of agitation, rejection, or adverse events. This invention continuously collects behavioral feedback, adverse event markers, and symptom change information during intervention implementation and writes this information back into a knowledge graph to dynamically update node relationships, enabling the knowledge graph to self-correct as the patient's state evolves. Compared to traditional static recommendation models, this invention can progressively optimize the priority and adaptation rules of intervention programs during long-term care, improving the stability and sustainability of intervention effects and preventing recommendation failure due to changes in patient state. This invention introduces a screening strategy prioritizing interest preferences and participation when generating personalized intervention recommendations, making the recommended programs not only reasonable from a medical and care perspective but also more advantageous in terms of patient subjective acceptance and participation enthusiasm, thus helping to improve adherence to non-pharmacological interventions and long-term implementation effects. Through a knowledge graph-driven dynamic semantic reasoning and feedback closed-loop mechanism, this invention achieves adaptive recommendations for dementia intervention programs in multiple scenarios and stages, achieving significant improvements in personalization level, implementation feasibility, safety, and long-term evolutionary capacity compared to background technologies, demonstrating good practical application value.

[0025] In one feasible implementation, the above-mentioned construction and updating of the dementia intervention knowledge graph based on the aforementioned individual characteristic data includes: The cognitive ability, behavioral symptoms, daily living skills, and interest preferences of the patients were extracted from the individual characteristic data mentioned above, and these data were mapped to patient nodes and symptom nodes in the dementia intervention knowledge graph mentioned above. Based on the characteristics of the patient nodes and symptom nodes mentioned above, and using the reasoning mechanism of the knowledge graph, the relationship between the intervention measure nodes and the environmental element nodes is automatically established, and the node weights are dynamically adjusted to construct a knowledge graph for dementia intervention. Based on the aforementioned individual characteristic data and patient behavioral feedback, the node relationships in the dementia intervention knowledge graph are updated regularly.

[0026] Exemplary cognitive function assessment results reflect the patient's cognitive level, such as memory and attention abilities. Behavioral and psychiatric symptom characteristics describe BPSD symptoms such as mood swings and agitation; daily living ability status characterizes the patient's self-care ability in daily life. Interest and preference information, through analysis of the patient's interests and habitual activities, helps understand what activities the patient prefers to participate in. By collecting this information, we can better understand the patient's specific situation and provide precise data support for the development of subsequent intervention plans.

[0027] Based on the aforementioned individual characteristic data, a knowledge graph for dementia intervention was constructed and updated. This graph includes patient nodes, symptom nodes, intervention measure nodes, and environmental element nodes, establishing interrelationships between these nodes. For example, the patient node carries the patient's identity information, cognitive abilities, and behavioral symptoms. The symptom node records various symptoms exhibited by the patient, such as cognitive impairment and emotional instability. The intervention measure node includes various feasible intervention programs, such as cognitive training and behavioral therapy; the environmental element node includes the patient's environmental conditions, such as living environment and care resources. By integrating this information into the knowledge graph, the patient's current state and its relationship with intervention measures and environmental elements can be accurately depicted, providing data support for subsequent reasoning and intervention.

[0028] Building upon this foundation, and based on the aforementioned knowledge graph, we perform intelligent semantic reasoning to determine the appropriate intervention plan based on the patient's current state. During the reasoning process, multi-level and contextual reasoning are combined, ensuring that the recommended plan considers not only the patient's short-term cognitive function and behavioral symptoms, but also contextual factors such as the patient's environment and emotional state. Through this comprehensive reasoning mechanism, we can more accurately select a set of candidate intervention plans suitable for the patient's current state.

[0029] After generating the candidate intervention set, we select the most suitable option based on the patient's individual needs, resulting in a personalized intervention recommendation. This recommendation is tailored not only to the patient's abilities and symptom characteristics but also incorporates information such as their interests and preferences to improve the acceptability and participation of the intervention. Furthermore, the selection of the intervention also considers the patient's environmental conditions to ensure that the intervention is feasible and effective in practice.

[0030] In this way, this embodiment realizes the recommendation of dementia intervention programs based on knowledge graphs, which can dynamically and accurately provide patients with personalized intervention programs, and continuously optimize and adjust them as the patient's condition changes to ensure the maximization of intervention effects.

[0031] In one feasible implementation, based on the characteristics of the patient nodes and symptom nodes, and using a knowledge graph reasoning mechanism, the relationship between intervention measure nodes and environmental element nodes is automatically established, and the node weights are dynamically adjusted to construct a dementia intervention knowledge graph, including: Based on the cognitive ability level, behavioral and mental symptom type, and living ability status corresponding to the above patient nodes, a set of candidate intervention measures nodes associated with the above patient nodes is determined. Based on the historical intervention effect correlation between the above symptom nodes and the above intervention measure nodes, calculate the adaptation weight of the above intervention measure nodes for different symptom types; Based on the care environment constraints represented by the above environmental element nodes, the feasibility of the above intervention measures nodes is screened. Based on the above adaptation weights and the above feasibility screening results, the above-mentioned intervention measure nodes and the above-mentioned environmental element nodes are established, and the above-mentioned adaptation weights are written into the above-mentioned association relationship to form a weighted relationship structure that can be used for subsequent semantic reasoning.

[0032] For example, based on the cognitive ability level, behavioral and psychiatric symptom types, and daily living ability status corresponding to the aforementioned patient nodes, the patient's current functional profile is quantitatively represented, and accordingly, a set of candidate intervention measures nodes matching the functional needs of the aforementioned patient nodes is selected from the set of intervention measure nodes. For example, the patient's cognitive ability, behavioral symptoms, and daily living ability status can be represented as follows: and The set of candidate interventions is determined using the following function: in, The function represents the set of candidate intervention nodes that are initially matched with the patient's current functional state. This is used to characterize the correspondence between the patient's ability status and the functional requirements of intervention measures, thereby avoiding intervention measures that are clearly mismatched with the patient's ability from entering the subsequent reasoning process.

[0033] Subsequently, based on the historical intervention effect correlation between the aforementioned symptom nodes and intervention measure nodes, the suitability of different intervention measure nodes for addressing different symptom types was quantitatively evaluated. Specifically, intervention measures in historical intervention records can be used... For symptoms To assess the improvement effect, the corresponding adaptation weights are calculated, which can be exemplarily calculated as follows: in, The adaptation weight of the intervention node is used to characterize the historical effectiveness of the intervention for the current symptom type; Indicates intervention measures For symptoms Historical intervention effectiveness evaluation values; These are weighting coefficients corresponding to historical samples, used to balance the impact of data from different time periods or sources on the current inference results. By introducing these adaptive weights, the selection of intervention measures no longer depends on static rules, but rather reflects the long-term statistical regularities of the actual intervention effects.

[0034] Based on this, and further considering the care environment constraints represented by the aforementioned environmental element nodes, the feasibility of the candidate intervention measures is screened. The environmental element nodes may include factors such as care facility type, available care resources, caregiver skill level, and safety constraints. For example, the environmental suitability screening of candidate intervention measures can be performed using the following function: in, This represents the set of intervention points that are feasible to implement in the current care environment. The function represents a set of environmental element nodes. This is used to eliminate interventions that are difficult to implement or pose safety risks under current environmental conditions, thereby improving the feasibility and safety of the recommendations in practical applications.

[0035] Finally, based on the aforementioned adaptation weights and feasibility screening results, a weighted association relationship is established between the intervention measure node and the environmental element node, and the adaptation weights are written into the association relationship to form a weighted relationship structure that can be used for subsequent semantic reasoning. For example, the adaptation weights and feasibility results can be fused to obtain the final weighted association relationship: in, This represents the final weighted association between intervention measure nodes and environmental element nodes. Indicates the first The adaptation weight of each intervention node, This indicates the feasibility assessment results of the intervention measure in the current environment. This represents the number of intervention measure nodes participating in the calculation. The weighted relation structure constructed in the above manner is written into the dementia intervention knowledge graph to support subsequent intelligent semantic reasoning processes based on multi-level reasoning and contextual reasoning.

[0036] In this embodiment, by dynamically constructing and updating the dementia intervention knowledge graph, not only can the intervention plan be accurately matched with the patient's abilities and symptoms, but the patient's environmental conditions are also fully considered, ensuring the high feasibility of the intervention plan. By weighted calculation and dynamic adjustment of the association relationships between intervention measure nodes, symptom nodes, and environmental element nodes, the recommendation system can be optimized in real time and continuously based on the patient's feedback and changes, thereby providing personalized and effective intervention plans.

[0037] In one feasible implementation, the node relationships in the dementia intervention knowledge graph are periodically updated based on the aforementioned individual characteristic data and patient behavioral feedback, including: During the implementation of the intervention program, behavioral response data, symptom change data, and adverse event marker data of the patients were collected during the execution of the intervention measures. Based on the above behavioral response data and the above symptom change data, the correlation between the above intervention measure nodes and the above symptom nodes is evaluated to generate the corresponding intervention effect evaluation value. Based on the above intervention effect evaluation values, the correlation weights between the above intervention measure nodes and the above symptom nodes are adjusted; The adjusted association weights are written back into the dementia intervention knowledge graph to update the reasoning basis of the knowledge graph, so that the subsequently generated set of candidate intervention programs can reflect the dynamic changes in the patient's condition.

[0038] For example, based on the aforementioned individual characteristic data and patient behavioral feedback, the node relationships in the dementia intervention knowledge graph are updated periodically to address a typical problem that arises during the long-term implementation of dementia intervention: the emotional state, environmental stimuli, and care resources of the same patient fluctuate continuously at different times, causing a certain intervention measure to be effective in the early stages but may fail or even trigger adverse events in the later stages. Therefore, it is necessary to transform the real feedback during the intervention process into calculable evidence and write this evidence back into the relational structure of the knowledge graph, so that subsequent reasoning can inherit the experience of what is effective, what is dangerous, and under what conditions it is effective for this patient, thereby forming an adaptive closed-loop optimization.

[0039] Specifically, during the implementation of the intervention program, the system continuously collects behavioral response data, symptom change data, and adverse event marker data from the patients as they perform the intervention measures. The behavioral response data may include quantifiable indicators such as participation, completion rate, number of refusal attempts, frequency of agitation, number of prompts, and number of task interruptions. The symptom change data may include changes in NPI item scores, changes in sleep quality, changes in emotional stability, and observational records related to the target symptoms. The adverse event marker data may include events such as fall risk, wandering risk, verbal / physical aggression, and significant anxiety / fear, along with their severity levels. Through continuous collection of these three types of data, the system can obtain a dynamic chain of evidence regarding the relationship between the intervention and symptoms, rather than relying solely on a one-time assessment or subjective impression.

[0040] Based on the aforementioned behavioral response data and symptom change data, the correlation between the aforementioned intervention measure nodes and symptom nodes is evaluated to generate corresponding intervention effect evaluation values. For example, to simultaneously characterize effectiveness and acceptability and incorporate adverse event risk into the evaluation, this embodiment can construct an intervention effect evaluation function with a risk penalty term and a time consistency term. This ensures that the evaluation not only reflects whether symptoms have improved but also whether the improvement is stable and whether it comes at the cost of high risk. For instance, the intervention effect evaluation value within one update cycle can be defined as: in, Indicates within the time window Internal, intervention nodes Acting on symptom nodes The obtained intervention effect evaluation value; The degree of improvement in the target symptoms can be represented by the decrease in symptom scores or the decrease in the frequency of symptom occurrence. A composite score indicating patient engagement or acceptability indicators, such as completion rate and duration of active participation; This represents a risk penalty item, which is mapped from adverse event labeling data; the higher the risk, the larger the penalty item. The penalty items for caregiving burden can be obtained by combining indicators such as the number of times caregivers remind them, the number of times intervention is interrupted, and the additional manpower input. These are the weighting coefficients for each item, used to balance the contributions of symptom improvement, participation, risk, and burden in the overall evaluation.

[0041] Through the above evaluation function, the system can comprehensively compare the real effects of different interventions on the same symptom under the same scale, and naturally distinguish between different situations such as effective and safe, effective but high risk, and poor participation leading to difficulty in sustainability.

[0042] After obtaining the above-mentioned intervention effect evaluation values, the correlation weights between the above-mentioned intervention measure nodes and the above-mentioned symptom nodes are adjusted according to these evaluation values. To avoid evaluation fluctuations, i.e., drastic weight oscillations, caused by simple linear weighting, this embodiment can employ an update rule with a smoothing factor. This allows the weights to adaptively adjust with changes in effect while maintaining a certain degree of stability, thus better aligning with the gradual nature of clinical intervention. For example, the following weight update formula can be used: in, Indicates time Intervention measures node Symptom nodes The correlation weight between them This indicates the updated association weight; This indicates the update step size or smoothing coefficient. The larger the value, the more sensitive it is to the latest feedback. This represents a normalization mapping function, such as the sigmoid function, used to map the intervention effect evaluation values ​​to a preset weight range, thereby ensuring that weight updates are controllable and easy to use for graph inference. This update method ensures that when an intervention consistently brings symptom improvement and low risk in the current period, the weight will be gradually increased; conversely, when adverse events occur or participation significantly decreases, the weight will be suppressed, thus solidifying the experience of seeking benefits and avoiding harm at the knowledge graph level.

[0043] Furthermore, to enable the knowledge graph to identify the difference between short-term, accidental improvements and long-term stable effectiveness, this embodiment can also introduce a time consistency constraint, which enhances the robustness of the weights by penalizing the volatility of evaluation values ​​across multiple periods. For example, a stability factor can be defined and participate in the update process: Based on this, the weight update is expanded to: in, Indicates that recently The larger the variance of the intervention effect evaluation value within each update cycle, the more unstable the effect is. This is a stability penalty coefficient; This is a stability factor, with a value range of [value range missing]. When the effect is more stable The closer it is to 1, the more fully the weights can be increased; when the effect fluctuates significantly... This will reduce the weights, making weight updates more conservative. Through this mechanism, the system can prioritize learning stable, effective, and sustainable intervention relationships, reducing the risk of mislearning caused by occasional fluctuations.

[0044] Finally, the adjusted association weights are written back into the dementia intervention knowledge graph to update the reasoning basis of the knowledge graph, so that the subsequently generated set of candidate intervention programs can reflect the dynamic changes in the patient's condition.

[0045] In one feasible implementation, the above-mentioned semantic reasoning based on the knowledge graph to determine a set of candidate intervention schemes matching the patient's current state includes: By conducting multi-level inference on short-term characteristics such as patients’ cognitive function, behavioral symptoms and living abilities, multi-level inference results can be obtained. By combining patients' long-term behavioral patterns, historical feedback data, and other situational reasoning, the system analyzes real-time changes in patients' emotional state, environmental factors, and social interactions to generate situational reasoning results. Based on the above multi-level reasoning results and the above scenario reasoning results, the matching degree between the patient's current state and multiple intervention measure nodes is calculated to form a set of candidate intervention plans.

[0046] In one feasible implementation, the above-mentioned multi-level reasoning is performed on short-term characteristics such as the patient's cognitive function, behavioral symptoms, and daily living abilities to obtain multi-level reasoning results, including: Based on the cognitive function assessment results of the above patients, a first-level reasoning feature representing the patient's current cognitive level is generated; Based on the above-mentioned behavioral and psychological symptom characteristics of patients, a second-level inference feature is generated to characterize the patient's behavioral risk and symptom intensity. Based on the patients' living ability status, a third-level inference feature representing the degree of functional limitation of the patients is generated. The first-level reasoning features, the second-level reasoning features, and the third-level reasoning features are hierarchically fused to form the multi-level reasoning results used to characterize the patient's short-term comprehensive state.

[0047] In one feasible implementation, the above-mentioned analysis of real-time changes in the patient's emotional state, environmental factors, and social interactions, combined with the patient's long-term behavioral patterns, historical feedback data, and other contextual reasoning, generates contextual reasoning results, including: Based on the historical behavioral feedback data of the patients mentioned above, the behavioral response patterns of the patients under different intervention scenarios were extracted to form long-term behavioral pattern characteristics. Based on the above environmental element nodes, the constraints and stimulating factors of the patient's current care environment are obtained to form environmental context characteristics. Based on the changes in patients' emotional state and social interactions, emotional and social characteristics that characterize patients' psychological and social adaptation status are generated. By jointly modeling the aforementioned long-term behavioral pattern characteristics, environmental context characteristics, and emotional and social characteristics, the aforementioned situational reasoning results are generated to characterize the patient's current external and psychological situation.

[0048] In one feasible implementation, based on the above-mentioned multi-level reasoning results and the above-mentioned scenario reasoning results, the matching degree between the patient's current state and multiple intervention measure nodes is calculated to form a set of candidate intervention plans, including: Based on the above multi-level reasoning results, the ability matching degree between the patient's short-term ability status and the functional requirements of the above intervention measures nodes is calculated. Based on the above scenario reasoning results, the scenario fit between the patient's current situation and the implementation conditions of the above intervention measures nodes is calculated; When there is a conflict between the above ability matching degree and the above situational adaptability degree, the above ability matching degree is adjusted punitively based on the preset conflict suppression rules; When the above capability matching degree is consistent with the above situational adaptability degree, the two are synergistically enhanced to generate the above comprehensive matching degree. Based on the comprehensive matching degree, the nodes of the above-mentioned intervention measures are screened and sorted to form the above-mentioned set of candidate intervention schemes.

[0049] For example, the semantic reasoning of the patient's current state based on the above knowledge graph to determine the set of candidate intervention programs that match the patient's state is not a simple rule matching or a single similarity retrieval, but rather incorporates the patient's short-term ability state and its situation into the same reasoning framework. First, a comprehensive short-term state of the patient is formed through multi-level reasoning, then an external and psychological situation profile is formed through contextual reasoning, and finally, the intervention measure nodes are matched and calculated in the knowledge graph from the dual perspectives of ability and condition, and a comprehensive matching degree is generated through conflict suppression and synergy enhancement mechanisms.

[0050] Specifically, when performing multi-level inference on short-term characteristics such as a patient's cognitive function, behavioral symptoms, and daily living abilities to obtain multi-level inference results, this embodiment first generates a first-level inference feature representing the patient's current cognitive level based on the aforementioned cognitive function assessment results. This first-level inference feature can be used to characterize the availability of the patient's abilities such as attention maintenance, instruction comprehension, memory retention, and orientation. Subsequently, based on the aforementioned behavioral and psychiatric symptom characteristics, a second-level inference feature representing the patient's behavioral risks and symptom intensity is generated. This second-level inference feature is used to characterize the category and intensity of BPSD-related risks such as agitation, anxiety, depression, hallucinations, and nocturnal behavioral disorders. Furthermore, based on the aforementioned daily living ability status, a third-level inference feature representing the degree of functional limitation of the patient is generated. This third-level inference feature is used to characterize the degree of dependence of the patient on caregiver assistance, prompts, and task breakdown granularity when completing daily activities. To ensure that the multi-level inference has clear hierarchical semantics and facilitates subsequent calculations, this embodiment can represent the aforementioned three-level inference features as vector forms. and And through hierarchical fusion functions, multi-level inference results are formed to characterize the patient's short-term comprehensive state. For example, a fusion formula with "risk gating" characteristics can be used: in, This represents the result of the above multi-level reasoning; This represents element-wise multiplication; This represents the behavioral risk gating function calculated from the second-level inference features mentioned above, and its output range is... This is used to suppress the cognitive contribution to executability when behavioral risk is high; This represents the characteristics of the first level of reasoning mentioned above; This represents the third level of reasoning characteristics mentioned above. By introducing the aforementioned risk gating mechanism, even if the patient's cognitive assessment is good, when the risk of behavioral agitation increases significantly, the multi-level reasoning results will still reflect a decrease in short-term executability, thereby avoiding recommendations for complex activities that are likely to trigger conflict or rejection in the present moment.

[0051] In generating scenario-based reasoning results, this embodiment uses joint modeling of long-term behavioral patterns, environmental context, and emotional and social characteristics to form a comprehensive result that characterizes the patient's external and psychological context. Specifically, based on the patient's historical behavioral feedback data, behavioral response patterns under different intervention scenarios are extracted, such as participation, refusal probability, and agitation trigger probability in scenarios involving musical stimulation, manual activities, and step-by-step dressing training, to form long-term behavioral pattern characteristics. Based on the aforementioned environmental element nodes, the constraints and stimuli of the patient's current care environment are obtained, such as noise, lighting, space availability, available time for caregivers, available items, and accessibility of potentially hazardous items, thus forming environmental contextual characteristics. Based on the changes in the patients' emotional state and social interactions, emotional and social characteristics M representing their psychological and social adaptation status are generated.

[0052] Based on this, the aforementioned long-term behavioral pattern characteristics, environmental context characteristics, and emotional and social characteristics are jointly modeled to generate a contextual reasoning result C. An example of this can be achieved using the following contextual accessibility modeling formula: in, This represents the result of the reasoning in the above scenario; It represents a contextual accessibility vector determined by long-term behavioral patterns and emotional and social states, used to characterize a patient's potential to accept different types of intervention activities under current psycho-behavioral trends. This represents the feature vector of the environmental context, and through element-wise constraints, the acceptable outcome is further converged to one that is feasible in the current environment. In this way, the contextual reasoning results can simultaneously reflect both the patient's willingness to act and the environment's permission to act, providing a unified contextual representation for subsequent matching calculations.

[0053] When calculating the matching degree between the patient's current state and multiple intervention measure nodes based on the above-mentioned multi-level reasoning results and scenario reasoning results, and forming a set of candidate intervention plans, this embodiment first calculates the capability matching degree between the patient's short-term capability status and the functional requirements of the intervention measure nodes based on the above-mentioned multi-level reasoning results. For example, the functional requirements of the intervention measure nodes can be represented as a requirement vector. And calculate the first one in the following way Capacity matching of each intervention measure: in, Indicates the degree of ability matching. This indicates the result of the above multi-level reasoning. Indicates the first Functional requirement vector corresponding to each intervention measure node The distance represents the norm distance; the smaller the distance, the closer the ability status is to the functional requirements, and thus the higher the ability matching degree. Subsequently, based on the above scenario reasoning results, the scenario fit degree between the patient's current situation and the implementation conditions of the intervention node is calculated. For example, the [number]th [intervention node] can be [calculated / exemplarily / etc.]. The conditions for implementing an intervention are represented as a condition vector. And calculate: in, Indicates context fit. This represents the result of the reasoning in the above scenario. Indicates the first The implementation condition vector of each intervention node is calculated as a cosine similarity, which measures the degree of consistency between the current situation and the implementation conditions.

[0054] After obtaining the ability matching degree and situational fit degree, this embodiment does not use a simple weighted summation, but introduces a nonlinear fusion mechanism of conflict suppression and synergistic enhancement, enabling the matching calculation to express common clinical contradictions such as short-term ability being permissible but situational incompatibility or situational permissibility but insufficient ability. For example, the conflict degree can be defined first: When there is a conflict between the above-mentioned capability matching degree and the above-mentioned situational adaptability degree, that is... Exceeding the preset conflict threshold Then, based on the preset conflict suppression rules, a penalty adjustment is made to the ability matching degree. For example, the following penalty function can be used: in, This indicates the ability matching degree after punishment. This is the conflict penalty coefficient. The higher the value, the stronger the penalty, thus avoiding outputting a high matching degree result under strong situational conflict; when the above ability matching degree is consistent with the above situational fit degree, that is... Then, the two are subjected to synergistic enhancement processing to generate a comprehensive matching degree. For example, the following synergistic enhancement formula can be used: in, Indicates the overall matching degree. For the synergistic enhancement coefficient, This is the geometric mean term, used to ensure that a lower value for either term would significantly reduce the overall match, avoiding an artificially high match due to one term being higher than the other. Additional gains are granted when the consistency between the two is higher, so that interventions that truly match both ability and context are ranked higher in the ranking. Finally, the intervention nodes are screened and ranked according to the above comprehensive matching degree to form a set of candidate intervention schemes, and the top-ranked intervention nodes are combined into candidate schemes and output to the subsequent personalized recommendation steps.

[0055] In this embodiment, multi-level reasoning decomposes the patient's short-term state into a two-layer structure of cognition, behavior, and function, and achieves hierarchical fusion through risk gating. Contextual reasoning models long-term response patterns, environmental constraints, and emotional and social states into contextual accessibility. Matching degree calculation avoids the risk of misrecommendation caused by simple weighting through conflict suppression and synergy enhancement, so that the set of candidate intervention programs can more realistically reflect the comprehensive conditions that the patient can do at present, the environment can do, and that doing so is safer and more likely to be effective, thereby improving the feasibility, acceptability, and safety of intervention recommendations.

[0056] In one feasible implementation, the node relationships in the dementia intervention knowledge graph are periodically updated based on the aforementioned individual characteristic data and patient behavioral feedback, including: During the intervention program implementation period, time-series analysis was conducted on the patient's behavioral response trends, symptom evolution trajectory, and frequency of adverse events during the implementation of the above intervention measures to generate feedback evolution characteristics that characterize the stability of the intervention and changes in risk. Based on the above feedback evolution characteristics, the correlation between the above intervention measure nodes and the above symptom nodes is classified into three types: stable enhancement type, fluctuation adaptation type and risk suppression type. For the aforementioned stable and enhanced associations, the association weights are incrementally updated to strengthen their priority in subsequent semantic reasoning. To address the aforementioned fluctuating adaptive correlations, a time decay or contextual condition constraint mechanism is introduced to conditionally adjust the correlation weights. For the aforementioned risk-suppressing associations, their association weights are reduced or frozen, and they are marked as risk-constrained associations in the aforementioned dementia intervention knowledge graph; The above-described classification and adjusted relationships are written back into the above-described dementia intervention knowledge graph, so that the knowledge graph forms a dynamic reasoning structure with risk perception ability and long-term self-evolution characteristics.

[0057] For example, based on the aforementioned individual characteristic data and patient behavioral feedback, regularly updating the node relationships in the dementia intervention knowledge graph is not merely a simple increase or decrease of association weights. Instead, it treats the feedback changes during the intervention implementation period as an evolutionary process. By conducting time-series analysis on behavioral response trends, symptom evolution trajectories, and the frequency of adverse events, it extracts feedback evolution characteristics that can simultaneously characterize the stability of the effect and the trend of risk. Based on this, it classifies and manages the association between intervention measure nodes and symptom nodes, thereby enabling the knowledge graph to have dynamic reasoning capabilities such as long-term learning, risk perception, and conditional reasoning.

[0058] Specifically, during the intervention program implementation period, the system continuously collects behavioral response data, symptom change data, and adverse event marker data of patients during the implementation of the above-mentioned intervention measures, and forms a time series within a preset time window. The behavioral response data may include participation, completion rate, number of refusal attempts, number of prompts, and number of agitation events. The symptom change data may include changes in NPI item scores or symptom frequency. The adverse event marker data may include fall risk, aggressive behavior, significant fear, or events related to getting lost. Based on the above time series, this embodiment performs time-series analysis on behavioral response trends, symptom evolution trajectories, and adverse event frequency to generate feedback evolution characteristics. For example, the feedback evolution characteristics can be constructed as a triple consisting of stability, improvement trend, and risk trend. in, Indicates during the update cycle Feedback evolution feature vector within, Indicates the stability of the intervention effect. Indicates the strength of the trend of symptom improvement. This indicates the intensity of risk changes. To obtain the aforementioned stability and trend, this embodiment can analyze the target symptom scoring sequence. With adverse event count sequence Perform regression and volatility analysis, such as on recent A time slice The trend of symptom improvement is defined as: in, Indicates the symptom scoring sequence The slope obtained from linear fitting is used; the more negative the slope, the faster the score decreases. Therefore, a negative sign is taken. The larger the value, the greater the improvement. Furthermore, to characterize whether the improvement is stable, stability is defined as a penalty term for fluctuations in the fitted residuals, for example: in, This represents the fitted predicted value. Represents variance. This is a stability penalty coefficient; a larger variance indicates stronger fluctuations. The smaller the value, the better. Furthermore, to characterize risk trends, trends can be extracted from the frequency of adverse events, for example: in, Represents the sequence of adverse event counts. The slope obtained from the fitting, when positive, indicates an increased risk. The larger the slope, the lower the risk if the slope is negative or close to 0. Therefore, take... This emphasizes the punitive significance of escalating risk. Through the aforementioned feedback evolution characteristics, the system can simultaneously incorporate whether symptoms are improving or worsening, whether the improvement is stable, and whether the risk is rising into a computable framework.

[0059] After obtaining the aforementioned feedback evolution characteristics, this embodiment classifies the association between intervention nodes and symptom nodes based on these characteristics, distinguishing them into stable enhancement, fluctuating adaptation, and risk suppression associations. To ensure the feasibility and interpretability of the classification rules, a classification determination function can be constructed, for example: in, Indicates the node of intervention measures Symptom nodes In the cycle Internal relationship types, To improve the threshold, To stabilize the threshold, This is the upper limit threshold for risk. The risk trigger threshold is usually greater than 100%. Based on this classification, the system can classify relationships that are significantly improved, stable, and low in risk as stable-enhancing relationships, relationships with significantly increased risk as risk-suppressing relationships, and relationships that fall between the two and are greatly affected by the situation as volatile-adaptive relationships.

[0060] When an association is determined to be stable and enhanced, this embodiment performs an incremental update of the association weight to strengthen its priority in subsequent semantic reasoning. For example, a monotonic enhancement update rule with an upper limit can be used: in, This indicates the correlation weight for the current period. This indicates the updated association weight. To enhance the step size factor, the improvement trend is multiplied by the stability, so that relationships with more obvious and stable improvement grow faster, while relationships with improvement but instability grow more slowly, thus ensuring that the enhancement logic has clinical rationality.

[0061] When the association is determined to be of the fluctuating adaptive type, this embodiment introduces a time decay or situational condition constraint mechanism to conditionally adjust the association weight, so that this type of relationship is not simply increased or decreased in subsequent reasoning, but can only be activated when specific situational conditions are met.

[0062] For example, a context-gated function can be established for fluctuation-adaptive relationships. ,in The current context feature vector (e.g., environmental stimulus level, caregiver availability, patient emotional state, etc.) is updated using the following conditional updates: in, To update the smoothing coefficients, For normalized mapping functions, As a risk penalty coefficient, This is used to indicate whether the current situation meets the effective triggering conditions for the intervention; when the situation does not meet these conditions... Approaching 0 causes the weights to decay or maintain low activity, when the situation is satisfied. Approaching 1 allows the weight to increase with the effect. Through this mechanism, fluctuating adaptive relations will not be misjudged as invalid and permanently suppressed, nor will they be over-promoted in inappropriate situations, thus achieving conditionally usable knowledge representation.

[0063] When an association is determined to be risk-suppressive, this embodiment reduces or freezes its association weight and marks it as a risk-constrained association in the dementia intervention knowledge graph to prevent the intervention from being continuously recalled or placed with high priority in subsequent reasoning.

[0064] For example, the following risk suppression update rule can be adopted: Freeze the system when the risk trigger conditions are met: like Then let And marked as in, To suppress the step size coefficient, To minimize the weights or freeze the lower limit, this is used to prevent short-term extreme noise from causing the weights to drop directly to zero and losing traceability. This is a risk flag. When the flag is 1, it can be used as a hard constraint or strong penalty in subsequent semantic reasoning to participate in the candidate solution selection, thereby forming a reasoning logic that prioritizes risk avoidance.

[0065] Finally, the relationships, after being categorized and adjusted as described above, are written back into the dementia intervention knowledge graph, enabling the knowledge graph to form a dynamic reasoning structure with risk perception capabilities and long-term self-evolutionary characteristics. Specifically, the written-back content includes not only the updated relationship weights... It can also include associated type tags. Stability indicators Risk Flag The corresponding context gating condition parameters enable subsequent semantic reasoning to directly utilize structured knowledge of stable enhancement priority, fluctuating adaptive context triggering, and risk suppression strong constraints when retrieving candidate intervention schemes, thereby achieving a safer, more sustainable, and more individualized adaptive recommendation effect.

[0066] Through the above mechanism, this embodiment realizes that the recommendation system can not only learn at the knowledge graph level, thereby significantly improving the creativity and engineering feasibility of the overall solution.

[0067] like Figure 2 As shown, this application proposes a knowledge graph-based dementia intervention program recommendation system 10, including: The acquisition unit 101 is used to acquire individual characteristic data of dementia patients, wherein the aforementioned individual characteristic data includes cognitive function assessment results, behavioral and mental symptom characteristics, living ability status, and interest and preference information; The construction and updating unit 102 is used to construct and update the dementia intervention knowledge graph based on the above-mentioned individual characteristic data, wherein the above-mentioned knowledge graph includes patient nodes, symptom nodes, intervention measure nodes, environmental element nodes and their interrelationships; The determining unit 103 is used to perform intelligent semantic reasoning on the patient's current state based on the above knowledge graph, and to determine a set of candidate intervention schemes that match the patient's current state. The intelligent semantic reasoning is generated based on multi-level reasoning and contextual reasoning. The generation unit 104 is used to generate personalized intervention recommendations for the dementia patients from the above set of candidate intervention programs.

[0068] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for recommending intervention programs for dementia based on knowledge graphs, characterized in that, The method comprises the following steps: obtaining individual characteristic data of a dementia patient, wherein the individual characteristic data comprises cognitive function evaluation results, behavioral and psychological symptom characteristics, life ability state and interest preference information; constructing and updating a dementia intervention knowledge graph based on the individual characteristic data, wherein the knowledge graph comprises patient nodes, symptom nodes, intervention measure nodes, environmental element nodes and their mutual relationships; based on the knowledge graph, intelligently performing semantic reasoning on the current state of the patient to determine a set of candidate intervention schemes matched with the state of the patient, wherein the intelligent semantic reasoning is generated based on multi-level reasoning and context reasoning; generating a personalized intervention recommendation result for the dementia patient from the set of candidate intervention schemes. 2.The knowledge graph-based dementia intervention scheme recommendation method of claim 1, wherein, The step of constructing and updating the dementia intervention knowledge graph based on the individual characteristic data comprises the following steps: extracting cognitive ability, behavioral symptoms, life ability and interest preference data of the patient from the individual characteristic data, and mapping these data as patient nodes and symptom nodes in the dementia intervention knowledge graph; based on the characteristics of the patient nodes and symptom nodes, automatically establishing relationships between intervention measure nodes and environmental element nodes based on the reasoning mechanism of the knowledge graph, and dynamically adjusting node weights to construct the dementia intervention knowledge graph; based on the individual characteristic data and the behavioral feedback of the patient, regularly updating the node relationships in the dementia intervention knowledge graph. 3.The knowledge graph-based dementia intervention scheme recommendation method of claim 2, characterized in that, The step of automatically establishing relationships between intervention measure nodes and environmental element nodes based on the characteristics of the patient nodes and symptom nodes, and dynamically adjusting node weights to construct the dementia intervention knowledge graph based on the reasoning mechanism of the knowledge graph comprises the following steps: based on the cognitive ability level, behavioral and psychological symptom type and life ability state corresponding to the patient node, determining a set of candidate intervention measure nodes associated with the patient node; based on the historical intervention effect association relationship between the symptom nodes and the intervention measure nodes, calculating the adaptation weight of the intervention measure nodes to different symptom types; combining the care environment constraint conditions represented in the environmental element nodes, performing implementability screening on the intervention measure nodes; based on the adaptation weight and the implementability screening result, establishing the association relationship between the intervention measure nodes and the environmental element nodes, and writing the adaptation weight into the association relationship to form a weighted relationship structure that can be used for subsequent semantic reasoning. 4.The knowledge graph-based dementia intervention scheme recommendation method of claim 2, characterized in that, The step of regularly updating the node relationships in the dementia intervention knowledge graph based on the individual characteristic data and the behavioral feedback of the patient comprises the following steps: during the implementation of the intervention scheme, collecting behavioral response data, symptom change data and adverse event marker data of the patient during the execution of the intervention measures; based on the behavioral response data and the symptom change data, performing effect evaluation on the association relationship between the intervention measure nodes and the symptom nodes to generate corresponding intervention effect evaluation values; based on the intervention effect evaluation values, adjusting the association weight between the intervention measure nodes and the symptom nodes; Write back the adjusted correlation weight to the dementia intervention knowledge graph to update the reasoning basis of the knowledge graph, so that the subsequent generated candidate intervention scheme set can reflect the dynamic changes of the patient state.

5. The knowledge graph-based dementia intervention program recommendation method according to any one of claims 1 to 4, characterized in that, The semantic reasoning of the knowledge graph based on the current state of the patient determines a set of candidate intervention schemes matched with the patient's state, which includes: Through multi-level reasoning on the patient's cognitive function, behavioral symptoms, and life ability, etc. Short-term characteristics to obtain multi-level reasoning results; Combine the patient's long-term behavior pattern, historical feedback data and other context reasoning to analyze the real-time changes of the patient's emotional state, environmental factors and social interaction to generate scenario reasoning results; Based on the multi-level reasoning results and the scenario reasoning results, calculate the matching degree of the patient's current state and multiple intervention measure nodes to form a candidate intervention scheme set. 6.The knowledge graph-based dementia intervention scheme recommendation method of claim 5, characterized in that, The multi-level reasoning on the patient's cognitive function, behavioral symptoms, and life ability, etc. Short-term characteristics to obtain multi-level reasoning results, including: Based on the patient's cognitive function evaluation results, generate first-level reasoning features representing the patient's current cognitive level; Based on the patient's behavioral and psychological symptom characteristics, generate second-level reasoning features representing the patient's behavioral risk and symptom intensity; Based on the patient's life ability state, generate third-level reasoning features representing the patient's functional restriction degree; Hierarchical fusion of the first-level reasoning features, the second-level reasoning features and the third-level reasoning features forms the multi-level reasoning results used to represent the patient's short-term comprehensive state. 7.The knowledge graph-based dementia intervention scheme recommendation method of claim 5, characterized in that, The combination of the patient's long-term behavior pattern, historical feedback data and other context reasoning to analyze the real-time changes of the patient's emotional state, environmental factors and social interaction to generate scenario reasoning results, including: Based on the patient's historical behavior feedback data, extract the patient's behavior response pattern under different intervention situations to form long-term behavior pattern features; Based on the environmental element node, obtain the constraint conditions and stimulating factors of the patient's current care environment to form environmental context features; Based on the patient's emotional state changes and social interaction, generate emotional and social features representing the patient's psychological and social adaptation state; Joint modeling of the long-term behavior pattern features, the environmental context features and the emotional and social features generates the scenario reasoning results used to depict the patient's current external and psychological situation. 8.The knowledge graph-based dementia intervention scheme recommendation method of claim 5, wherein, Based on the multi-level reasoning results and the scenario reasoning results, calculate the matching degree of the patient's current state and multiple intervention measure nodes to form a candidate intervention scheme set, which includes: Based on the multi-level reasoning results, calculate the ability matching degree between the patient's short-term ability state and the functional requirements of the intervention measure nodes; Based on the scenario reasoning results, calculate the situation adaptation degree between the patient's current situation and the implementation conditions of the intervention measure nodes; When the ability matching degree and the situation adaptation degree conflict, perform a punitive adjustment on the ability matching degree based on a preset conflict suppression rule; When the capability matching degree and the situation adaptation degree are consistent, the two are synergistically enhanced to generate the comprehensive matching degree; According to the comprehensive matching degree, the intervention measure nodes are screened and sorted to form the candidate intervention scheme set. 9.The knowledge graph-based dementia intervention scheme recommendation method of claim 2, wherein, Based on the individual characteristic data and the behavior feedback of the patient, the node relationship in the dementia intervention knowledge graph is regularly updated, including: During the implementation period of the intervention scheme, the behavior response trend of the patient in the process of executing the intervention measure, the symptom evolution trajectory, and the frequency of adverse events are analyzed in time sequence to generate feedback evolution features representing intervention stability and risk changes; Based on the feedback evolution features, the association relationship between the intervention measure nodes and the symptom nodes is typed and determined, and is divided into stable enhancement type, fluctuation adaptation type and risk inhibition type association relationship; For the stable enhancement type association relationship, incremental update of the association weight is performed to strengthen its priority in subsequent semantic reasoning; For the fluctuation adaptation type association relationship, a time decay or situation condition constraint mechanism is introduced to conditionally adjust the association weight; For the risk inhibition type association relationship, its association weight is reduced or frozen, and is marked as a risk constraint association in the dementia intervention knowledge graph; The above typed and adjusted association relationship is written back to the dementia intervention knowledge graph, so that the knowledge graph forms a dynamic reasoning structure with risk perception ability and long-term self-evolution characteristics. 10.A knowledge graph-based dementia intervention program recommendation system, characterized by, Comprise: An acquisition unit is configured to acquire individual characteristic data of a dementia patient, wherein the individual characteristic data comprises cognitive function assessment results, behavioral and psychological symptom characteristics, life ability state, and interest preference information; A construction and update unit is configured to construct and update a dementia intervention knowledge graph based on the individual characteristic data, wherein the knowledge graph comprises patient nodes, symptom nodes, intervention measure nodes, environmental element nodes, and their mutual relationships; A determination unit is configured to perform intelligent semantic reasoning on the current state of the patient based on the knowledge graph, and determine a candidate intervention scheme set matched with the state of the patient, wherein the intelligent semantic reasoning is generated based on multi-level reasoning and situation reasoning; A generation unit is configured to generate a personalized intervention recommendation result for the dementia patient from the candidate intervention scheme set.

Images