Man-machine conversation data processing method for chronic disease management and related equipment

Abstract

The invention discloses a man-machine conversation data processing method for chronic disease management and related equipment, and relates to the field of data processing. The method comprises the steps of firstly obtaining a chronic disease consultation intention instruction of a user; secondly, identifying semantic key elements such as medical terms, symptom description, time information and behavior actions by using a natural language processing technology, and generating a semantic relation graph after determining the incidence relation of the elements; determining main intention information corresponding to the chronic disease consultation intention instruction based on the graph; and then determining the intention confidence coefficient of the main intention information, if the intention confidence coefficient is greater than or equal to a set confidence coefficient threshold value, determining a chronic disease consultation answer according to the main intention information, and feeding back the answer to the user. Through the processing of the steps of the method, user requirements can be accurately grasped, and reliable and accurate consultation answers can be provided, so that the user experience is enhanced, and the service quality of chronic disease management is improved.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a human-computer dialogue data processing method and related equipment for chronic disease management. Background Art

[0002] At present, chronic diseases (or chronic diseases) have become one of the serious problems affecting public health. The chronic disease management platform realizes personalized management of chronic disease patients by integrating advanced information technologies such as electronic health records, mobile health monitoring, data analysis, etc. The chronic disease management platform can provide real-time data tracking, automated reminders, customized health advice and telemedicine consultation, thereby improving the efficiency and quality of chronic disease management, enhancing patients' self-management ability, and promoting the continuity and coordination of medical services.

[0003] Among the related technologies, many chronic disease management platforms have integrated human-computer dialogue functions, providing users and patients with a convenient and efficient way of interaction. They can understand and process patients' problems and needs through natural language, provide instant feedback and personalized suggestions, thereby enhancing patient engagement, improving treatment compliance, and improving the overall efficiency and effectiveness of chronic disease management.

[0004] However, the current chronic disease management platform still has problems with precision in understanding patient intentions, resulting in inaccurate consultation answers output to users by human-computer dialogue, affecting the user experience and reducing the service quality of chronic disease management. Summary of the invention

[0005] In view of the above-mentioned technical problems and defects, the purpose of the present invention is to provide a human-computer dialogue data processing method and related equipment for chronic disease management, which can accurately understand the user's chronic disease consultation intention instructions and provide the user with more accurate consultation answers, so as to enhance the user experience and improve the service quality of chronic disease management.

[0006] To achieve the above-mentioned objectives, in a first aspect, the present invention provides a human-computer dialogue data processing method for chronic disease management, comprising: obtaining a chronic disease consultation intention instruction issued by a user; using natural language processing technology to identify and process the chronic disease consultation intention instruction to obtain multiple semantic key elements, the semantic key elements including medical terms, symptom descriptions, time information and behavioral actions; determining the element association relationship between the multiple semantic key elements; generating a semantic relationship graph of the chronic disease consultation intention instruction based on the element association relationship, the semantic relationship graph including multiple nodes and node connection relationships between the nodes, the nodes being used to represent the semantic key elements; determining the main intention information corresponding to the chronic disease consultation intention instruction based on the semantic relationship graph; determining the intention confidence of the main intention information; when the intention confidence is greater than or equal to the set confidence threshold, determining the chronic disease consultation answer based on the main intention information; and feeding back the chronic disease consultation answer to the user.

[0007] The present invention adopts the above method, by obtaining the chronic disease consultation intention instruction issued by the user, using natural language processing technology to identify semantic key elements including medical terms, symptom descriptions, time information and behavioral actions, and determine the element association relationship between them, and then generate a semantic relationship graph. This series of operations can clarify and structure the user's vague consultation intention to accurately understand the user's chronic disease consultation intention. Based on this, the main intention information and its intention confidence are determined. Only when the confidence reaches the set threshold is the chronic disease consultation answer determined and fed back to the user. This effectively guarantees the accuracy and reliability of the answer provided, making the entire chronic disease management human-computer dialogue process more scientific and orderly, and can better meet the actual needs of users, provide users with more accurate consultation answers, enhance user experience, and improve the service quality of chronic disease management.

[0008] In combination with some embodiments of the first aspect, in some embodiments, the main intention information corresponding to the chronic disease consultation intention instruction is determined based on the semantic relationship graph, including: determining, according to the semantic relationship graph, that the chronic disease consultation intention instruction corresponds to multiple possible intention information; determining, in each possible intention information, the information weight of the node and the relationship closeness of the connection relationship between each node; determining the criticality score of each possible intention information based on the information weight and the relationship closeness; and determining the main intention information from multiple possible intention information based on the criticality score.

[0009] The technical solution of the above embodiment is adopted to determine the main intent information by determining multiple possible intent information, as well as the information weights and connection closeness of the nodes therein, and then determining the key score based on these factors. This enables a comprehensive analysis of possible intents from multiple angles when processing chronic disease consultation intention instructions, improves the accuracy and comprehensiveness of determining the main intent information, avoids the limitations of single factor judgment, lays the foundation for the subsequent accurate provision of chronic disease consultation answers, better meets the actual needs of users, and enhances the user experience in human-computer dialogue for chronic disease management.

[0010] In combination with some embodiments of the first aspect, in some embodiments, the criticality score of each possible intention information is determined based on the information weight and the relationship closeness, including: determining the information entropy based on the node information weight vector of the information weight, and the information entropy is used to represent the uncertainty of the node information weight vector; determining the weighted centrality measure based on the node information weight vector and the node relationship closeness matrix of the relationship closeness, and the weighted centrality measure is used to represent the importance of the node in the semantic relationship graph; determining each criticality score based on the information entropy and the weighted centrality measure.

[0011] The technical solution of the above embodiment is adopted, and information entropy is used to represent the uncertainty of the node information weight vector, and the weighted centrality measurement represents the importance of the node in the semantic relationship graph, and then the criticality score of each possible intent information is determined based on the two. This method can more deeply explore the key information in the semantic relationship graph and quantify the impact of different factors on the importance of intent information. By introducing information theory and graph analysis methods, a scientific means is provided for accurately evaluating the criticality of possible intent information, which helps to more accurately determine the main intent information and improve the pertinence and effectiveness of answers to chronic disease consultations.

[0012] In combination with some embodiments of the first aspect, in some embodiments, each criticality score is determined according to information entropy and weighted centrality measurement, including: calculating the criticality score according to a preset criticality score formula, and the criticality score formula includes:

[0013]

[0014]

[0015] Among them, S i is the criticality score of the i-th possible intention information, W i is the node information weight vector of the i-th possible intention information, C i is the node relationship closeness matrix of the i-th possible intention information, λ is the balance parameter, H(W i ) represents information entropy, D(C i ,W i ) represents the weighted centrality measure, n i is the number of nodes in the i-th possible intention information, ϵ is the experimental parameter, j and k are the node numbers, m is the index variable, and W ij and W ik Respectively represent the information weights of the jth node and the kth node in the i-th possible intention information, C ijk represents the closeness of the relationship between the jth node and the kth node, C ijm Represents the closeness of the relationship between the j-th node and the m-th node.

[0016] By adopting the technical solution of the above embodiment, the role and relationship of each parameter in the calculation are clarified through a specific criticality scoring formula. The balance parameter allows the information entropy and weighted centrality measurement to adjust the weights according to actual conditions when calculating the criticality score, adapting to different application scenarios and data characteristics. The various parameters in the formula can fully consider multiple aspects such as node information weight, relationship closeness, and uncertainty, providing a calculable and adjustable method for accurately determining the criticality of possible intent information, further improving the accuracy and reliability of determining the main intent information.

[0017] In combination with some embodiments of the first aspect, in some embodiments, determining the intent confidence of the main intent information includes: determining a confidence evaluation factor of the main intent information, the confidence evaluation factor including the completeness of the main intent information, the clarity of the semantic key elements, and the degree of match between the main intent information and known chronic disease information; determining the intent confidence of the main intent information based on the confidence evaluation factor.

[0018] By adopting the technical solution of the above embodiment, the confidence evaluation factors of the main intent information are determined, including completeness, clarity of semantic key elements, and matching degree with known chronic disease information, and the reliability of the main intent information is evaluated in multiple dimensions. This helps to fully understand the quality of the main intent information and avoid erroneous results caused by inaccurate judgment of a single factor. By comprehensively considering these factors, the credibility of the main intent information can be determined more accurately, providing a strong guarantee for providing high-quality answers to chronic disease consultations and enhancing users' trust in the chronic disease management human-computer dialogue system.

[0019] In combination with some embodiments of the first aspect, in some embodiments, determining the intent confidence of the main intent information according to the confidence evaluation factor includes: calculating the intent confidence according to a preset confidence evaluation formula, and the confidence evaluation formula includes:

[0020]

[0021]

[0022]

[0023] Where T represents the confidence of intent, I(W,P) represents the integrity measure of the main intent information, L(W) represents the clarity of the key semantic elements, Q(M) represents the matching degree between the main intent information and the known chronic disease information, H(W) represents the information entropy of the main intent information, W is the node information weight vector, P is the occurrence probability of the node, M is the matching degree matrix between the main intent information and the known chronic disease information, n is the number of nodes of the main intent information, j is the node sequence number, and W jrepresents the information weight of the jth node, P j represents the probability of the jth node appearing in the main intent information, M j represents the matching degree between the jth node and the known chronic disease information, and α, β and γ are balance parameters.

[0024] Using the technical solution of the above embodiment, the integrity measure of the main intent information, the clarity of the semantic key elements, and the matching degree with the known chronic disease information are quantified and calculated through a specific confidence evaluation formula to obtain the intent confidence. The various parameters and calculation methods in the formula can scientifically reflect the performance of the main intent information in different aspects, and provide an operational method for accurately evaluating the intent confidence. The setting of the balance parameter allows the weights of different evaluation factors to be flexibly adjusted when calculating the confidence, adapting to different situations and needs, and improving the accuracy and adaptability of the confidence evaluation.

[0025] In combination with some embodiments of the first aspect, in some embodiments, after determining the intention confidence of the main intention information, it also includes: when the intention confidence is less than the confidence threshold, sending an intention clarification prompt message to the user; receiving a secondary chronic disease consultation intention instruction issued by the user, the secondary chronic disease consultation intention instruction is generated by the user based on the intention clarification prompt information; determining the main intention information based on the secondary chronic disease consultation intention instruction.

[0026] By adopting the technical solution of the above embodiment, when the confidence of the intention is less than the threshold, an intention clarification prompt message is sent to the user, and the user's secondary chronic disease consultation intention instruction is received to determine the main intention information. This process can obtain more information through interaction with the user when the initial main intention information is not reliable enough, thereby improving the accuracy of the main intention information. This interactive mechanism enhances the adaptability and user-friendliness of the system, avoids incorrect answers caused by unclear initial information, and improves the service quality and user satisfaction of the chronic disease management human-computer dialogue system.

[0027] In a second aspect, an embodiment of the present invention provides an electronic device, comprising: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code includes computer instructions, and the one or more processors call the computer instructions to enable the electronic device to execute the method described in the first aspect or the second aspect, and any possible implementation method of the first aspect or the second aspect.

[0028] In a third aspect, the present invention provides a computer-readable storage medium comprising instructions, which, when executed on the electronic device, causes the electronic device to execute the method described in the first aspect or the second aspect, and any possible implementation of the first aspect or the second aspect.

[0029] In a fourth aspect, the present invention provides a computer program product comprising instructions, which, when the computer program product is run on the electronic device, enables the electronic device to execute the method described in the first aspect or the second aspect, and any possible implementation of the first aspect or the second aspect.

[0030] It is understandable that the electronic device provided in the second aspect, the storage medium provided in the third aspect, and the computer program product provided in the fourth aspect are all used to execute the method provided in the present invention. Therefore, the beneficial effects that can be achieved can refer to the beneficial effects in the corresponding method, which will not be repeated here.

[0031] One or more technical solutions provided by the present invention have at least the following technical effects or advantages: 1. Accurately understand user intentions and improve the accuracy of consultation answers: Use natural language processing technology to identify and process chronic disease consultation intention instructions, obtain semantic key elements and construct a semantic relationship graph, and then determine the main intent information through multi-step analysis. For example, the information weight and relationship closeness of the nodes in each possible intent information will be comprehensively considered to determine the criticality score, so as to accurately screen out the main intent information. Subsequently, the intent confidence will be determined from multiple aspects such as completeness, clarity, and matching with known chronic disease information to further verify the reliability of the main intent information. Through such in-depth and multi-angle analysis, the accuracy of understanding the user's true intentions is greatly improved, thereby providing users with more accurate and reliable chronic disease consultation answers that are more in line with their needs.

[0032] 2. Scientifically quantify evaluation factors to enhance objectivity of judgment: Whether it is the determination of criticality scores or the calculation of intent confidence, corresponding scientific quantitative formulas are introduced. For example, in the criticality scoring formula, concepts such as information entropy and weighted centrality measurement are involved. Through clear parameters and calculation methods, the uncertainty of the node information weight vector and the importance of the node in the semantic relationship graph are comprehensively considered. In the calculation of intent confidence, factors such as integrity measurement, clarity of semantic key elements, and matching degree with known chronic disease information are also quantified according to specific formulas. These quantitative methods avoid subjective and arbitrary judgments, making the evaluation of intent information more objective and scientific, and ensuring the rigor of the entire chronic disease management dialogue data processing process.

[0033] 3. Equipped with an interactive feedback mechanism to improve user experience: When the intent confidence of the determined main intent information is lower than the set threshold, the server will send an intent clarification prompt to the user, receive the secondary chronic disease consultation intention instruction generated by the user based on the prompt, and then re-determine the main intent information. This interactive feedback mechanism fully takes into account the situation where the initial information may be inaccurate and incomplete, and provides users with opportunities to supplement and correct, so that the chronic disease management platform can better adapt to the expression habits of different users and complex and diverse consultation scenarios. It avoids erroneous replies caused by information errors, allowing users to truly feel the humanity and flexibility of the system, and effectively improves the user experience in the human-computer dialogue process of chronic disease management. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] The drawings herein are incorporated into and constitute a part of the specification, showing embodiments consistent with the present invention, and together with the specification, are used to explain the principles of the present invention. Obviously, the drawings described below are only some embodiments of the present invention, and for those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative work. In the drawings: Figure 1 It is a schematic diagram of the architecture of a chronic disease management platform according to an embodiment of the present invention; Figure 2 It is a flow chart of a method for processing human-computer dialogue data for chronic disease management according to an embodiment of the present invention; Figure 3 It is a flowchart of another method for processing human-computer dialogue data for chronic disease management according to an embodiment of the present invention; Figure 4 It is a schematic diagram of the architecture of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0035] The terms used in the following embodiments of the present invention are only for the purpose of describing specific embodiments, and are not intended to be limiting of the present invention. As used in the specification of the present invention, the singular expressions "a", "a", "above", "the" and "this" are intended to also include plural expressions, unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used in the present invention refers to any or all possible combinations comprising one or more of the listed items.

[0036] In the following, the terms "first" and "second" are used for descriptive purposes only and should not be understood as implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the embodiments of the present invention, unless otherwise specified, the meaning of "plurality" is two or more.

[0037] It should also be noted that, unless otherwise clearly specified and limited, in the embodiments of the present invention, the terms such as "setting" and "connection" should be understood in a broad sense. For example, "connection" can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, it can be the internal connection of two components; it can be a wired communication connection or a wireless communication connection. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to the specific circumstances. The embodiments of the present invention are described in detail below.

[0038] The embodiment of the present invention provides a human-computer dialogue data processing method for chronic disease management, which provides users with more accurate consultation answers, enhances the user experience, and improves the service quality of chronic disease management. First, after obtaining the chronic disease consultation intention instruction issued by the user, natural language processing technology is used for recognition and processing, and semantic key elements including medical terms, symptom descriptions, time information, and behavioral actions are accurately extracted. Then, the relationship between these key elements is deeply analyzed to construct a semantic relationship graph. The semantic relationship graph clearly shows each node (i.e., semantic key elements) and the connection relationship between them, laying a solid foundation for accurately grasping the user's intention. After determining the main intention information based on the semantic relationship graph, the intention confidence is determined. When the confidence is greater than or equal to the set threshold, the chronic disease consultation answer is determined based on the main intention information. This process ensures that the answer provided to the user has high accuracy and reliability.

[0039] In this way, users can get accurate answers to their specific questions, avoiding vague or inaccurate responses, which greatly enhances the user experience. At the same time, it improves the service quality of chronic disease management, making it more competitive in the field of chronic disease management and providing users with better quality and more efficient services.

[0040] The method of this embodiment can be used in a chronic disease management platform, such as Figure 1 As shown, the chronic disease management platform includes a server and a user. The user conducts a human-computer dialogue with the server through the user to obtain information related to chronic diseases. The server includes a chronic disease management server (hereinafter referred to as the server).

[0041] Combine the following Figure 2 The human-computer dialogue data processing method of this embodiment is specifically described, and includes the following steps: Step 201, obtaining a chronic disease consultation intention instruction issued by a user.

[0042] The server first establishes a connection channel with the user terminal, which can achieve stable data transmission through the Internet protocol. When the user enters the consultation content on the user terminal of the chronic disease management platform (such as mobile applications, web interfaces, etc.), the user terminal will convert the user's text, voice, etc. input into a specific data format and send this data to the server through the network connection.

[0043] The server continuously monitors requests and data transmissions from the user. Once the user's input information is received, it will be identified as a chronic disease consultation intention instruction. At the same time, the server will perform preliminary verification and processing on the received data to ensure the integrity and validity of the data, so that these instructions can be accurately analyzed and processed later, providing users with high-quality chronic disease consultation services.

[0044] Step 202: Use natural language processing technology to identify and process the chronic disease consultation intention instructions to obtain multiple semantic key elements.

[0045] Among them, semantic key elements include medical terms, symptom descriptions, time information and behavioral actions.

[0046] Medical terms are words or phrases with specific professional meanings in the medical field, which are used to accurately describe concepts such as human body structure, physiological functions, disease names, treatment methods, and drugs, such as "hypertension", "coronary heart disease", "insulin", etc.

[0047] Symptom description refers to the patient's specific description of his or her own physical discomfort or abnormal sensations, usually including subjective feelings of pain, discomfort, functional abnormality, such as "splitting headache", "chest tightness and shortness of breath", "blurred vision", etc.

[0048] Time information is a specific description of the time point, time period, etc. of events or conditions related to chronic diseases. It is used to clarify the time context of the development of the disease, treatment process, etc., such as "last night", "a month ago", "for three days", etc.

[0049] Behavioral actions refer to the specific actions or activities taken by patients or relevant personnel in the process of chronic disease management, including treatment behaviors, living habits, etc., such as "taking antihypertensive drugs", "insisting on exercising", "regular check-ups", etc.

[0050] In this step, the server will first perform text preprocessing on the received chronic disease consultation intention instructions. This includes removing noise data, such as the irregular use of special characters and punctuation marks, and performing word segmentation operations to divide the instruction text into independent word units.

[0051] Next, the server uses natural language processing technologies such as named entity recognition to identify medical terms. Pre-trained medical terminology dictionaries and models are used to match and classify words in the text to determine which are specific disease names, drug names, and other medical terms. For the identification of symptom descriptions, the server uses pattern matching and semantic analysis methods. For example, it looks for common words and expressions that describe physical discomfort, such as "headache", "fatigue", "dyspnea", etc., and combines the context to determine whether it is indeed a symptom description.

[0052] In terms of identifying time information, the server uses a time expression recognition algorithm. It can recognize clear time words such as "yesterday", "a week ago", "recently", and expressions containing specific dates and times, and convert them into a standard time format for subsequent processing. For the recognition of behavioral actions, the server analyzes verbs and verb-object phrases in the text. For example, in expressions such as "see a doctor", "take medicine", and "measure blood pressure", it can accurately identify behavioral actions such as "see a doctor", "take medicine", and "measure". Through these comprehensive natural language processing technologies, the server can extract semantic key elements such as medical terms, symptom descriptions, time information, and behavioral actions from chronic disease consultation intention instructions.

[0053] Step 203: determine the element association relationship between multiple semantic key elements.

[0054] Specifically, natural language processing technology, especially named entity recognition (NER) technology, can be used to extract key chronic disease medical terms, symptom descriptions, time information, behavioral actions and other entities from the text of patient or user consultation. These entities are semantic key elements.

[0055] Next, relationship extraction techniques are applied to identify the relationships between these entity pairs, such as cause-effect relationships, relationship between symptoms and diseases, relationship between treatments and drugs, etc. This can be achieved through supervised learning models that have been previously trained with a large amount of labeled data.

[0056] The context in which the entities appear is then analyzed to determine the implicit relationships between them. For example, certain symptoms may have a strong association with a specific disease, and this association can be inferred from the patient's description.

[0057] Finally, medical rules and logic are applied, which may be based on clinical guidelines or the knowledge of medical experts, to determine the reasonable associations between entities (semantic key elements). This process ensures that the platform can accurately capture the key information in patient consultations and understand their deep meaning. In this way, the element association relationships between semantic key elements are obtained.

[0058] In some embodiments, the element association relationship between multiple semantic key elements can also be determined by constructing a multi-dimensional semantic understanding architecture.

[0059] The multi-dimensional semantic understanding architecture is a natural language processing technology framework that deeply understands and interprets the meaning of text data by comprehensively considering information at multiple levels in the language, such as vocabulary, syntax, and semantics. This architecture can identify and parse key elements in multiple dimensions such as medical terms, symptom descriptions, time information, and behavioral actions in the text, and place them in a broader context to determine the complex relationships and interactions between them, thereby providing more accurate intent recognition and more effective user interaction for the chronic disease management platform.

[0060] In this embodiment, first, the multidimensional semantic understanding architecture uses natural language processing technology to perform in-depth analysis on the input text, which includes word segmentation, part-of-speech tagging and named entity recognition of the text, so as to extract key medical terms, symptom descriptions and other semantic elements; then, dependency syntax analysis and semantic role labeling techniques are applied to determine the grammatical structure and semantic role of these elements in the sentence, and reveal the direct or indirect relationship between them; then, in combination with the domain-specific knowledge graph, the extracted entities are linked to the corresponding concepts in the graph, and the predefined entity relationships in the graph are used to identify and infer the complex associations between entities; finally, through machine learning models, especially deep learning models such as recurrent neural networks or transformer networks, these multi-dimensional semantic information are comprehensively analyzed and learned, so as to achieve accurate identification and understanding of the association relationship between multiple semantic key elements in the user's consultation intention.

[0061] Step 204: Generate a semantic relationship graph of the chronic disease consultation intention instruction based on the element association relationship.

[0062] Among them, the semantic relationship graph includes multiple nodes and the node connection relationship between each node, and the node is used to represent the semantic key elements. The semantic relationship graph is a graphical data structure that represents the semantic key elements (such as nouns, terms or concepts) in the text as nodes, and the semantic connections between these elements (such as causal, attribute, classification and other relationships) as edges, thereby constructing a network that intuitively displays the complex connections and hierarchical structures between information. In this embodiment, the semantic relationship graph enables the server to capture and understand the relationship between key elements such as medical terms, symptom descriptions, time information and behavioral actions in patient consultations, and then analyze and interpret the patient's intentions, providing a basis for providing accurate medical advice and decision support.

[0063] In this step, a node is first created for each identified semantic key element. These nodes represent entities (semantic key elements) in the patient consultation, such as diseases, symptoms, drugs, etc. Then, edges between nodes are constructed based on the associations between entities. Edges represent the relationships between entities, such as "cause", "treatment", "related symptoms", etc. Next, the nodes and edges are organized into a directed or undirected graph, which reflects the structured representation of semantic elements in the consultation text.

[0064] At the same time, the generated graph is optimized to remove unnecessary edges or nodes and merge similar entities to improve the readability and analysis efficiency of the graph. Additional semantic information is added to the graph, such as confidence score, association strength, etc., which can help further analyze and interpret the patient's consultation intention.

[0065] Finally, the semantic relationship graph is presented to users such as patients or medical professionals in a graphical manner, allowing them to intuitively understand the complex semantic structure of patient consultations.

[0066] This step enables the server to extract valuable information from the natural language of patient consultation and present it in a structured form, thus laying a solid foundation for providing accurate medical advice and decision support.

[0067] Step 205, determining the main intention information corresponding to the chronic disease consultation intention instruction based on the semantic relationship graph.

[0068] Among them, the main intention information is the user's main and core consultation purpose, demand or problem in the chronic disease consultation intention instruction.

[0069] Specifically, the server first interprets the generated semantic relationship graph in depth, analyzing the nodes in the graph one by one, that is, each semantic key element, and the connection relationship between the nodes.

[0070] For medical term nodes, the server will make judgments based on known medical knowledge and common chronic disease patterns. For example, if the medical term node "diabetes" appears in the graph and is closely connected to symptom description nodes such as "high blood sugar" and "drinking and eating more" and behavior action nodes such as "insulin injection", the server will preliminarily infer that the main intention may be related to symptom control or treatment plan consultation for diabetes.

[0071] For symptom description nodes, the server analyzes them based on their severity, frequency of occurrence, and degree of relevance to medical terms. If multiple symptom description nodes point to the same medical term, then this medical term is likely to be the core of the main intent.

[0072] Time information nodes can also provide clues for determining the main intention. If the time information shows that the symptoms appear at a relatively concentrated time or have a clear sequence with specific treatment behaviors, this may suggest that the patient has doubts about the changes in the condition or the effect of treatment within a specific time period.

[0073] Behavior nodes reflect the patient’s actual actions and focus. For example, if there are behavior nodes such as “consult a doctor” and “check drug side effects”, it means that the patient may have doubts about the current treatment plan or need more information support.

[0074] The server comprehensively considers these factors and conducts a comprehensive analysis of the semantic relationship graph to determine the main intent information corresponding to the chronic disease consultation intention instruction, laying the foundation for the subsequent accurate provision of chronic disease consultation answers.

[0075] Step 206, determining the intent confidence of the main intent information.

[0076] The intention confidence refers to a quantitative evaluation value of the reliability or credibility of the above main intention information. The larger the value of the intention confidence, the higher the reliability or credibility of the main intention information.

[0077] Specifically, the server will first make a preliminary assessment based on the structure and content of the semantic relationship graph. If the nodes in the semantic relationship graph are clear and unambiguous, the connection relationship is tight and conforms to medical logic, then the confidence level will be relatively high. For example, when a medical term node is tightly connected to multiple related symptom description nodes and behavior action nodes, and this connection relationship has a high degree of rationality in medical knowledge, the confidence level will be increased.

[0078] The server will also analyze historical data. If a similar semantic relationship graph was accurately interpreted as a specific main intention in past consultations, and the subsequent consultation answers received positive feedback from users, the confidence level of the current semantic relationship graph with a similar structure will also be increased accordingly.

[0079] At the same time, the server will use machine learning algorithms to learn from a large number of processed consultation cases and continuously optimize the confidence determination method. These algorithms can automatically adjust the confidence calculation weights based on different characteristics, such as the number of nodes, the strength of the connection, the commonness of medical terms, etc.

[0080] In addition, the server will also consider the user's interactive behavior. If the user further emphasizes or supplements certain key elements after making a consultation intent instruction, this will also affect the determination of confidence. For example, if the user emphasizes the severity of a symptom after describing the symptom, the confidence of the main intent information related to the symptom may be adjusted accordingly.

[0081] By combining these factors, the server can more accurately determine the intent confidence of the main intent information.

[0082] Step 207: When the intention confidence is greater than or equal to the set confidence threshold, the answer to the chronic disease consultation is determined based on the main intention information.

[0083] Specifically, if the intent confidence is greater than or equal to the set confidence threshold, it indicates that the server has sufficient confidence in the identified main intent information and believes that it accurately reflects the user's true intent.

[0084] In this embodiment, the server can use natural language processing technology to deeply analyze the main intent information. Through lexical analysis, syntactic analysis, and semantic analysis, key medical terms, symptom descriptions, behavioral actions, and time information are extracted. For example, key information such as "diabetes", "drinking more", "insulin injection", and "the past week" are identified.

[0085] Next, the server queries the pre-built medical knowledge base and knowledge graph. In the medical knowledge base, the extracted medical terms are used to find relevant disease definitions, symptoms, treatment methods and other detailed information. The knowledge graph is used to determine the association between different key elements. For example, if the main intent information contains "diabetes" and "blurred vision", the association that diabetes may cause blurred vision can be found in the knowledge graph, thus providing a basis for subsequent answer generation.

[0086] At the same time, the server can also use machine learning algorithms to learn from a large number of historical chronic disease consultation cases. By analyzing the consultation answers corresponding to similar main intent information, the most likely answer direction of the current main intent information is predicted. For example, if many consultations on "skin itching in diabetic patients" in historical cases have received advice on blood sugar control and skin care, then when there are similar situations in the new main intent information, the server can refer to these suggestions to generate answers.

[0087] In addition, the server can also perform reasoning in conjunction with the rule engine. According to preset medical rules, such as the possibility rules for specific symptom combinations corresponding to specific diseases, the applicable condition rules for different treatment methods, etc., the main intent information is inferred and judged. For example, if symptoms such as "high blood pressure", "headache", and "palpitations" appear in the main intent information, the rule engine can infer that it may be necessary to further check blood pressure and adjust antihypertensive drugs based on the combination of these symptoms.

[0088] The server can also integrate and optimize the results obtained from the above technical means to generate accurate, comprehensive and easy-to-understand answers to chronic disease consultations to meet user needs.

[0089] Step 208: Feedback the answers to the chronic disease consultation to the user.

[0090] Specifically, the server feeds back the chronic disease consultation answer to the user in the form of text, voice or graphics through its communication interface, ensuring that the user can receive timely, accurate and easy-to-understand medical information and advice to meet their consultation needs and support their chronic disease self-management process.

[0091] In some embodiments, the server will organize and format the answers so that they are presented in a clear and easy-to-read form. If the answer contains medical terms, it will be explained in plain language as much as possible. Then, according to the channel used by the user, such as chronic disease management applications, web interfaces, or text messages, choose the appropriate feedback method. If it is an application or web interface, the answer will be displayed directly on the user's consultation interface, and the readability and comprehensibility of the answer may be enhanced by point-by-point explanations, pictures and texts, etc. For SMS feedback, the answer will be simplified into concise and clear text content to ensure that users can quickly obtain key information. At the same time, the server will also record the time and content of the feedback for subsequent user satisfaction surveys and service quality improvements. In addition, if the answer involves further action suggestions, such as medical examinations, adjusting drug dosages, etc., the server will give clear prompts and instructions in the feedback to guide users to take the right actions.

[0092] Through these methods, the server can promptly and accurately feedback answers to chronic disease consultations to users, providing strong support for users' chronic disease management.

[0093] This embodiment constructs a human-computer dialogue data processing method for chronic disease management through the above steps, which realizes providing users with more accurate and personalized chronic disease consultation services, thereby significantly improving user experience and the service quality of chronic disease management. First, through natural language processing technology, such as entity recognition and intent recognition, key semantic elements are extracted from the user's chronic disease consultation instructions, including medical terms, symptom descriptions, time information, and behavioral actions. Then, these semantic key elements are used to determine the association between them, and a detailed semantic relationship graph is constructed. The nodes in the graph represent the extracted semantic elements, and the edges represent the relationship between them. Based on this semantic relationship graph, the main intention information of the user's consultation can be accurately identified, and the confidence of this intention can be calculated. When the confidence reaches or exceeds the preset threshold, an accurate chronic disease consultation answer will be retrieved or generated based on the main intention information, and fed back to the user in an appropriate manner. This method not only improves the accuracy of the answer, but also enhances the user's sense of participation and satisfaction through timely and effective feedback, making the chronic disease management service more humane and intelligent.

[0094] Combine the following Figure 3To further illustrate the method of this embodiment, it specifically includes the following steps: Step 301, obtaining a chronic disease consultation intention instruction issued by a user.

[0095] This step refers to the description in the above embodiment and will not be repeated here.

[0096] Step 302: Use natural language processing technology to identify and process the chronic disease consultation intention instructions to obtain multiple semantic key elements.

[0097] This step refers to the description in the above embodiment and will not be repeated here.

[0098] Step 303: determine the element association relationship between multiple semantic key elements.

[0099] This step refers to the description in the above embodiment and will not be repeated here.

[0100] Step 304: Generate a semantic relationship graph of the chronic disease consultation intention instruction based on the element association relationship.

[0101] This step refers to the description in the above embodiment and will not be repeated here.

[0102] Step 305, determining, based on the semantic relationship graph, that the chronic disease consultation intention instruction corresponds to multiple possible intention information.

[0103] Among them, possible intention information refers to the description of multiple potential and logical user consultation purposes or needs extracted based on the semantic relationship graph during the analysis and processing of the user's chronic disease consultation intention instructions. Possible intention information is a series of possibilities obtained by comprehensively interpreting each node and its connection relationship in the semantic relationship graph, combining medical knowledge and natural language processing technology.

[0104] For example, when nodes such as "diabetes", "blood sugar fluctuations", and "diet adjustment" and corresponding connections appear in the semantic relationship graph, possible intent information may include various speculations such as how to make dietary adjustments when diabetic patients have blood sugar fluctuations, and the impact of dietary adjustments on blood sugar fluctuations. These speculations may not necessarily be completely certain to be the user's true intentions, but they all have certain rationality and possibility, and further analysis and screening are needed to clarify the user's main consulting intention.

[0105] In this step, the server first analyzes the nodes in the semantic relationship graph, which represent semantic key elements such as medical terms, symptom descriptions, time information, and behavioral actions. For each node, the server will combine medical knowledge and common chronic disease scenarios to understand it. For example, when seeing the medical term node "diabetes", the server will think of various symptoms, treatment methods, and complications related to it. Next, analyze the connection relationship between the nodes. Different connection relationships may imply different intention directions. If there is a connection between the "diabetes" node and the "hyperglycemia" node, and there is also a "diet control" node connected to it, this may mean that the user's intention is to consult diabetic patients on how to control their diet in the case of high blood sugar. The server also considers the role of time information nodes in the relationship graph. If there are time nodes such as "recent" and "long-term" connected to specific symptom or behavior action nodes, it may mean that the user has questions about the changes in the disease or the treatment effect in different time periods.

[0106] In addition, the server can also integrate multiple nodes and their connection relationships to perform combined analysis. For example, the combination of the "hypertension" node, the "dizziness" node, and the "drug adjustment" node may correspond to the user's intention to ask whether a hypertensive patient needs to adjust his or her medication after experiencing dizziness.

[0107] By comprehensively interpreting the semantic relationship graph and trying various combinations, the server can determine multiple possible intent information, providing rich possibilities for further screening and determining the main intent information.

[0108] Step 306, determining the information weight of the nodes in each possible intention information and the closeness of the connection relationship between each node.

[0109] Among them, information weight refers to the numerical value assigned to each node to represent its importance when analyzing possible intention information, based on factors such as the importance of the semantic key elements represented by the node in the medical knowledge system, the frequency of occurrence, and the degree of relevance to the core issues.

[0110] Relationship closeness refers to the closeness of the connection between nodes in the semantic relationship graph, which is determined by analyzing the logical strength of the connection, its rationality in the medical context, and its criticality to the expression of intent.

[0111] In this embodiment, multiple factors need to be considered to determine the information weight, such as the importance of the medical terminology node in the medical knowledge system, the severity and frequency of occurrence of the symptom description node, etc. Specifically, natural language processing technology can be used to analyze the semantic key elements represented by each node. If it is a medical terminology node, the medical knowledge base will be queried, and a certain weight will be assigned according to the importance, commonness and correlation of the term with other diseases in the field of chronic diseases. For example, "diabetes" is a common and widely-affected chronic disease, and its node information weight is relatively high. For symptom description nodes, the severity of the symptoms, the frequency of occurrence, and the contribution to the diagnosis of the disease will be considered. For example, "severe headache", a symptom that is more serious and may indicate a variety of serious diseases, will have a higher information weight than general symptoms such as "mild fatigue".

[0112] The closeness of the relationship is determined by analyzing the strength and rationality of the node connection relationship. It can be determined by their distance in the semantic relationship graph, the strength of the connected edge, or the correlation score. Specifically, graph analysis algorithms and medical logic reasoning can be used. If the connection between two nodes has a clear causal relationship, accompanying relationship, or diagnostic relationship in medical knowledge, then their closeness of relationship will be higher. For example, there is usually a strong causal relationship between the "hypertension" node and the "dizziness" node, and the closeness of their connection relationship will be correspondingly improved. At the same time, the direction and strength of the connection will also be considered. If one node has a strong influence or directionality on another node, then the connection strength is large and the closeness of the relationship will be higher. In addition, by analyzing a large number of medical literature and clinical cases, extracting common node connection patterns, it can also assist in determining the closeness of the relationship. For example, in many cases, if the "specific drug treatment" node and the "symptom relief" node frequently appear at the same time, then the closeness of the connection relationship between them will be considered high. Through these technical means, the information weight of the nodes in each possible intention information and the closeness of the connection relationship between each node can be determined more accurately.

[0113] By adopting the above scheme, the server can determine a reasonable information weight for each node in the possible intention information, as well as the accurate relationship closeness of the connection relationship between each node.

[0114] Step 307, determining the criticality score of each possible intention information according to the information weight and the relationship closeness.

[0115] First, for information weight, different weight coefficients are assigned to different types of nodes using mathematical models. For example, medical terminology nodes may be assigned a higher basic weight based on their importance and prevalence in the field of chronic diseases, while the weight of symptom description nodes may be adjusted based on the severity and specificity of the symptoms. By quantifying the information of each node, an initial set of weight values ​​is obtained.

[0116] For relationship closeness, graph algorithms and statistical analysis methods are used to determine the strength of the connection relationship. If the connection between two nodes appears frequently in a large number of medical cases or has a clear medical logical relationship, then their relationship closeness will be high. The relationship closeness can be quantified by calculating indicators such as the frequency of the connection and the correlation coefficient.

[0117] After that, a weighted summation method can be used to assign a certain weight ratio to the information weight and the relationship closeness, and then the node weight and connection relationship closeness in each possible intent information can be comprehensively calculated. For example, if the node information weight contained in a possible intent information is high, and the connection relationship closeness between the nodes is also strong, then the criticality score of this possible intent information will be correspondingly high.

[0118] At the same time, machine learning algorithms can also be introduced to learn from a large number of labeled chronic disease consultation cases, optimize the calculation methods of information weight and relationship closeness, and their weight distribution in determining the criticality score, so as to more accurately evaluate the importance of each possible intention information and obtain a more reliable criticality score.

[0119] In some embodiments, step 307 may specifically include: S3071, determining information entropy according to the node information weight vector of the information weight.

[0120] Among them, information entropy is used to represent the uncertainty of the node information weight vector.

[0121] The server first organizes the node information weights in each possible intent information into a node information weight vector. The node information weight vector reflects the weight distribution of different nodes in a specific possible intent information.

[0122] For a given node information weight vector (hereinafter referred to as weight vector), the server calculates the probability corresponding to each weight value, that is, the proportion of the weight value in the entire vector. Then, according to the relevant mathematical definition of information entropy in information theory, it is calculated based on these probability values.

[0123] If the weight values ​​in the weight vector are evenly distributed, that is, the information weights of each node are not much different, then the information entropy will be high, which means that the uncertainty of the weight vector is large. On the contrary, if the information weights of some nodes are significantly higher than other nodes, the distribution of the weight vector is more concentrated, the information entropy will be low, and the uncertainty is small.

[0124] By calculating information entropy, the server can understand the uncertainty of the node information weight in specific possible intent information, providing an important reference factor for the subsequent determination of criticality scores.

[0125] S3072, determining a weighted centrality measure according to the node information weight vector and the node relationship closeness matrix of the relationship closeness.

[0126] Among them, the weighted centrality measure is used to represent the importance of a node in the semantic relationship graph.

[0127] Specifically, the server first constructs a node relationship closeness matrix, which reflects the closeness of the connection relationship between nodes in the semantic relationship graph. At the same time, it combines the node information weight vector obtained previously to determine the weighted centrality measure.

[0128] The server can use a variety of graph analysis algorithms. For example, an improved algorithm based on node degree centrality, closeness centrality, or betweenness centrality can be used to incorporate the node information weight vector into the calculation. Taking degree centrality as an example, if a node has a close connection with many other nodes and the node's information weight is high, then its value in the weighted centrality metric will be correspondingly high. In this way, the server comprehensively considers the node's information weight and the closeness of the connection relationship to determine the importance of each node in the semantic relationship graph, that is, the weighted centrality metric.

[0129] The weighted centrality measure can reflect the criticality of a node in specific possible intent information, which is of great significance for the subsequent determination of the criticality score.

[0130] S3073, determine each criticality score based on information entropy and weighted centrality measurement.

[0131] Among them, the criticality score can comprehensively reflect the importance of nodes in possible intent information and the uncertainty of weights.

[0132] Specifically, the server can design a comprehensive calculation formula to integrate information entropy and weighted centrality metrics. For example, the weighted centrality metric can be adjusted according to the size of the information entropy. If the information entropy is high, it means that the uncertainty of the node information weight is large. At this time, the weight of the weighted centrality metric in the criticality score can be appropriately reduced to avoid the impact of uncertainty on the score. On the contrary, if the information entropy is low, it means that the weight distribution is more concentrated, and the weight of the weighted centrality metric can be increased. In this way, the server can determine a reasonable criticality score for each possible intent information based on information entropy and weighted centrality metrics.

[0133] In some embodiments, the criticality score may be calculated based on a preset criticality score formula, and the criticality score formula includes:

[0134]

[0135]

[0136] Among them, S i Score the criticality of the i-th possible intention information; W i is the node information weight vector of the i-th possible intention information, which is an n i ×1 column vector; C i is the node relationship closeness matrix of the i-th possible intention information, which is an n i ×n i A symmetric matrix; λ is a balance parameter, which is a decimal between 0 and 1. It is used to adjust the weight between information entropy and weighted centrality measure. λ can flexibly adjust the proportion of the two in the score according to the actual situation. When the value of λ is different, the key factors will also be different, which has strong adaptability.

[0137] H(W i ) represents information entropy; D(C i ,W i ) represents the weighted centrality measure; n i is the number of nodes in the i-th possible intention information; ϵ is an experimental parameter, which is a positive number. In this embodiment, it is preferably a decimal close to 0, for example, it can be a positive number less than 0.01 to avoid the situation where the denominator is zero; j and k are node numbers, and m is an index variable; W ijrepresents the information weight of the jth node in the i-th possible intention information; W ik represents the information weight of the kth node in the i-th possible intention information; C ijk Represents the closeness of the relationship between the jth node and the kth node; C ijm Represents the closeness of the relationship between the j-th node and the m-th node.

[0138] Furthermore, information entropy H(W i ) is defined by formula (2), and the node information weight vector W i The uncertainty of the node information weight vector W i The weight of each node in ij The more evenly distributed, The larger the value of i ) is larger, indicating that the uncertainty of the node weight in the possible intention information is higher, that is, it is difficult to determine which node is more critical.

[0139] The denominator (1-W ij +ϵ) is to avoid ij When it approaches 0 or 1, the log2 function has undefined or abnormally large values, which plays a role in smoothing and stabilizing the calculation, so that the calculation of information entropy can be carried out reasonably under various weight distribution conditions.

[0140] W ij represents the information weight of the jth node in the i-th possible intention information, and (1-W ij ) can be regarded as the degree to which the node does not have this weight. ij Divide by (1-W ij +ϵ), can highlight the impact of the differences between different weights on information entropy. ij When it is close to 1, the numerator is larger and the denominator is smaller, and the contribution to information entropy will be greater, indicating that the node is more important in this possible intention information, and its weight has higher certainty and lower uncertainty; on the contrary, when W ij When it is close to 0, the numerator is small, the denominator is large, and the contribution to information entropy is small, indicating that the weight uncertainty of the node is high.

[0141] Weighted centrality measure D(C i ,W i ) is defined by referring to formula (3), which comprehensively considers the node information weight vector and the node relationship closeness matrix to measure the importance of the node in the semantic relationship graph.

[0142] For the jth node, first calculate its relationship with other nodes C ijk and the corresponding node weight W ik The sum of the products is divided by the sum of the closeness of the relationship between the node and all nodes , and obtain a normalized importance measure. Then add it to the information weight W of node j itself ij Multiplying and summing them up comprehensively considers the information weight of the node and the closeness of its connection with other nodes, which can more accurately reflect the importance of the node in the entire semantic relationship graph.

[0143] If a node’s information weight W ij Large, and with other important nodes (information weight W ik Larger and more closely related C ijk If the connection between the key nodes is tight and the weighted centrality is high, the contribution of the node in the weighted centrality measure will be greater, thus highlighting the critical impact of the key node on the entire possible intent information.

[0144] The criticality scoring formula of this embodiment combines equations (1), (2), and (3) to comprehensively and reasonably integrate factors such as node information weight and relationship closeness from different angles, and can more accurately evaluate the criticality of each possible intent information, providing a reliable basis for subsequent intent analysis and processing.

[0145] Step 308, determining the main intention information from the multiple possible intention information according to the criticality score.

[0146] Specifically, the possible intent information with the highest criticality score may be selected as the main intent information. This is because this information is considered to be the most accurate reflection of the core content of the user's chronic disease consultation intention instruction among all possible intent information. For example, if among multiple possible intent information, the possible intent information about "consulting about the adjustment plan for hypertension medication" has the highest criticality score, then the server will determine it as the main intent information.

[0147] In this way, the server can accurately determine the user's most important consultation intention from many possibilities, laying the foundation for providing accurate chronic disease consultation answers in the future.

[0148] Step 309, determining the confidence evaluation factor of the main intention information.

[0149] Among them, the confidence evaluation factors include the completeness of the main intention information, the clarity of the semantic key elements, and the matching degree between the main intention information and the known chronic disease information.

[0150] The completeness of the main intent information refers to whether the main intent information covers sufficiently comprehensive and key content, including medical terms, symptoms, behaviors, time, etc., without missing important information. The server will use natural language processing technology to analyze the main intent information. First, extract the key elements, such as medical terms, symptom descriptions, behavioral actions, and time information. Then check whether these key elements are comprehensive and cover all important aspects of the user's questions. For example, if a user asks about diabetes, the server will determine whether the main intent information also contains key information such as blood sugar levels, medication status, diet status, and onset time. If more key information is missing, the completeness is low.

[0151] The clarity of the key semantic elements indicates whether the key semantic elements in the main intent information, such as symptom descriptions and professional terms, are clear, unambiguous, and easy to understand. The server also uses semantic analysis technology in natural language processing. The semantic key elements in the main intent information are carefully analyzed to determine whether their expressions are clear, accurate, and whether they are prone to ambiguity. For example, if the description of the symptoms in the main intent information is "feeling unwell", this description is vague and has a low degree of clarity; if it is "numbness and weakness in the left leg for three days", this description is clearer.

[0152] The matching degree between the main intent information and the known chronic disease information refers to the degree of consistency between the main intent information and the common chronic disease symptoms, diagnosis, treatment and other information in the pre-stored chronic disease knowledge base. The server will compare the main intent information with the pre-built chronic disease knowledge base. Through knowledge graph technology and text matching algorithm, find the corresponding content of the key elements in the main intent information in the knowledge base. If the symptoms, disease names, treatment methods, etc. in the main intent information are highly matched with the content in the knowledge base, then the matching degree is high. For example, if the main intent information is about the daily diet recommendations for patients with hypertension, and there is a wealth of knowledge about the dietary management of hypertension in the knowledge base, the two can correspond well and the matching degree will be high. Through these technical means, the server can more accurately determine the confidence evaluation factor of the main intent information.

[0153] Step 310, determining the intent confidence of the main intent information according to the confidence evaluation factor.

[0154] Specifically, the server will first assign different weights to each confidence evaluation factor. For example, it may be considered that the integrity of the main intent information has a greater impact on the confidence, so a higher weight is given; while the clarity of the key semantic elements and the matching degree of the main intent information with the known chronic disease information have relatively low weights. This weight allocation is usually determined based on the analysis of a large number of chronic disease consultation cases and the experience of medical experts.

[0155] Next, the server will conduct a quantitative assessment of each evaluation factor. For the completeness of the main intent information, the server will score according to the degree of missing key information. If the main intent information contains all important medical terms, symptom descriptions, behavioral actions, and time information, the completeness score will be higher; if some key information is missing, the score will be reduced accordingly. For the clarity of semantic key elements, the server will use natural language processing technology to analyze whether the expression of key elements is clear and unambiguous. If the expression is clear, the score will be high; if there are vague or ambiguous places, the score will be reduced. For the matching degree of the main intent information with known chronic disease information, the server will use knowledge graphs and text matching algorithms to calculate the similarity between the main intent information and the content in the chronic disease knowledge base. The higher the similarity, the higher the matching score.

[0156] Finally, the server multiplies the score of each evaluation factor by its corresponding weight, and then sums them up to get a comprehensive score. This score is the intent confidence of the main intent information. If the score is high, it means that the reliability and accuracy of the main intent information are high, and the server can be more confident in providing chronic disease consultation answers based on this main intent information in subsequent processing. Conversely, if the score is low, the server may need to further interact with the user to obtain more information to improve the confidence of the main intent information. In this way, the server can more accurately determine the intent confidence of the main intent information and provide users with more reliable chronic disease consultation services.

[0157] In some embodiments, the intent confidence may be calculated according to a preset confidence evaluation formula, where the confidence evaluation formula includes:

[0158]

[0159]

[0160]

[0161] Among them, T represents the intention confidence; I(W,P) represents the integrity measure of the main intention information, and its function expression refers to formula (5); L(W) represents the clarity of the key semantic elements, and its function expression refers to formula (6); Q(M) represents the matching degree between the main intention information and the known chronic disease information. Its function expression refers to formula (7); H(W) represents the information entropy of the main intention information, and its function expression can refer to formula (2); W is the node information weight vector; P is the probability of occurrence of the node; M is the matching matrix between the main intention information and the known chronic disease information; n is the number of nodes of the main intention information, j is the node sequence number; W j represents the information weight of the jth node; P j represents the probability of the jth node appearing in the main intent information. NLP techniques, such as term frequency-inverse document frequency (TF-IDF) or information retrieval models, can be used to evaluate the actual occurrence of nodes in the text (occurrence probability or frequency). If a node often appears in the intent content conveyed by the user, its occurrence probability P j On the contrary, if it rarely appears, the probability is low. For example, when asking questions about drug treatment in common chronic disease consultations, the probability of the node "drug dosage" appearing in the corresponding context may be relatively high; M j Indicates the matching degree between the jth node and the known chronic disease information; α, β and γ are balance parameters, which can adjust the weights of I(W,C), C(W) and Q(M) in the confidence calculation according to actual conditions, making the formula highly flexible and adaptable.

[0162] In this embodiment, formula (5) is used to calculate the integrity metric of the main intent information, which comprehensively measures the integrity of the main intent information from two aspects: node information weight and occurrence probability. Through reasonable normalization means, the calculation result is scientific, reasonable and easy to use, which provides an effective quantitative basis for more accurately judging the quality of the main intent information.

[0163] Molecular part Combining the information weight of a node with its probability of appearance means that not only the importance of the node itself is taken into account, but also the possibility of it actually appearing in the user's main intent information. By multiplying the two and then summing them, the contribution of all nodes to the overall integrity under these two factors is comprehensively considered. If the probability of occurrence of those nodes with high weights is also high, then the value of the numerator will be relatively large, which to a certain extent reflects that the main intent information is more complete in terms of key information inclusion.

[0164] Denominator It plays a normalization role, making the integrity metric value within a reasonable and comparable range, and avoiding the situation where the integrity metric value is too large or too small and difficult to measure and compare due to factors such as different levels of weights and different numbers of nodes.

[0165] In formula (6), The information entropy is converted into a measure of clarity by calculating the information entropy. When the information entropy is low, the value of C(W) is high, indicating that the clarity of the semantic key elements is high. At the same time, log2n as the denominator plays a normalization role, making the clarity value within a reasonable range.

[0166] Formula (7) calculates the average matching degree of all nodes in the main intention information with the known chronic disease information. If each node has a high matching degree with the known chronic disease information, then the average matching degree will be high, indicating that the main intention information is highly consistent with the known chronic disease information. By summing up the matching degrees of all nodes and then taking the average, the matching of each part of the main intention information with the known chronic disease information is taken into account, making the calculation of the matching degree more comprehensive and accurate.

[0167] This embodiment comprehensively considers the integrity of the main intent information, the clarity of the semantic key elements, and the matching degree with the known chronic disease information from multiple perspectives through the above formula, and can reasonably calculate the intent confidence of the main intent information, providing a reliable basis for chronic disease consultation.

[0168] Step 311, determine whether the intention confidence is greater than or equal to the confidence threshold.

[0169] If yes, proceed to step 314. If no, proceed to step 312.

[0170] Step 312: Send a prompt message to the user to clarify the intention.

[0171] Specifically, when the intent confidence of the main intent information calculated by the server is less than the confidence threshold, it means that the server is not sufficiently confident about the accuracy and reliability of the main intent information currently determined. At this time, the server needs to send an intent clarification prompt to the user. The intent clarification prompt is a prompt with targeted inquiries or guidance issued by the server to guide users to provide more accurate and complete information when the intent confidence of the main intent information in chronic disease consultation is less than the threshold.

[0172] First, the server will analyze the reasons for the low confidence, which may be the lack of completeness of the main intent information, unclear semantic key elements, or low matching with known chronic disease information. Then, the corresponding intent clarification prompt information is generated according to the specific reasons. For example, if the completeness is insufficient, the prompt information may ask the user whether there are other key contents such as symptoms, behavioral actions or time information that are not provided; if the semantic key elements are unclear, the user may be asked to further explain some vague statements; if the match is low, the user may be asked whether there is a specific chronic disease background or specific focus.

[0173] The server will send intent clarification prompts to the user through the channels used by the user, such as message push from the chronic disease management application, pop-up windows on the web interface, or text messages, to guide the user to provide more accurate and complete information, thereby improving the accuracy of subsequent determination of the main intent information.

[0174] Step 313: Receive a second chronic disease consultation intention instruction from the user.

[0175] Among them, the secondary chronic disease consultation intention instruction is generated by the user based on the intention clarification prompt information.

[0176] After receiving the intention clarification prompt, the user will further think about and clarify his or her question based on the prompt. Then, the user will re-express his or her chronic disease consultation intention in a clearer and more accurate way, generating a secondary chronic disease consultation intention instruction.

[0177] The server monitors and receives secondary instructions from users in real time through an interface that interacts with the user. The server uses natural language processing technology to quickly parse and analyze secondary instructions and extract key information, similar to the previous analysis process, but this time it will pay more attention to the association with the intent clarification prompt information. For example, if the user provides a detailed explanation of the ambiguous statement in the previous prompt information in the secondary instruction, the server will focus on analyzing the newly provided information, and combine it with the previous analysis results to re-determine the main intent information and calculate the intent confidence.

[0178] By receiving the user's secondary chronic disease consultation intention instructions, the server can continuously optimize its understanding of the user's intentions and improve the accuracy and quality of the chronic disease consultation services provided to the user.

[0179] Then return to step 302. The second chronic disease consultation intention instruction is processed again in the manner of steps 302-311.

[0180] Step 314, determining the answer to the chronic disease consultation based on the main intention information.

[0181] This step refers to the description in the above embodiment and will not be repeated here.

[0182] Step 315, feeding back the answers to the chronic disease consultation to the user.

[0183] This step refers to the description in the above embodiment and will not be repeated here.

[0184] Through the above method steps, this embodiment uses natural language processing technology to deeply analyze the user's chronic disease consultation intention instructions, sorts out semantic key elements and associations to construct a semantic relationship diagram to clarify the main intention information, and uses scientific quantitative formulas to determine the criticality score and intention confidence, to ensure the accuracy and objectivity of intention judgment in multiple dimensions; at the same time, when the intention confidence is insufficient, a user interaction feedback mechanism is set up to further improve the main intention information. The entire solution can accurately understand the user's intention, scientifically and rigorously evaluate information, and fully consider user interaction, effectively improving the quality of consulting services and user experience in chronic disease management.

[0185] The method provided in the above embodiment can be executed by an electronic device, and the electronic device specifically includes a chronic disease management server. The electronic device in the embodiment of the present invention is described below from the perspective of hardware processing. Figure 4 , is a schematic diagram of a physical device structure of an electronic device in an embodiment of the present invention.

[0186] It should be noted that Figure 4 The structure of the electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0187] like Figure 4 As shown, the electronic device includes a central processing unit (CPU) 401, which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM) 402 or the program loaded from the storage part 408 to the random access memory (RAM) 403, such as executing the method described in the above embodiment. In RAM 403, various programs and data required for system operation are also stored. CPU 401, ROM 402 and RAM 403 are connected to each other through bus 404. Input / output (I / O) interface 405 is also connected to bus 404.

[0188] The following components are connected to the I / O interface 405: an input section 406 including an audio input device, a button switch, etc.; an output section 407 including a liquid crystal display (LCD) and an audio output device, an indicator light, etc.; a storage section 408 including a hard disk, etc.; and a communication section 409 including a network interface card such as a LAN (Local Area Network) card, a modem, etc. The communication section 409 performs communication processing via a network such as the Internet. A drive 410 is also connected to the I / O interface 405 as needed. A removable medium 411, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 410 as needed so that a computer program read therefrom is installed into the storage section 408 as needed.

[0189] In particular, according to an embodiment of the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present invention includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a computer program for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through the communication part 409, and / or installed from a removable medium 411. When the computer program is executed by the central processing unit (CPU) 401, various functions defined in the present invention are performed.

[0190] It should be noted that specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in combination with an instruction execution system, apparatus, or device.

[0191] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present invention. Each box in the flowchart or block diagram may represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of a code contains one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions marked in the box may also occur in an order different from that marked in the accompanying drawings.

[0192] Specifically, the electronic device of this embodiment includes a processor and a memory, the memory is coupled to one or more processors, the memory is used to store computer program code, the computer program code includes computer instructions, and one or more processors call the computer instructions to enable the electronic device to execute the method provided by the above embodiment.

[0193] As another aspect, the present invention further provides a computer-readable storage medium, which may be included in the electronic device described in the above embodiment; or may exist independently without being assembled into the electronic device. The above storage medium carries one or more computer programs, and when the above one or more computer programs are executed by a processor of the electronic device, the electronic device implements the method provided in the above embodiment.

[0194] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention.

[0195] As used in the above embodiments, the term "when..." may be interpreted to mean "if..." or "after..." or "in response to determining..." or "in response to detecting...", depending on the context. Similarly, the phrases "upon determining..." or "if (the stated condition or event) is detected" may be interpreted to mean "if determining..." or "in response to determining..." or "upon detecting (the stated condition or event)" or "in response to detecting (the stated condition or event)", depending on the context.

[0196] Those skilled in the art can understand that to implement all or part of the processes in the above-mentioned embodiments, the processes can be completed by computer programs to instruct related hardware, and the programs can be stored in computer-readable storage media. When the programs are executed, they can include the processes of the above-mentioned method embodiments. The aforementioned storage media include: ROM or random access memory RAM, magnetic disk or optical disk and other media that can store program codes.

Claims

1. A human-computer dialogue data processing method for chronic disease management, characterized in that: include: Obtain the chronic disease consultation intention instruction issued by the user; Using natural language processing technology to identify and process the chronic disease consultation intention instruction, multiple semantic key elements are obtained, and the semantic key elements include medical terms, symptom descriptions, time information and behavioral actions; Determining an element association relationship between a plurality of the semantic key elements; Generate a semantic relationship graph of the chronic disease consultation intention instruction according to the element association relationship, wherein the semantic relationship graph includes a plurality of nodes and node connection relationships between the nodes, and the nodes are used to represent the semantic key elements; Determine the main intention information corresponding to the chronic disease consultation intention instruction based on the semantic relationship graph; determining an intent confidence level of the primary intent information; When the intention confidence is greater than or equal to the set confidence threshold, the chronic disease consultation answer is determined according to the main intention information; The answer to the chronic disease consultation is fed back to the user.

2. The method according to claim 1, characterized in that The determining, based on the semantic relationship graph, the main intention information corresponding to the chronic disease consultation intention instruction includes: According to the semantic relationship graph, determining that the chronic disease consultation intention instruction corresponds to multiple possible intention information; Determine the information weight of the node and the closeness of the connection relationship between the nodes in each possible intention information; Determining a criticality score of each of the possible intention information according to the information weight and the relationship closeness; The main intention information is determined from the plurality of possible intention information according to the criticality score.

3. The method according to claim 2, characterized in that Determining the criticality score of each of the possible intention information according to the information weight and the relationship closeness includes: Determining information entropy according to the node information weight vector of the information weight, wherein the information entropy is used to represent the uncertainty of the node information weight vector; Determining a weighted centrality metric according to the node information weight vector and the node relationship closeness matrix of the relationship closeness, wherein the weighted centrality metric is used to represent the importance of the node in the semantic relationship graph; The criticality scores are determined according to the information entropy and the weighted centrality measure.

4. The method according to claim 3, characterized in that Determining each of the criticality scores according to the information entropy and the weighted centrality measure includes: The criticality score is calculated according to a preset criticality score formula, and the criticality score formula includes:

5.

6.

7. Among them, S i is the criticality score of the i-th possible intention information, W i is the node information weight vector of the i-th possible intention information, C i is the node relationship closeness matrix of the i-th possible intention information, λ is the balance parameter, H(W i ) represents the information entropy, D(C i ,W i ) represents the weighted centrality measure, n i is the number of nodes in the i-th possible intention information, ϵ is the experimental parameter, j and k are the node numbers, m is the index variable, and W ij and W ik Respectively represent the information weights of the jth node and the kth node in the i-th possible intention information, C ijk represents the closeness of the relationship between the jth node and the kth node, C ijm Represents the closeness of the relationship between the j-th node and the m-th node.

8. The method according to any one of claims 2 to 4, characterized in that: The determining the intent confidence of the main intent information includes: Determining a confidence evaluation factor of the main intention information, wherein the confidence evaluation factor includes the completeness of the main intention information, the clarity of the semantic key elements, and the matching degree between the main intention information and known chronic disease information; The intent confidence of the main intent information is determined according to the confidence evaluation factor.

9. The method according to claim 5, characterized in that The determining the intention confidence of the main intention information according to the confidence evaluation factor includes: The intention confidence is calculated according to a preset confidence evaluation formula, wherein the confidence evaluation formula includes:

10.

11.

12.

13. Wherein, T represents the confidence of the intention, I(W,P) represents the integrity measure of the main intention information, L(W) represents the clarity of the semantic key elements, Q(M) represents the matching degree between the main intention information and the known chronic disease information, H(W) represents the information entropy of the main intention information, W is the node information weight vector, P is the occurrence probability of the node, M is the matching degree matrix between the main intention information and the known chronic disease information, n is the number of nodes of the main intention information, j is the node sequence number, W j represents the information weight of the jth node, P j represents the probability of the jth node appearing in the main intent information, M j represents the matching degree between the jth node and the known chronic disease information, and α, β and γ are balance parameters.

14. The method according to claim 1, characterized in that After determining the intent confidence of the main intent information, the method further includes: When the intention confidence is less than the confidence threshold, sending an intention clarification prompt message to the user; receiving a secondary chronic disease consultation intention instruction issued by the user, wherein the secondary chronic disease consultation intention instruction is generated by the user according to the intention clarification prompt information; The main intention information is determined according to the secondary chronic disease consultation intention instruction.

15. An electronic device, characterized in that: including one or more processors and memory; The memory is coupled to the one or more processors, and the memory is used to store computer program codes, wherein the computer program codes include computer instructions, and the one or more processors call the computer instructions to enable the electronic device to execute the method according to any one of claims 1 to 7.

16. A computer-readable storage medium storing computer instructions, characterized in that: When the computer instructions are executed on an electronic device, the electronic device is caused to execute the method as claimed in any one of claims 1 to 7.

17. A computer program product, characterized in that When the computer program product is executed on an electronic device, the electronic device is enabled to execute the method according to any one of claims 1 to 7.

Images