Psychological disorder intervention device for cardiovascular patients based on knowledge graph

Abstract

The present invention provides a device for intervention of psychological disorders in patients with cardiovascular diseases based on a knowledge graph, which includes: a multi-source heterogeneous data acquisition unit for acquiring multi-source heterogeneous data related to cardiovascular diseases and psychological disorders; a preprocessing unit for preprocessing the multi-source heterogeneous data, extracting key information, and constructing an initial knowledge graph; an expansion and optimization unit for expanding and optimizing the initial knowledge graph using a large language model, and constructing a knowledge graph containing entities of diseases, symptoms, treatments, and patient psychological states and their relationships; a mining unit for mining potential disease comorbidity relationships and personalized intervention strategies contained in the knowledge graph through knowledge reasoning and graph neural network technology; and a dynamic update unit for establishing a dynamic update mechanism to ensure timely updating of the knowledge graph. Based on the knowledge graph constructed by the present invention, accurate analysis and personalized intervention of psychological disorders in patients with cardiovascular diseases can be achieved, providing intelligent support for clinical decision-making.

Description

Technical Field

[0001] The present invention relates to the field of medical information, and particularly to a device for intervening in psychological disorders of cardiovascular disease patients based on a knowledge graph. Background Art

[0002] Cardiovascular disease patients often suffer from psychological disorders, but the current understanding of the relationship between the two is still insufficient, making it difficult to achieve accurate diagnosis and personalized intervention. Traditional methods mainly rely on single data sources and simple statistical analysis, and cannot comprehensively grasp the complex disease comorbidity mechanism. How to effectively integrate multi-source heterogeneous data and construct an association knowledge network between cardiovascular disease and psychological disorders is a major challenge. At the same time, there are also technical bottlenecks in accurately extracting key information from massive unstructured medical texts and establishing semantic connections.

[0003] In addition, psychological disorders have diverse manifestations and complex influencing factors, making it difficult to characterize their development laws with traditional models. How to utilize advanced artificial intelligence technologies, especially deep learning and graph neural networks, to mine potential disease correlations and predict individual risks is an urgent problem to be solved. On the other hand, how to ensure data quality and interpretability during the construction of the knowledge graph, and how to closely integrate with clinical practice and gain expert recognition are also key difficulties.

[0004] Generally speaking, how to break through the limitations of existing technologies, establish an intelligent analysis system that integrates multidisciplinary knowledge and has self-learning ability, and achieve accurate assessment and personalized management of the mental health of cardiovascular disease patients is a major technical challenge in this field. Summary of the Invention

[0005] In view of this, the purpose of the present invention is to provide a device for intervening in psychological disorders of cardiovascular disease patients based on a knowledge graph to improve the above problems.

[0006] The present invention provides a device for intervening in psychological disorders of cardiovascular disease patients based on a knowledge graph, which includes:

[0007] A multi-source heterogeneous data acquisition unit for acquiring multi-source heterogeneous data related to cardiovascular disease and psychological disorders;

[0008] A preprocessing unit for preprocessing the multi-source heterogeneous data, extracting key information, and constructing an initial knowledge graph;

[0009] An expansion and optimization unit for expanding and optimizing the initial knowledge graph by using a pre-trained large language model, and constructing a knowledge graph including entities and relationships of diseases, symptoms, treatments, and patients' psychological states;

[0010] A mining unit, used to mine potential disease comorbidity relationships and personalized intervention strategies contained in the knowledge graph through knowledge reasoning and graph neural network technology, so as to implement intervention in psychological disorders of cardiovascular patients; wherein the knowledge graph is presented through a visual and interactive interface to support user query, reasoning and knowledge acquisition;

[0011] The dynamic update unit is used to establish a dynamic update mechanism, use the large language model to continuously learn new knowledge, and is regularly reviewed and optimized by a cross-domain expert team to ensure timely updating of the knowledge graph.

[0012] Preferably, the multi-source heterogeneous data acquisition unit is specifically used for:

[0013] Acquire structured, semi-structured and unstructured heterogeneous data from evidence-based guidelines, clinical practice guidelines, medical databases, authoritative literature, online platforms and clinical health manuals;

[0014] The acquired heterogeneous data is cleaned, integrated and standardized to extract key information elements, including patient information, disease information, symptom manifestations, treatment plans, and psychological state descriptions, thereby providing a high-quality data foundation for subsequent knowledge graph construction.

[0015] Preferably, the preprocessing unit is specifically used for:

[0016] Define and classify core entities such as patients, diseases, symptoms, treatments, and psychological states;

[0017] Use natural language processing techniques to perform named entity recognition and relation extraction to identify entities and determine the semantic relationships between entities;

[0018] A standardized knowledge graph is constructed based on the extracted entities and relationships, forming nodes for diseases, symptoms, treatments, and patient psychological states, as well as multiple associations between nodes.

[0019] Preferably, the expansion optimization unit is specifically used for:

[0020] On unstructured text data such as medical literature, guidelines, and case reports, we use pre-trained large language models for continuous learning to capture deep semantic knowledge in the fields of cardiovascular disease and psychological disorders. Pre-trained large language models include BERT and GPT.

[0021] Through transfer learning, the knowledge learned by the large language model is transferred to the initial knowledge graph, expanding the entity, relationship and attribute information therein, optimizing the structure and semantic representation of the knowledge graph, and improving the comprehensiveness and accuracy of the knowledge graph.

[0022] Preferably, the excavation unit is specifically used for:

[0023] Apply path-based reasoning and rule-based reasoning techniques to the knowledge graph of patient-disease-symptom-treatment to discover new disease comorbidity patterns and etiological mechanisms;

[0024] Graph neural networks are used to perform representation learning and node classification on knowledge graphs to predict patients' risk of psychological disorders and provide personalized psychological intervention recommendations.

[0025] Preferably, path-based reasoning and rule-based reasoning techniques are applied to the knowledge graph consisting of patient-disease-symptom-treatment to discover new disease comorbidity patterns and etiological mechanisms, including:

[0026] Obtain a set of rules related to disease comorbidity from the knowledge graph. For each rule, calculate the brightness value based on its usage frequency and accuracy in historical reasoning. The brightness value calculation formula is: brightness value = α × usage frequency + β × accuracy, where α and β are preset weight coefficients;

[0027] In the process of disease comorbidity pattern discovery, high-brightness rules are selected from the rule sequence and applied to the knowledge graph. If the data in the knowledge graph meets the conditions of multiple rules at the same time, the rule with the highest brightness value is selected for reasoning first;

[0028] When conflicts between rules are detected, the security policy is activated according to the preset conflict resolution priority to resolve the conflicting rules and retain the rules that meet the security policy. The conflict resolution priority is determined based on the brightness value, timeliness and scope of application of the rules. The security policy includes three methods: rule merging, rule replacement and rule deletion.

[0029] Using the rule set processed by the security strategy, reasoning operations are performed in the knowledge graph to obtain preliminary reasoning results of the disease comorbidity pattern, and the rules used in each reasoning step and their brightness values ​​are recorded. The preliminary reasoning results include possible disease comorbidity relationships, relationship strengths, and confidence levels.

[0030] The reasoning process, the rule sets used, the rule brightness value, the application of the resolution strategy, the preliminary reasoning results, the final reasoning results and the confidence score are stored in the reasoning log of the knowledge graph to support subsequent rule optimization; the confidence score is calculated based on the rule brightness value, the number of rules and the reasoning depth used in the reasoning process.

[0031] Preferably, the knowledge graph is presented through a visual and interactive interface to support user query, reasoning and knowledge acquisition, including:

[0032] Design a visual layout based on the knowledge graph, and intuitively display the complex relationships between entities such as diseases, symptoms, treatments, and patients' psychological states through nodes, edges, and color visual elements;

[0033] Develop interactive functions to allow users to explore the knowledge graph through keyword search, semantic query, and intelligent question-answering, achieving accurate and convenient knowledge retrieval and acquisition;

[0034] Embedded with a knowledge reasoning engine, users can ask questions in natural language, and the system can infer answers from the knowledge graph based on the questions to support decision-making.

[0035] Preferably, the dynamic updating unit is specifically used for:

[0036] Design knowledge graph update process and quality assessment indicators to ensure the standardization and controllability of updates;

[0037] Continuously acquire the latest research results in the fields of cardiovascular disease and psychology, use large language models to automatically extract structured knowledge, and dynamically expand the knowledge graph;

[0038] A review team composed of multidisciplinary experts in cardiovascular disease, psychology, and knowledge engineering will be formed to review and verify the new knowledge to ensure the scientificity and authority of the knowledge graph update.

[0039] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0040] The present invention first obtains and standardizes data related to psychological disorders of cardiovascular patients from multiple data sources, extracts and classifies core entities using natural language processing technology, and then uses the BERT+BiLSTM-CRF model to extract the relationship between entities and construct a knowledge graph of psychological disorders. Then, based on the graph, a graph neural network model is used for reasoning to explore potential comorbidity relationships and personalized intervention strategies. The knowledge graph is then evaluated and improved in combination with evidence-based medicine and expert feedback, and finally a visualization result is generated and an interactive query interface is provided. By integrating multi-source data, deep learning and knowledge graph technology, the present invention achieves accurate analysis and personalized intervention of psychological disorders in cardiovascular patients, providing intelligent support for clinical decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 A schematic diagram of the structure of a knowledge graph-based intervention device for psychological disorders in cardiovascular patients provided in the first embodiment of the present invention. DETAILED DESCRIPTION

[0042] In order to further understand the content of the present invention, the present invention is described in detail in conjunction with the accompanying drawings and embodiments. The present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It is understood that the specific embodiments described herein are only used to explain the relevant inventions, rather than to limit the invention. It is also necessary to explain that, for ease of description, only the parts related to the invention are shown in the accompanying drawings.

[0043] like Figure 1 As shown, the first embodiment of the present invention provides a device for intervening in psychological disorders of cardiovascular patients based on a knowledge graph, which includes:

[0044] The multi-source heterogeneous data acquisition unit 110 is used to acquire multi-source heterogeneous data related to cardiovascular diseases and psychological disorders.

[0045] Specifically, multi-source heterogeneous data related to cardiovascular disease and psychological disorders can be obtained, such as evidence-based guidelines, clinical practice guidelines, medical databases (such as PubMed, Cochrane), authoritative literature, online platforms and clinical health manuals. By cleaning, integrating and standardizing the acquired heterogeneous data, key information elements can be extracted. The key information elements include patient information, disease information, symptom manifestations, treatment plans, and psychological state descriptions, thereby providing a high-quality data foundation for the subsequent construction of a knowledge graph.

[0046] For example, crawler technology can be used to obtain literature data related to cardiovascular diseases and psychological disorders from medical databases such as PubMed and Cochrane, and regular expressions and dictionary matching methods can be used to clean and structure the data, extract structured fields such as patient basic information, disease diagnosis, symptom manifestations, treatment plans, etc., and build a standardized medical knowledge base.

[0047] The preprocessing unit 120 is used to preprocess the multi-source heterogeneous data, extract key information, and construct an initial knowledge graph.

[0048] In this embodiment, natural language processing technology can be used to perform entity recognition and relationship extraction on standardized structured data, identify core entities such as patients, psychological disorders, symptoms, and treatment methods, and extract relationships between entities, such as "suffering from" between patients and psychological disorders, "manifestations" between psychological disorders and symptoms, etc., to obtain entity relationship triples. A knowledge graph is constructed based on entity relationship triples, and entities are classified, such as psychological disorders are divided into subcategories such as anxiety and depression, to form a hierarchical knowledge structure. At the same time, attributes are defined for entities, such as the patient's age, gender, medical history, severity and duration of psychological disorders, etc., and values ​​are assigned to entity attributes through data mapping to obtain an initial knowledge graph of psychological disorders in cardiovascular patients containing entities, relationships and attributes.

[0049] In this embodiment, for example, a BERT+BiLSTM-CRF model may be used to identify entities in text based on natural language processing technology and automatically classify them:

[0050]

[0051] Among them, x is the input text sequence, y is the output label sequence, W and b are model parameters, and entity recognition can be effectively realized through this formula. By defining these categories and their relationships (such as causal relationships, inclusion relationships, similarity relationships, etc.), a hierarchical knowledge structure is formed to support subsequent reasoning and application.

[0052] The expansion and optimization unit 130 is used to expand and optimize the initial knowledge graph using a pre-trained large language model to construct a knowledge graph containing disease, symptom, treatment, patient psychological state entities and their relationships.

[0053] Specifically, in this embodiment:

[0054] First, on unstructured text data such as medical literature, guidelines, and case reports, continuous learning is performed using pre-trained language models to capture deep semantic knowledge in the fields of cardiovascular disease and psychological disorders; pre-trained language models may include BERT and GPT; then, the knowledge learned by the language model is transferred to the initial knowledge graph through transfer learning, the entity, relationship, and attribute information therein is expanded, the structure and semantic representation of the knowledge graph is optimized, and the comprehensiveness and accuracy of the knowledge graph are improved.

[0055] The mining unit 140 is used to mine the potential disease comorbidity relationships and personalized intervention strategies contained in the knowledge graph through knowledge reasoning and graph neural network technology, so as to implement intervention in psychological disorders of cardiovascular patients; wherein the knowledge graph is presented through a visual and interactive interface to support user query, reasoning and knowledge acquisition.

[0056] In this embodiment, a graph neural network model is constructed based on the entities and relationships in the knowledge graph, where each node of the graph neural network model represents an entity and the edge represents the relationship between entities. The graph neural network model is initialized, an initial hidden state is assigned to each node, and the node hidden state is initialized using random initialization or pre-trained word vectors.

[0057]

[0058] Update the hidden state of each node, where h v is the hidden state of node v, σ is the activation function, W uv is the weight matrix from node u to node v, h u is the hidden state of node u, b v is the bias term of node v, and N(v) is the set of neighbor nodes of node v. The hidden states of all nodes are updated iteratively until the preset number of iterations or convergence conditions are reached to obtain the final node hidden state representation.

[0059] In this embodiment, the updated node hidden state is used to mine potential disease comorbidity relationships in the knowledge graph through tasks such as link prediction or node classification, that is, to determine whether there is a comorbidity relationship between two disease nodes. According to the attribute information of the patient node (such as age, gender, medical history, etc.) and the psychological disorder node connected to it, the node representation of the graph neural network model is used to predict the patient's psychological disorder risk and obtain a personalized risk assessment result. Combined with the patient's psychological disorder risk assessment results and the treatment method nodes in the knowledge graph, the reasoning ability of the graph neural network is used to generate personalized intervention strategies for patients, including recommending appropriate treatment methods and lifestyle adjustment suggestions.

[0060] In this embodiment, further, visualization technology can be used to intuitively display the entities, relationships and attributes in the knowledge graph in the form of nodes and edges, and an interactive interface can be designed to allow users to click, drag and other operations to realize query and browsing of the knowledge graph content.

[0061] Specifically, graph database technologies such as Neo4j can be used to store entity, relationship and attribute information in the knowledge graph, and efficient knowledge retrieval and query can be achieved through the Cypher query language. Obtain entity, relationship and attribute data in the knowledge graph, use visualization libraries such as D3.js, design appropriate layout algorithms, generate visualization results of the knowledge graph, use nodes to represent entities, edges to represent relationships between entities, and use visual encodings such as color and size to represent the attributes of entities and relationships. In the visualization interface, an interactive query function is provided. Users can query and retrieve content in the knowledge graph by searching keywords, clicking nodes and edges, etc. If the user clicks on an entity node, other entities and relationships related to the entity are highlighted. According to the user's query conditions, knowledge graph embedding technologies such as TransE are used to find entities with semantic similarity to the query conditions in the vector space, sort them according to similarity, return the top-k related entities, and present them to the user. For complex user queries, question-answering technology based on knowledge graphs is used to convert natural language queries into structured query statements through semantic parsing, perform reasoning and matching on the knowledge graph, and obtain query results.

[0062] In the visualization interface, a knowledge card function is provided. When a user clicks on an entity, the knowledge card of the entity pops up to display the detailed attribute information of the entity, as well as the entities and relationships directly connected to it. The knowledge graph embedding technology is adopted to provide a related entity recommendation function in the visualization interface. When a user views an entity, other entities with similar semantics are calculated and recommended based on the embedding vector of the entity to help users discover potential related knowledge. Through log analysis and user feedback collection, the user's query behavior and interest preferences are analyzed, and recommendation algorithms such as collaborative filtering are adopted to provide users with personalized knowledge recommendations to improve the pertinence and satisfaction of knowledge acquisition. The query logs and user feedback of the knowledge graph are mined and analyzed to identify the common query patterns and knowledge needs of users, optimize the structure and content of the knowledge graph, and iteratively optimize the design and interaction of the visualization interface to improve the user experience. The dynamic update unit 150 is used to establish a dynamic update mechanism, use a large language model to continuously learn new knowledge, and is regularly reviewed and optimized by a cross-domain expert team to ensure the timely update of the knowledge graph.

[0063] In this embodiment, a regular and comprehensive update strategy for the knowledge graph can be formulated, such as monthly or quarterly updates, to instantly update hot issues and emerging research results to ensure the timeliness and accuracy of the knowledge graph.

[0064] Among them, the newly added knowledge is verified and quality controlled, and the scientificity and rigor of the new knowledge can be ensured through consistency checks with the existing knowledge graph and review and demonstration by domain experts.

[0065] When evaluating the performance of graph neural network models in predicting the risk of psychological disorders in patients with cardiovascular disease and generating personalized intervention strategies, methods such as cross-validation can be used to calculate indicators such as the model's accuracy, precision, recall rate, and F1 value, and continuously optimize the model to improve its effectiveness in practical applications.

[0066] In summary, the embodiment of the present invention first obtains and standardizes the data related to psychological disorders of cardiovascular patients from multiple data sources, uses natural language processing technology to extract and classify core entities, and then uses the BERT+BiLSTM-CRF model to extract the relationship between entities and construct a knowledge graph of psychological disorders. Then, based on the graph, a graph neural network model is used for reasoning to explore potential comorbidity relationships and personalized intervention strategies. The knowledge graph is then evaluated and improved in combination with evidence-based medicine and expert feedback, and finally a visualization result is generated and an interactive query interface is provided. By integrating multi-source data, deep learning and knowledge graph technology, the present invention achieves accurate analysis and personalized intervention of psychological disorders in patients with cardiovascular diseases, providing intelligent support for clinical decision-making.

[0067] Some preferred embodiments of the present invention are further described below:

[0068] In the mining unit 140 of the above embodiment, in the knowledge graph composed of patient-disease-symptom-treatment, rule-based reasoning technology can discover new disease comorbidity patterns and etiology mechanisms, but in actual applications, there may be conflicts and inconsistencies between different rules. For example, some rules may indicate that a patient suffers from anxiety and depression at the same time, while other rules believe that the two diseases cannot occur at the same time. At this point, how to reasonably resolve conflicts between rules and ultimately give consistent reasoning results has become a key technical issue.

[0069] To this end, in a preferred embodiment of the present invention, the mining unit 140 is specifically used for:

[0070] First, a set of rules related to disease comorbidity is obtained from the knowledge graph. For each rule, the brightness value is calculated according to its usage frequency and accuracy in historical reasoning. The brightness value calculation formula is: brightness value = α×usage frequency + β×accuracy, where α and β are preset weight coefficients.

[0071] Then, the rule set is sorted in descending order according to the calculated brightness value to obtain a rule sequence arranged from high to low according to the rule brightness. The rule sequence is stored in the form of an ordered list, and each element contains the rule content and the corresponding brightness value.

[0072] Among them, firstly, traverse each rule in the rule set, calculate the brightness value of each rule, and obtain a rule brightness value list. Then, according to the rule brightness value list, use the quick sort algorithm to sort the rule set in descending order to obtain a rule sequence arranged from high to low according to the rule brightness. Then create an empty ordered list to store the sorted rule sequence. Then traverse the sorted rule sequence, obtain the rule content and the corresponding brightness value for each rule, and add the rule content and the brightness value as an element to the ordered list. Then determine whether the number of elements in the ordered list is equal to the number of rules in the rule set. If they are equal, output the ordered list as the final rule sequence; otherwise, re-execute the above steps until the number of elements in the ordered list is equal to the number of rules in the rule set. Finally, for each element in the ordered list, extract the rule content and the corresponding brightness value, store the rule content and the brightness value in two independent arrays respectively, obtain the rule content array and the brightness value array, use the rule content array and the brightness value array as the final storage form of the rule sequence, and output the rule sequence.

[0073] Next, in the process of disease comorbidity pattern discovery, high-brightness rules are selected from the rule sequence and applied to the knowledge graph. If the data in the knowledge graph meets the conditions of multiple rules at the same time, the rule with the highest brightness value is prioritized for reasoning.

[0074] In this embodiment, after obtaining the rule sequence, the rule with the highest brightness value is selected from the rule sequence, and the rule condition of the rule is matched with the data in the knowledge graph. If the knowledge graph data meets the rule condition, the rule is used to infer the knowledge graph to obtain the inference result; otherwise, the next step is entered: the matched rules are removed from the rule sequence. If the rule sequence is empty, all inference results are output and the algorithm ends; otherwise, the step is returned to: the rule condition of the rule is matched with the data in the knowledge graph.

[0075] After traversing the rule sequence, the inference results are statistically analyzed to obtain the comorbidity patterns and association strengths between diseases. Then, based on the comorbidity patterns and association strengths, the association rule mining algorithm is used to further mine the association rules between diseases. Finally, the disease comorbidity patterns, association strengths, and association rules are added as knowledge to the disease comorbidity knowledge graph to complete the discovery of disease comorbidity patterns and the update and iteration of the knowledge graph.

[0076] Next, when a conflict is detected between rules, the security policy is initiated according to the preset conflict resolution priority to resolve the conflicting rules and retain the rules that comply with the security policy. The conflict resolution priority is determined based on the brightness value, timeliness and scope of application of the rules. The security policy includes three methods: rule merging, rule replacement and rule deletion.

[0077] Specifically, first obtain the rule conflict resolution priority preset in the system. The priority is determined based on the attributes of the rule such as brightness value, timeliness and scope of application, and form a rule priority sorting list. Then monitor the rules in the system in real time, extract the brightness value, timeliness, scope of application and other attributes of each rule, input the attribute value into the machine learning model for analysis, and determine whether there is a rule conflict. If a rule conflict is detected, the brightness value, timeliness and scope of application of the conflicting rules are compared according to the rule priority sorting list to determine the rule with a higher priority. Then, according to the preset resolution security policy, the conflicting rules are processed by rule merging, rule replacement or rule deletion, and the rules with higher priority and in compliance with the security policy are retained.

[0078] During the rule merging process, natural language processing technology is used to analyze the text content of conflicting rules, extract key information, use clustering algorithms to classify similar rules, and automatically generate new merged rules.

[0079] During the rule replacement process, the correlation between conflicting rules and security policies is analyzed through deep learning algorithms, and replacement plans are automatically recommended to replace conflicting rules with lower priority with rules that comply with security policies.

[0080] During the rule deletion process, a reinforcement learning algorithm is used to autonomously learn and decide on the conflicting rules that need to be deleted based on historical data and security policy requirements, continuously optimize the rule base, and improve the security and efficiency of the system.

[0081] Next, the rule set processed by the security strategy is used to perform reasoning operations in the knowledge graph to obtain preliminary reasoning results of the disease comorbidity pattern. The rules used in each reasoning step and their brightness values ​​are recorded. The preliminary reasoning results include possible disease comorbidity relationships, relationship strengths, and confidence levels.

[0082] Specifically, firstly, a set of disease comorbidity-related rules that have been safely processed is obtained to construct a disease comorbidity knowledge graph. Then, according to the constructed disease comorbidity knowledge graph, a rule-based reasoning algorithm, such as the Rete algorithm, is used to perform reasoning operations.

[0083] During the reasoning operation, the rules used in each reasoning step and the brightness value of the rule are recorded. The brightness value indicates the importance of the rule. Through the reasoning operation, the preliminary results of the disease comorbidity pattern are obtained, including the possible comorbidity relationship between diseases, the strength of the comorbidity relationship, and the confidence of the reasoning results.

[0084] The preliminary inference results are post-processed, and the threshold filtering method is used to screen out the strongly correlated disease comorbidity relationships according to the preset relationship strength threshold and confidence threshold. The screened disease comorbidity relationships are compared and verified with medical field knowledge, and the inference results are adjusted and optimized in combination with clinical practice to obtain the final disease comorbidity model. The inference results of the disease comorbidity model are presented in a graphical way, with the rule brightness information of each inference step, which is convenient for medical experts to analyze and interpret.

[0085] Finally, the reasoning process, the rule set used, the rule brightness value, the application of the resolution strategy, the preliminary reasoning results, the final reasoning results and the confidence score are stored in the reasoning log of the knowledge graph to support subsequent rule optimization; the confidence score is calculated based on the rule brightness value, the number of rules and the reasoning depth used in the reasoning process.

[0086] Specifically, firstly, a rule set used in the inference process is obtained, a brightness value of each rule is determined, and the rule set and the brightness value are stored in the inference log.

[0087] Then determine whether the reasoning process uses a resolution strategy. If a resolution strategy is used, the type and parameters of the resolution strategy are also recorded in the reasoning log.

[0088] Next, the preliminary reasoning result and the final reasoning result obtained in the reasoning process are obtained, and both results are stored in corresponding fields of the reasoning log.

[0089] Next, the comprehensive confidence score is calculated using the weighted average method based on the rule brightness value, number of rules, and reasoning depth recorded in the reasoning log.

[0090] Next, the calculated confidence score is also stored in the inference log and associated with the inference result.

[0091] Next, statistical analysis is performed on the data in the reasoning log to determine the correlation between the frequency of use of different rules and the accuracy of the reasoning results.

[0092] Finally, based on the correlation analysis between the frequency of rule usage and the accuracy of the results, the rule set is optimized, the brightness value of the rule is adjusted or the rule is modified to form an optimized rule set and update the knowledge graph.

[0093] In summary, this embodiment, through the integration of rule brightness and elimination safety strategy, can take into account the importance and consistency of rules when discovering disease comorbidity patterns, and improve the reliability and accuracy of reasoning results. At the same time, this also provides new ideas for the construction and optimization of knowledge graphs, which helps to further explore the hidden associations and rules in medical data.

[0094] The description of the above embodiments is only used to help understand the technical solutions and core ideas of the present application. Ordinary technical personnel in this field should understand that they can still modify the technical solutions recorded in the aforementioned embodiments, or replace some of the technical features therein with equivalents. However, these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A device for intervening psychological disorders in patients with cardiovascular disease based on knowledge graph, characterized in that: include: A multi-source heterogeneous data acquisition unit, used to acquire multi-source heterogeneous data related to cardiovascular diseases and psychological disorders; A preprocessing unit, used to preprocess the multi-source heterogeneous data, extract key information, and construct an initial knowledge graph; An expansion and optimization unit, used to expand and optimize the initial knowledge graph using a pre-trained large language model to construct a knowledge graph containing entities of diseases, symptoms, treatments, and patient mental states and their relationships; A mining unit is used to mine potential disease comorbidity relationships and personalized intervention strategies contained in the knowledge graph through knowledge reasoning and graph neural network technology, so as to implement intervention on psychological disorders of cardiovascular patients; wherein the knowledge graph is presented through a visual and interactive interface to support user query, reasoning and knowledge acquisition; the mining unit is specifically used to: Apply path-based reasoning and rule-based reasoning techniques to the knowledge graph of patient-disease-symptom-treatment to discover new disease comorbidity patterns and etiological mechanisms; Use graph neural networks to perform representation learning and node classification on knowledge graphs, predict patients' risk of psychological disorders, and provide personalized psychological intervention suggestions; apply path-based reasoning and rule-based reasoning techniques on the knowledge graph composed of patients-diseases-symptoms-treatments to discover new disease comorbidity patterns and etiology mechanisms, including: obtaining a set of rules related to disease comorbidity from the knowledge graph, and calculating the brightness value of each rule based on its frequency of use and accuracy in historical reasoning. The brightness value calculation formula is: brightness value = α×frequency of use + β×accuracy, where α and β are preset weight coefficients; in the process of disease comorbidity pattern discovery, select high-brightness rules from the rule sequence and apply them to the knowledge graph. If the data in the knowledge graph meets the conditions of multiple rules at the same time, the rule with the highest brightness value is selected for reasoning first; When conflicts between rules are detected, the security policy is initiated according to the preset conflict resolution priority to resolve the conflicting rules and retain the rules that meet the security policy. The conflict resolution priority is determined based on the brightness value, timeliness and scope of application of the rules. The security policy includes three methods: rule merging, rule replacement and rule deletion. The rule set processed by the security policy is used to perform reasoning operations in the knowledge graph to obtain preliminary reasoning results of the disease comorbidity pattern, and the rules and their brightness values ​​used in each reasoning step are recorded. The preliminary reasoning results include possible disease comorbidity relationships, relationship strengths and confidence levels. The reasoning process, the rule set used, the rule brightness value, the application of the resolution strategy, the preliminary reasoning results, the final reasoning results and the confidence score are stored in the reasoning log of the knowledge graph to support subsequent rule optimization. The confidence score is calculated based on the rule brightness value, number of rules and reasoning depth used in the reasoning process. The dynamic update unit is used to establish a dynamic update mechanism, use the large language model to continuously learn new knowledge, and is regularly reviewed and optimized by a cross-domain expert team to ensure timely updating of the knowledge graph.

2. The device for intervening in psychological disorders of cardiovascular patients based on knowledge graph according to claim 1, characterized in that: The multi-source heterogeneous data acquisition unit is specifically used for: Acquire structured, semi-structured and unstructured heterogeneous data from evidence-based guidelines, clinical practice guidelines, medical databases, authoritative literature, online platforms and clinical health manuals; The acquired heterogeneous data is cleaned, integrated and standardized to extract key information elements, including patient information, disease information, symptom manifestations, treatment plans, and psychological state descriptions, thereby providing a high-quality data foundation for subsequent knowledge graph construction.

3. The device for intervening in psychological disorders of cardiovascular patients based on knowledge graph according to claim 2, characterized in that: The pre-processing unit is specifically used for: Define and classify core entities such as patients, diseases, symptoms, treatments, and psychological states; Use natural language processing techniques to perform named entity recognition and relation extraction to identify entities and determine the semantic relationships between entities; A standardized knowledge graph is constructed based on the extracted entities and relationships, forming nodes for diseases, symptoms, treatments, and patient psychological states, as well as multiple associations between nodes.

4. The device for intervening in psychological disorders of cardiovascular patients based on knowledge graph according to claim 1, characterized in that: The expansion optimization unit is specifically used for: On unstructured text data such as medical literature, guidelines, and case reports, we use pre-trained large language models for continuous learning to capture deep semantic knowledge in the fields of cardiovascular disease and psychological disorders. Pre-trained large language models include BERT and GPT. Through transfer learning, the knowledge learned by the large language model is transferred to the initial knowledge graph, expanding the entity, relationship and attribute information therein, optimizing the structure and semantic representation of the knowledge graph, and improving the comprehensiveness and accuracy of the knowledge graph.

5. The device for intervening in psychological disorders of cardiovascular patients based on knowledge graph according to claim 1, characterized in that: The knowledge graph is presented through a visual and interactive interface to support user query, reasoning and knowledge acquisition, including: Design a visual layout based on the knowledge graph, and intuitively display the complex relationships between diseases, symptoms, treatments, and patients' psychological states through nodes, edges, and color visual elements; Develop interactive functions to allow users to explore the knowledge graph through keyword search, semantic query, and intelligent question-answering, achieving accurate and convenient knowledge retrieval and acquisition; Embedded with a knowledge reasoning engine, users can ask questions in natural language, and the system can infer answers from the knowledge graph based on the questions to support decision-making.

6. The device for intervening in psychological disorders of cardiovascular patients based on knowledge graph according to claim 1, characterized in that: The dynamic update unit is specifically used for: Design knowledge graph update process and quality assessment indicators to ensure the standardization and controllability of updates; Continuously acquire the latest research results in the fields of cardiovascular disease and psychology, use large language models to automatically extract structured knowledge, and dynamically expand the knowledge graph; A review team composed of multidisciplinary experts in cardiovascular disease, psychology, and knowledge engineering will be formed to review and verify the new knowledge to ensure the scientificity and authority of the knowledge graph update.

Images