Chronic Disease Data Analysis Method and System Based on Natural Language Processing and Integrated Training

Abstract

The present invention discloses a chronic disease data analysis method and system based on natural language processing and integrated training, which relates to the field of artificial intelligence and includes: First, using natural language processing technology to generate a first chronic disease instance without labeled chronic disease target values from unstructured medical texts, and based on this, obtaining an initially trained initial analysis model through contrastive learning in an integrated training framework. Then, obtaining a second chronic disease instance with labeled chronic disease target values, and performing multi-model integration optimization on the initial analysis model to obtain a trained chronic disease recognition model. Finally, using this model to identify the chronic disease symptoms to be analyzed, obtaining the chronic disease data analysis result, and realizing the effective analysis of chronic disease data.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular, to a chronic disease data analysis method and system based on natural language processing and integrated training. Background Art

[0002] With the development of medical informatization, a large amount of unstructured medical text data contains rich chronic disease information, but it is difficult to directly utilize. Traditional chronic disease data analysis methods have limitations in processing such complex data, such as being unable to fully mine the potential information in the text, and the accuracy and generalization ability of the models are insufficient. At the same time, a single model is difficult to comprehensively analyze the diversity and complexity of chronic disease data. Summary of the Invention

[0003] The purpose of the present invention is to provide a chronic disease data analysis method and system based on natural language processing and integrated training.

[0004] In a first aspect, an embodiment of the present invention provides a chronic disease data analysis method based on natural language processing and integrated training, including:

[0005] Performing entity recognition and relationship extraction on unstructured medical text through natural language processing technology to generate a first chronic disease instance without labeled chronic disease target values;

[0006] Obtaining a pre-trained initial analysis model, where the pre-trained initial analysis model is obtained through contrastive learning based on the first chronic disease instance by an integrated training framework, and the integrated training framework integrates the feature representations output by multiple heterogeneous models;

[0007] Obtaining a second chronic disease instance corresponding to a disease analysis index, where the second chronic disease instance is a chronic disease instance with labeled chronic disease target values;

[0008] Performing multi-model integration optimization on the pre-trained initial analysis model based on the second chronic disease instance to obtain a completed-trained chronic disease recognition model corresponding to the disease analysis index;

[0009] Obtaining the chronic disease symptoms to be analyzed, and performing recognition processing on the chronic disease symptoms to be analyzed based on the completed-trained chronic disease recognition model to obtain the chronic disease data analysis result of the chronic disease symptoms to be analyzed.

[0010] In a possible implementation manner, the method further includes:

[0011] Obtaining a first chronic disease instance, where the first chronic disease instance includes multiple first chronic disease symptom instances;

[0012] Determine the first chronic disease symptom instance topology of each first chronic disease symptom instance according to the disease description information of each first chronic disease symptom instance, and determine the contrast learning feature encoding of each first chronic disease symptom instance according to the topology of each first chronic disease symptom instance;

[0013] Perform recognition processing on the topology of each first chronic disease symptom instance based on a preset basic model to obtain the first feature vector instance corresponding to the topology of each first chronic disease symptom instance;

[0014] Perform recognition on each first feature vector instance based on the trained feature recognition component to obtain the first symptom inference result corresponding to each first chronic disease symptom instance;

[0015] Perform integrated optimization training on the preset basic model according to the contrast learning feature encoding and the first symptom inference result of each first chronic disease symptom instance to obtain the initially trained analysis model.

[0016] In a possible implementation manner, the determining the contrast learning feature encoding of each first chronic disease symptom instance according to the topology of each first chronic disease symptom instance includes:

[0017] Obtain a plurality of pre-set symptom topology patterns, where each symptom topology pattern corresponds to a symptom combination structure with clinical diagnostic significance;

[0018] Associate the topology of each first chronic disease symptom instance with the plurality of symptom topology patterns to obtain the correlation coefficient parameter of the topology of each first chronic disease symptom instance;

[0019] Determine the contrast learning feature encoding of each first chronic disease symptom instance according to the correlation coefficient parameter corresponding to the topology of each first chronic disease symptom instance.

[0020] In a possible implementation manner, the associating the topology of each first chronic disease symptom instance with the plurality of symptom topology patterns to obtain the correlation coefficient parameter of the topology of each first chronic disease symptom instance includes:

[0021] Associate the topology of the target first chronic disease symptom instance with each symptom topology pattern to obtain the correlation coefficient corresponding to each symptom topology pattern;

[0022] Use the symptom topology pattern with the correlation coefficient exceeding the correlation coefficient threshold as the target symptom topology pattern corresponding to the topology of the target first chronic disease symptom instance;

[0023] Use the target symptom topology pattern corresponding to the topology of the target first chronic disease symptom instance as the correlation coefficient parameter of the topology of the target first chronic disease symptom instance.

[0024] In a possible implementation manner, determining the contrast learning feature encoding of each first chronic disease symptom instance according to the correlation coefficient parameter corresponding to each first chronic disease symptom instance topology includes:

[0025] Obtain the core feature encoding of the target symptom topology pattern corresponding to the target first chronic disease symptom instance topology;

[0026] Determine the control feature encoding in the multiple symptom topology patterns that is not associated with the target chronic disease symptom instance topology according to the core feature encoding;

[0027] Perform feature coupling processing on the core feature encoding and the control feature encoding to obtain the contrast learning feature encoding of the target first chronic disease symptom instance topology.

[0028] In a possible implementation manner, the preset basic model includes multiple physical sign interaction networks and at least one physical sign fusion network. Based on the preset basic model, performing identification processing on each first chronic disease symptom instance topology to obtain the first feature vector instance corresponding to each first chronic disease symptom instance topology includes:

[0029] According to the disease description information of the target chronic disease symptom instance, obtain the physical sign feature vector of each symptom entity in the target first chronic disease symptom instance topology, the associated physical sign entity of each symptom entity, and the association strength vector of the pathological association relationship between each symptom entity and the associated physical sign entity;

[0030] Perform a feature transformation operation on the physical sign feature vector of each symptom entity to obtain the basic physical sign vector of each symptom entity;

[0031] Fuse the basic physical sign vector of the associated physical sign entity and the association strength vector to obtain a fusion vector;

[0032] Obtain the first weight tensor of the first feature aggregation module for interactive aggregation processing, and perform a linear transformation operation on the first weight tensor and the fusion vector to obtain a first transformation coefficient tensor;

[0033] Process the first transformation coefficient tensor based on a preset first non-linear mapping function to obtain the first-order interactive aggregation vector of the symptom entity;

[0034] Perform physical sign aggregation processing on the input physical sign vector of each symptom entity and the first-order interactive aggregation vector to obtain the first-order physical sign vector of each symptom entity;

[0035] Based on the target sign interaction network, perform interaction aggregation processing on the previous-order sign vectors of the associated sign entities and the association strength vectors to obtain the target-order interaction aggregation vectors of each symptom entity;

[0036] Perform sign aggregation processing on the previous-order sign vectors of each symptom entity and the target-order interaction aggregation vectors to obtain the target-order sign vectors of each symptom entity;

[0037] Use the target-order sign vectors of each symptom entity as the feature vectors of each symptom entity;

[0038] Based on the sign fusion network, process the feature vectors of each symptom entity to obtain the first feature vector instance corresponding to the target first chronic disease symptom instance topology.

[0039] In a possible implementation manner, the performing sign aggregation processing on the input sign vectors of each symptom entity and the first-order interaction aggregation vectors to obtain the first-order sign vectors of each symptom entity includes:

[0040] Obtain the second weight tensor of the second feature aggregation module for performing sign aggregation processing;

[0041] Perform a linear transformation operation on the second weight tensor and the first-order interaction aggregation vectors to obtain a second transformation coefficient tensor;

[0042] Based on a pre-set second non-linear mapping function, process the input sign vectors of each symptom entity and the second transformation coefficient tensor to obtain the first-order sign vectors of each symptom entity.

[0043] In a possible implementation manner, the obtaining the sign feature vectors of each symptom entity, the associated sign entities of each symptom entity, and the association strength vectors of the pathological association relationships between each symptom entity and the associated sign entities in the target first chronic disease symptom instance topology according to the disease description information of the target chronic disease symptom instance includes:

[0044] According to the disease description information of the target first chronic disease symptom instance, obtain the clinical index data of each sub-symptom instance in the target first chronic disease symptom instance;

[0045] Perform feature coupling processing on the clinical index data of each sub-symptom instance to obtain the sign feature vectors of each symptom entity in the target first chronic disease symptom instance topology;

[0046] Use the control sub-symptoms having a clinical association path with each sub-symptom instance as the associated sub-symptoms of each sub-symptom instance;

[0047] Obtain the clinical index data of the clinical association path between each of the sub-symptom instances and the associated sub-symptoms;

[0048] Perform feature coupling processing on the clinical index data of the clinical association path to obtain the association strength vector of the pathological association relationship between each symptom entity and the associated physical sign entity.

[0049] In a possible implementation manner, the model optimization of the initially trained initial analysis model based on the second chronic disease instance to obtain the trained chronic disease recognition model corresponding to the disease analysis index includes:

[0050] Obtain the second chronic disease symptom instance topology of each second chronic disease symptom instance in the second chronic disease instance;

[0051] Based on the initially trained initial analysis model, perform recognition processing on each second chronic disease symptom instance topology to obtain the second feature vector instance corresponding to each second chronic disease symptom instance topology;

[0052] Based on the trained feature recognition component, perform recognition on each second feature vector instance to obtain the second symptom inference result corresponding to each second chronic disease symptom instance;

[0053] According to the chronic disease target value of each second chronic disease symptom instance and the second symptom inference result, perform integrated optimization training on the initially trained initial analysis model to obtain the trained chronic disease recognition model corresponding to the disease analysis index.

[0054] In a second aspect, an embodiment of the present invention provides a server system, including a server, and the server is used to execute the method described in the first aspect.

[0055] Compared with the prior art, the beneficial effects provided by the present invention include: adopting a chronic disease data analysis method and system based on natural language processing and integrated training disclosed in the present invention, generating a first chronic disease instance without labeled chronic disease target values from unstructured medical texts by using natural language processing technology, and obtaining an initially trained initial analysis model through contrast learning based on this integrated training framework. Then obtain a second chronic disease instance with labeled chronic disease target values, perform multi-model integrated optimization on the initial analysis model, and obtain a trained chronic disease recognition model. Finally, use this model to identify the chronic disease symptoms to be analyzed and obtain the chronic disease data analysis result, realizing the effective analysis of chronic disease data. Description of the Drawings

[0056] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, without creative efforts, other related drawings can also be obtained based on these drawings.

[0057] Figure 1 It is a schematic flowchart of the steps of the chronic disease data analysis method based on natural language processing and integrated training provided by the embodiments of the present invention;

[0058] Figure 2 It is a schematic block diagram of the structure of the computer device provided by the embodiments of the present invention. Specific Embodiments

[0059] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Usually, the components of the embodiments of the present invention described and illustrated in the drawings here can be arranged and designed in various different configurations.

[0060] The following will describe in detail the specific embodiments of the present invention with reference to the accompanying drawings.

[0061] To solve the technical problems in the foregoing background art, Figure 1 It is a schematic flowchart of the chronic disease data analysis method based on natural language processing and integrated training provided by the embodiments of the present disclosure. The chronic disease data analysis method based on natural language processing and integrated training will be introduced in detail below.

[0062] Step S201: Perform entity recognition and relationship extraction on unstructured medical texts through natural language processing technology to generate a first chronic disease instance without labeled chronic disease target values;

[0063] Step S202: Obtain a pre-trained initial analysis model, where the pre-trained initial analysis model is obtained through contrastive learning based on the first chronic disease instance through an integrated training framework, and the integrated training framework integrates the feature representations output by multiple heterogeneous models;

[0064] Step S203: Obtain a second chronic disease instance corresponding to the disease analysis index, where the second chronic disease instance is a chronic disease instance with labeled chronic disease target values;

[0065] Step S204: Based on the second chronic disease instance, perform multi-model integration optimization on the pre-trained initial analysis model to obtain a trained chronic disease recognition model corresponding to the disease analysis index;

[0066] Step S205: Obtain the chronic disease symptoms to be analyzed, and perform recognition processing on the chronic disease symptoms to be analyzed based on the trained chronic disease recognition model to obtain the chronic disease data analysis result of the chronic disease symptoms to be analyzed.

[0067] In an embodiment of the present invention, exemplarily, a large number of patient medical records are stored in the electronic medical record system of a hospital. These medical records mostly exist in the form of unstructured text, such as documents scanned after being handwritten by doctors, or disease descriptions freely input by doctors in the system. After receiving these unstructured medical texts, the server uses entity recognition algorithms in natural language processing technology to identify various entities related to chronic diseases from the texts, such as disease names and related physical sign indicators like "hypertension", "diabetes", "blood glucose value", "blood pressure value", etc. At the same time, through relationship extraction algorithms, the relationships between these entities are sorted out, such as "patient - has - hypertension", "hypertension - is related to - blood pressure value". After such processing, the server integrates the extracted and sorted information to generate a first chronic disease instance without marked chronic disease target values. For example, in a medical record that describes "the patient has felt dizzy recently, the measured blood pressure is 160 / 100 mmHg, and has a history of hypertension for many years", the server identifies entities such as "dizziness", "hypertension", "blood pressure value 160 / 100 mmHg", and relationships such as "patient - has a history of - hypertension", "hypertension - is associated with - blood pressure value" to form a first chronic disease instance. However, at this time, there is no marked chronic disease target value such as whether the hypertension diagnosis standard is met in this instance. The server has previously used a large number of similar generated first chronic disease instances for preliminary training through an integrated training framework. In this process, the server is like a knowledgeable researcher, integrating multiple different types (heterogeneous) of models, such as convolutional neural networks (CNNs), recurrent neural networks (RNNs) in deep learning, and decision tree models in traditional machine learning. These models each extract features and learn from the first chronic disease instances from different perspectives. For example, the CNN model may be better at capturing local feature patterns in the text, while the RNN model has a better grasp of the sequential information of the text. The server integrates the feature representations output by these heterogeneous models and then conducts contrastive learning. In contrastive learning, the server compares the features between different instances to find the differential features between similar and different instances. For example, by comparing the first chronic disease instances generated from the medical records of different hypertensive patients, it is found that those instances with similar blood pressure values and symptoms have certain common features, which are significantly different from the instance features of hypotensive patients. After such a training process, the server obtains an initially trained initial analysis model, which already has the ability to preliminarily analyze chronic disease-related information. The professional doctor team in the hospital has made detailed annotations on a part of the chronic disease instances according to clinical diagnosis criteria and past experience, forming second chronic disease instances. These instances not only contain information extracted from the medical records similar to the first chronic disease instances, but also marked clear chronic disease target values. For example, for the second chronic disease instance corresponding to the analysis index of hypertension disease, it is marked whether the patient is diagnosed with hypertension (yes / no), and the classification of hypertension (grade one, grade two, grade three, etc.).The server obtains these labeled second chronic disease instances from the hospital's database, providing a key basis for subsequent optimization of the initial analysis model. After obtaining the second chronic disease instances, the server uses the integrated training framework again. First, it converts the relevant information of each second chronic disease instance into a format that the model can process, such as generating a second chronic disease symptom instance topology (similar to the processing method of the first chronic disease symptom instance topology). Then, it uses the initially trained initial analysis model to identify and process each second chronic disease symptom instance topology, obtaining a second feature vector instance corresponding to each topology, and allowing the initial analysis model to conduct a preliminary analysis on these instances with standard answers. Next, based on the already trained feature recognition component, it identifies each second feature vector instance, obtaining a second symptom inference result corresponding to each second chronic disease symptom instance, that is, the final judgment of the initial analysis model on this instance. After that, the server conducts integrated optimization training on the initial analysis model according to the chronic disease target value (that is, the standard answer marked by the doctor) of each second chronic disease symptom instance and the second symptom inference result given by the model. For example, if the initial analysis model determines that a certain hypertension instance is grade 1 hypertension, but the doctor marks it as grade 2 hypertension, the server will adjust the parameters of the model to make the model more accurately identify similar situations. After such an optimization training process for a large number of second chronic disease instances, the server finally obtains a fully trained chronic disease recognition model for the disease analysis indicators (such as hypertension diagnosis and grading), and this model becomes more accurate and reliable in identifying relevant chronic disease symptoms. When a new patient comes to the hospital for treatment, the doctor records the patient's description of chronic disease-related symptoms, forms the chronic disease symptoms information to be analyzed, and uploads it to the server. For example, the new patient describes "often thirsty and fatigued recently, and the blood sugar value is found to be 11 mmol / L during a physical examination". Previously, multiple heterogeneous models (such as the sign interaction network and the sign fusion network) have been used to conduct integrated training on the first chronic disease instances to obtain an initial analysis model with preliminary analysis capabilities. On this basis, second chronic disease instances with labeled chronic disease target values are obtained, and these instances are like a "standard answer" set. During multi-model integrated optimization, first convert the second chronic disease instances into a format that the model can process (such as generating symptom instance topologies), use the initial analysis model to process and obtain second feature vector instances, and then obtain second symptom inference results through the feature recognition component. Subsequently, according to the difference between the chronic disease target value and the inference result, the parameters of the initial analysis model are adjusted. For example, in a hypertension instance, if the model's determined grade does not match the doctor's label, the parameters are adjusted. Through a large number of such optimization trainings, integrating the advantages of multiple models, the recognition accuracy of the model for chronic disease symptoms is improved, and finally a fully trained, more accurate and reliable chronic disease recognition model is obtained. After the server obtains these chronic disease symptoms to be analyzed, it immediately calls the fully trained chronic disease recognition model. The model conducts identification and processing on the chronic disease symptoms to be analyzed.In this process, the model analyzes factors such as the relationships between symptoms and the similarity to past instances. Eventually, the server obtains the chronic disease data analysis result of the chronic disease symptoms to be analyzed, such as determining that the patient may have diabetes, and gives relevant risk assessments or suggestions for further examinations, etc., providing strong support for the doctor's diagnosis.

[0068] In an embodiment of the present invention, the following implementation manners are further provided.

[0069] Obtain a first chronic disease instance, where the first chronic disease instance includes a plurality of first chronic disease symptom instances;

[0070] Determine the topology of each first chronic disease symptom instance according to the disease description information of each first chronic disease symptom instance, and determine the contrast learning feature encoding of each first chronic disease symptom instance according to the topology of each first chronic disease symptom instance;

[0071] Perform an identification process on the topology of each first chronic disease symptom instance based on a preset basic model to obtain a first feature vector instance corresponding to the topology of each first chronic disease symptom instance;

[0072] Perform an identification on each first feature vector instance based on a feature identification component that has completed training to obtain a first symptom inference result corresponding to each first chronic disease symptom instance;

[0073] Perform integrated optimization training on the preset basic model according to the contrast learning feature encoding of each first chronic disease symptom instance and the first symptom inference result to obtain an initial analysis model that has completed training.

[0074] In an embodiment of the present invention, exemplarily, when the server processes chronic disease data, it first obtains the first chronic disease instances. These instances are the results generated from the previous entity recognition and relationship extraction of unstructured medical texts through natural language processing technology. Each first chronic disease instance contains multiple first chronic disease symptom instances. For example, in the data of a certain hospital, a first chronic disease instance may cover various symptoms of a hypertensive patient, such as dizziness, headache, palpitation, etc., which respectively constitute different first chronic disease symptom instances. Then, the server determines the topology of each first chronic disease symptom instance according to the disease description information of the instance. Taking the symptom instance of "dizziness" as an example, the server will analyze the relevant information, such as the frequency of dizziness, the duration, whether it is accompanied by other symptoms, etc., and construct these information into a topological structure to show the connection between the symptom and its related factors. After that, the contrastive learning feature encoding is determined according to this topology. The server first obtains a plurality of pre-set symptom topology patterns, which are symptom combination structures with clinical diagnostic significance summarized by medical experts based on a large amount of clinical experience. The server associates the topology of the "dizziness" symptom instance with these patterns. For example, it is found that the dizziness symptom has a high correlation with the pattern of "frequent dizziness and accompanied by elevated blood pressure", so as to obtain the correlation coefficient parameter, and then determine the contrastive learning feature encoding, which highlights the association characteristics between the symptom and a specific pattern. Then, the server performs recognition processing on the topology of each first chronic disease symptom instance based on a pre-set basic model. The pre-set basic model contains multiple sign interaction networks and at least one sign fusion network. Taking the topology of the "headache" symptom instance as an example, the server obtains the sign feature vector of each symptom entity (such as headache intensity, onset location, etc.), the associated sign entity (such as whether it is related to fatigue), and the association strength vector of the pathological association relationship between them according to the disease description. Through a series of operations, such as performing feature transformation on the sign feature vector, fusing the basic sign vector and the association strength vector of the associated sign entity, etc., finally, the first feature vector instance corresponding to the topology of this symptom instance is obtained, and this vector instance comprehensively reflects various aspects of the symptom. After that, the server performs recognition on each first feature vector instance based on the feature recognition component that has completed training. This component is like a professional diagnostic assistant, analyzing the first feature vector instance. For example, for the first feature vector instance corresponding to the "palpitation" symptom, after recognition, it gives the first symptom inference result, judging the possible disease tendency or severity corresponding to the palpitation symptom. Finally, the server performs integrated optimization training on the pre-set basic model according to the contrastive learning feature encoding and the first symptom inference result of each first chronic disease symptom instance.For example, for the symptom of "chest tightness", if the contrast learning feature encoding shows a correlation with a certain typical symptom pattern, but the first symptom inference result deviates from the expectation, the server will adjust the parameters of the preset basic model. After repeated optimization of a large number of first chronic disease symptom instances, the initially trained analysis model is finally obtained, making the analysis of chronic disease symptoms more accurate.

[0075] In an embodiment of the present invention, determining the contrast learning feature encoding of each first chronic disease symptom instance according to the topology of each first chronic disease symptom instance can be implemented through the following example.

[0076] Obtain a plurality of pre-set symptom topology patterns, where each symptom topology pattern corresponds to a symptom combination structure with clinical diagnostic significance;

[0077] Associate each first chronic disease symptom instance topology with the plurality of symptom topology patterns to obtain the correlation coefficient parameter of each first chronic disease symptom instance topology;

[0078] Determine the contrast learning feature encoding of each first chronic disease symptom instance according to the correlation coefficient parameter corresponding to each first chronic disease symptom instance topology.

[0079] In an embodiment of the present invention, for example, when the server processes the topology of the first chronic disease symptom instance to determine the contrast learning feature encoding, it first obtains a plurality of pre-set symptom topology patterns. These patterns are formulated by medical experts based on long-term clinical practice and research results, and each pattern corresponds to a symptom combination structure with clinical diagnostic significance. For example, in the symptom topology pattern related to cardiovascular diseases, there may be a combination pattern of "chest pain + palpitation + aggravated after exercise", which is of great significance for judging cardiovascular diseases such as coronary heart disease; there is also a pattern of "dizziness + abnormally elevated blood pressure + headache", which is closely related to hypertensive emergencies.

[0080] Next, the server associates each first chronic disease symptom instance topology with these pre-set multiple symptom topology patterns, thereby obtaining the correlation coefficient parameters of each first chronic disease symptom instance topology. Suppose there is currently a first chronic disease symptom instance topology described as "a patient often feels palpitations, and the symptoms worsen after fatigue, accompanied by mild chest pain." The server compares this topology with each symptom topology pattern. When associated with the "chest pain + palpitations + aggravation after exercise" pattern, because "palpitations" are similar to "palpitations", "aggravated after fatigue" and "aggravated after exercise" are related, and "mild chest pain" also meets some of the characteristics of this pattern, a higher correlation coefficient is calculated by the algorithm; when compared with other irrelevant patterns (such as symptom topology patterns for respiratory diseases), the correlation coefficient is lower. In this way, the server finds the most relevant symptom topology pattern for the first chronic disease symptom instance topology and determines its correlation coefficient parameters.

[0081] Finally, the server determines the comparative learning feature coding of each first chronic disease symptom instance according to the correlation coefficient parameter corresponding to each first chronic disease symptom instance topology. Continuing with the above example, after the server obtains the correlation coefficient parameter associated with the "chest pain + palpitations + aggravation after exercise" mode, it first extracts the core feature coding of the mode, that is, the coding representing the key symptom features of "chest pain, palpitations, and aggravation after exercise". Then, the control feature coding that is not associated with the topology of the target chronic disease symptom instance in multiple symptom topology patterns is determined, such as the feature coding of other respiratory disease symptom topology patterns. The server performs feature coupling processing on the core feature coding and the control feature coding, and combines the coding representing the key features of the relevant pattern with the coding representing the features of the irrelevant pattern with a specific algorithm, thereby generating the comparative learning feature coding of the first chronic disease symptom instance. This coding not only highlights the correlation between the symptom instance and the specific clinical diagnosis pattern, but also strengthens its unique characteristics by comparing with the irrelevant pattern, providing more targeted information for subsequent analysis and diagnosis.

[0082] In the embodiment of the present invention, associating each first chronic disease symptom instance topology with the multiple symptom topology patterns to obtain the association coefficient parameter of each first chronic disease symptom instance topology can be implemented through the following examples.

[0083] Associating the target first chronic disease symptom instance topology with each symptom topology pattern to obtain a correlation coefficient corresponding to each symptom topology pattern;

[0084] The symptom topology pattern whose correlation coefficient exceeds the correlation coefficient threshold is used as the target symptom topology pattern corresponding to the target first chronic disease symptom instance topology;

[0085] Take the target symptom topology pattern corresponding to the target first chronic disease symptom instance topology as the correlation coefficient parameter of the target first chronic disease symptom instance topology.

[0086] In an embodiment of the present invention, by way of example, when the server processes the association between the first chronic disease symptom instance topology and the symptom topology pattern, it operates by taking a specific target first chronic disease symptom instance topology as an example. Suppose this target first chronic disease symptom instance topology describes the symptoms of a patient: "The patient has frequently felt dizzy recently, accompanied by tinnitus, and the dizziness worsens when standing." First, the server associates this target first chronic disease symptom instance topology with each pre-set symptom topology pattern to obtain the correlation coefficient corresponding to each symptom topology pattern. For example, among many symptom topology patterns, there is a "dizziness + elevated blood pressure + headache" pattern for hypertension, a "tinnitus + hearing loss + ear pain" pattern for ear diseases, and a "dizziness when standing + fatigue + sudden drop in blood pressure" pattern for orthostatic hypotension, etc. The server uses a specific algorithm to calculate the correlation coefficient by analyzing factors such as the matching degree of the symptoms in the target instance topology with the symptoms in each pattern and the similarity of the logical relationships between the symptoms. For the "dizziness + elevated blood pressure + headache" pattern, since the target instance has the symptom of "dizziness" but does not mention elevated blood pressure and headache, the calculated correlation coefficient is relatively low; for the "tinnitus + hearing loss + ear pain" pattern, although there is the symptom of "tinnitus", there is a lack of descriptions related to hearing loss and ear pain, so the correlation coefficient is not high either; while for the "dizziness when standing + fatigue + sudden drop in blood pressure" pattern, "dizziness when standing" is a perfect match, and "dizziness" and "fatigue" often occur together in some cases of orthostatic hypotension, so the correlation coefficient corresponding to this pattern is relatively high. Next, the server sets a correlation coefficient threshold, which is obtained based on medical experience and a large amount of data statistics and is used to screen out the pattern most relevant to the target instance topology. The server takes the symptom topology pattern with a correlation coefficient exceeding the correlation coefficient threshold as the target symptom topology pattern corresponding to the target first chronic disease symptom instance topology. Suppose the correlation coefficient threshold is set to 0.6. In the above calculation, the correlation coefficient of the "dizziness when standing + fatigue + sudden drop in blood pressure" pattern reaches 0.7, exceeding the threshold, while the correlation coefficients of other patterns do not reach it, so the "dizziness when standing + fatigue + sudden drop in blood pressure" pattern is determined as the target symptom topology pattern. Finally, the server takes the target symptom topology pattern corresponding to the target first chronic disease symptom instance topology as the correlation coefficient parameter of this target first chronic disease symptom instance topology. This is because this target symptom topology pattern can best represent the characteristic tendency of the current target first chronic disease symptom instance topology. Subsequent operations such as contrast learning feature encoding based on this correlation coefficient parameter (i.e., the target symptom topology pattern) can provide more targeted and accurate information for analyzing this target first chronic disease symptom instance, helping to more accurately judge the patient's condition and the chronic diseases they may have.

[0087] In an embodiment of the present invention, determining the contrast learning feature encoding of each first chronic disease symptom instance according to the correlation coefficient parameter corresponding to each first chronic disease symptom instance topology can be implemented through the following examples.

[0088] Obtain the core feature encoding of the target symptom topology pattern corresponding to the target first chronic disease symptom instance topology;

[0089] Determine the control feature encoding that is not associated with the target chronic disease symptom instance topology among the multiple symptom topology patterns according to the core feature encoding;

[0090] Perform feature coupling processing on the core feature encoding and the control feature encoding to obtain the contrast learning feature encoding of the target first chronic disease symptom instance topology.

[0091] In an embodiment of the present invention, by way of example, taking a server processing a specific topological example of the first chronic disease symptom as an example, assume that the target topological example of the first chronic disease symptom is described as "the patient often feels chest tightness, which worsens after activity and is occasionally accompanied by palpitations". The server has determined that its corresponding target symptom topological pattern is "chest tightness + aggravated after activity + palpitations", and this pattern has a relatively high correlation with coronary heart disease. First, the server obtains the core feature encoding of the target symptom topological pattern corresponding to the target topological example of the first chronic disease symptom. For the target symptom topological pattern of "chest tightness + aggravated after activity + palpitations", its core feature encoding is a set of code information that can represent these three key symptoms and their mutual relationships. The server, according to the preset encoding rules, converts "chest tightness", "aggravated after activity", and "palpitations" into specific feature codes respectively, and combines them into the core feature encoding according to their logical relationships in the pattern. For example, "chest tightness" corresponds to code A, "aggravated after activity" corresponds to code B, and "palpitations" corresponds to code C. The core feature encoding may be arranged in the order of ABC and carry metadata indicating their association relationships. Next, the server determines the control feature encoding in multiple symptom topological patterns that is not associated with the target topological example of the chronic disease symptom according to the core feature encoding. The server traverses all the preset symptom topological patterns to find patterns that are quite different from the current target symptom topological pattern. For example, for the symptom topological pattern of "cough + expectoration + dyspnea" for respiratory diseases, its relevance to the current target pattern related to coronary heart disease is low. The server also converts "cough", "expectoration", and "dyspnea" into feature codes according to the encoding rules, assuming they are D, E, and F respectively, and forms the control feature encoding DEF and related association metadata. Finally, the server performs feature coupling processing on the core feature encoding and the control feature encoding to obtain the comparative learning feature encoding of the target topological example of the first chronic disease symptom. The server uses a specific algorithm to fuse the core feature encoding ABC and the control feature encoding DEF. For example, through a weighted fusion method, a higher weight is assigned to each code in the core feature encoding, and a lower weight is assigned to the codes in the control feature encoding, and then they are combined in a certain order and mathematical operations are performed to generate a new encoding. Assume that the operation rule is to perform weighted addition on the code values at the corresponding positions to obtain a new set of numerical values, and then through normalization processing, finally form the comparative learning feature encoding of the target topological example of the first chronic disease symptom. This encoding not only highlights the core features closely related to the target symptom topological pattern but also reflects the differences from other irrelevant patterns through the control feature encoding, enabling subsequent analysis to more accurately distinguish this symptom example from other situations and assisting the server to perform more accurate chronic disease analysis and diagnosis based on this encoding.

[0092] In an embodiment of the present invention, the preset basic model includes a plurality of physical sign interaction networks and at least one physical sign fusion network. The recognition processing of each first chronic disease symptom instance topology based on the preset basic model to obtain the first feature vector instance corresponding to each first chronic disease symptom instance topology can be implemented through the following examples.

[0093] According to the disease description information of the target chronic disease symptom instance, obtain the physical sign feature vector of each symptom entity in the target first chronic disease symptom instance topology, the associated physical sign entity of each symptom entity, and the association strength vector of the pathological association relationship between each symptom entity and the associated physical sign entity;

[0094] Perform a feature transformation operation on the physical sign feature vector of each symptom entity to obtain the basic physical sign vector of each symptom entity;

[0095] Based on the plurality of physical sign interaction networks, perform an aggregation process on the basic physical sign vector of each symptom entity, the basic physical sign vector of the associated physical sign entity, and the association strength vector to obtain the feature vector of each symptom entity;

[0096] Based on the physical sign fusion network, process the feature vector of each symptom entity to obtain the first feature vector instance corresponding to the target first chronic disease symptom instance topology.

[0097] In an embodiment of the present invention, by way of example, assume that the server is processing a chronic disease symptom instance of a patient, and the target chronic disease symptom instance is described as "the patient has long suffered from hypertension, recently had dizziness symptoms, and the blood pressure value increased significantly during dizziness, and was accompanied by occasional palpitations". The server performs recognition processing on this target first chronic disease symptom instance topology based on a preset basic model to obtain a corresponding first feature vector instance. For the symptom entity of "dizziness", the server extracts relevant information from the disease description information. Through the analysis and learning of a large amount of clinical data, the server knows that dizziness may be related to the blood pressure value. Therefore, the "blood pressure value" is the associated physical sign entity of the "dizziness" symptom entity. It can be seen from the description that the blood pressure increases significantly during dizziness. The server quantifies this degree of association into an association strength vector according to a specific algorithm, such as [0.8] (the higher the value, the stronger the association). At the same time, for "dizziness" itself, the server converts the characteristics of dizziness (such as the frequency and degree of dizziness) into a physical sign feature vector according to clinical knowledge and data, assumed to be [0.6, 0.4], which represent the quantified values of the frequency and degree of dizziness respectively. For the symptom entity of "palpitations", the server analyzes and learns that it may be associated with physical signs related to heart function. In this case, the associated physical sign entity can be the "heart rate". Since only occasional palpitations are mentioned in the description, the server determines its association strength vector with the heart rate, for example, [0.5]. The physical sign feature vector of "palpitations" itself is quantified as [0.3, 0.7] according to the frequency and intensity of palpitations. As a symptom entity, the associated physical sign entity of "hypertension" can be the "blood pressure control situation", and the association strength vector is assumed to be [0.9] because hypertension is closely related to blood pressure control. The physical sign feature vector of "hypertension" can be quantified as [0.7, 0.5] according to the course of hypertension and the range of blood pressure fluctuations. For the physical sign feature vector [0.6, 0.4] of "dizziness", the server uses a preset conversion algorithm, which can be a linear transformation including a weight matrix. After calculation, the basic physical sign vector is obtained, assumed to become [0.5, 0.6]. This conversion process is to convert the original physical sign feature vector into a form more suitable for subsequent processing, highlighting the key features related to disease diagnosis. The physical sign feature vector [0.3, 0.7] of "palpitations" undergoes the same feature conversion operation to obtain a basic physical sign vector, such as [0.4, 0.8]. The physical sign feature vector [0.7, 0.5] of "hypertension" is converted to obtain a basic physical sign vector, assumed to be [0.8, 0.4]. Taking "dizziness" as an example, multiple physical sign interaction networks will consider the basic physical sign vector [0.5, 0.6] of "dizziness", the basic physical sign vector corresponding to the "blood pressure value" (associated physical sign entity) (assuming the blood pressure value is quantified as [0.7, 0.3] according to some measurement data), and the association strength vector [0.8]. The physical sign interaction network passes through a specific aggregation algorithm, which can be weighted summation combined with a non-linear transformation.After multiplying each element of the "dizziness" basic physical sign vector by the association strength vector, weighted addition is performed with the corresponding elements of the "blood pressure value" basic physical sign vector, and then through non-linear function processing, the feature vector of the "dizziness" symptom entity is obtained, assumed to be [0.65, 0.55]. For "palpitation", combining its basic physical sign vector [0.4, 0.8], the basic physical sign vector of the "heart rate" (associated physical sign entity) (assumed to be [0.6, 0.2]) and the association strength vector [0.5], through the aggregation processing of the physical sign interaction network, the feature vector of the "palpitation" symptom entity is obtained, such as [0.45, 0.75]. "Hypertension" combines its own basic physical sign vector [0.8, 0.4], the basic physical sign vector of the "blood pressure control situation" (associated physical sign entity) (assumed to be [0.9, 0.1]) and the association strength vector [0.9], and through the aggregation processing, the feature vector is obtained, assumed to be [0.88, 0.35]. The server inputs the feature vectors of the three symptom entities of "dizziness" [0.65, 0.55], "palpitation" [0.45, 0.75], and "hypertension" [0.88, 0.35] into the physical sign fusion network. The physical sign fusion network will comprehensively consider the relationships between these feature vectors, and through a series of operations, such as matrix multiplication, convolution operations, etc., fuse the features of each symptom entity. Finally, a unified vector is output, that is, the first feature vector instance corresponding to the topology of the target first chronic disease symptom instance, assumed to be [0.6, 0.5, 0.7, 0.4]. This first feature vector instance comprehensively reflects the features of each symptom entity and their mutual relationships in the topology of the target first chronic disease symptom instance, providing a key data basis for subsequent symptom inference and disease analysis.

[0098] In the embodiment of the present invention, the aggregation processing of the basic physical sign vector of each symptom entity, the basic physical sign vector of the associated physical sign entity, and the association strength vector based on the multiple physical sign interaction networks to obtain the feature vector of each symptom entity can be executed through the following examples.

[0099] Based on the first physical sign interaction network, the basic physical sign vector of the associated physical sign entity and the association strength vector are interactively aggregated to obtain the first-order interactive aggregation vector of each symptom entity;

[0100] Perform physical sign aggregation processing on the input physical sign vector of each symptom entity and the first-order interactive aggregation vector to obtain the first-order physical sign vector of each symptom entity;

[0101] Based on the target physical sign interaction network, the previous-order physical sign vector of the associated physical sign entity and the association strength vector are interactively aggregated to obtain the target-order interactive aggregation vector of each symptom entity;

[0102] Perform sign aggregation processing on the previous-order sign vectors of each of the symptom entities and the target-order interaction aggregation vectors to obtain the target-order sign vectors of each of the symptom entities;

[0103] Use the target-order sign vectors of each of the symptom entities as the feature vectors of each of the symptom entities.

[0104] In an embodiment of the present invention, by way of example, assume that the server is processing a target first chronic disease symptom instance topology described as "the patient has recently experienced headache symptoms, the severity of the headache increases with the increase in blood pressure, and at the same time is accompanied by neck stiffness, and the neck stiffness is related to the frequency of headache attacks". For the "headache" symptom entity, its associated sign entity is "blood pressure". Assume that the basic sign vector of "blood pressure" obtained through the previous steps is [0.7, 0.3] (representing the quantitative characteristics of systolic blood pressure and diastolic blood pressure respectively), and the association strength vector between headache and blood pressure is [0.8] (indicating the degree of association that the severity of headache increases with the increase in blood pressure). The first sign interaction network will, according to a preset algorithm, such as multiplying each element of the association strength vector by the basic sign vector of "blood pressure", to obtain [0.56, 0.24], which is the first-order interaction aggregation vector of the "headache" symptom entity. This vector reflects the initial influence characteristics of blood pressure on headache based on the association strength. For the "neck stiffness" symptom entity, its associated sign entity can be regarded as "the frequency of headache attacks". Assume that the basic sign vector of "the frequency of headache attacks" is [0.6] (simply quantified as a value of the attack frequency), and the association strength vector between neck stiffness and the frequency of headache attacks is [0.7]. The first sign interaction network multiplies the two through the same or a similar algorithm to obtain [0.42], which is the first-order interaction aggregation vector of the "neck stiffness" symptom entity, reflecting the initial influence of the frequency of headache attacks on neck stiffness. Taking "headache" as an example, its input sign vector (i.e., the basic sign vector) is [0.5, 0.6] (representing the quantitative characteristics of the severity and duration of headache respectively). The server obtains a specific weight tensor for sign aggregation processing, assumed to be [0.4, 0.6]. Perform a linear transformation operation on the first-order interaction aggregation vector [0.56, 0.24] and the weight tensor, that is, multiply the corresponding elements and then add them together, to obtain [0.56 * 0.4 + 0.24 * 0.6] = [0.368]. Then, based on a pre-set non-linear mapping function (such as the Sigmoid function), process the input sign vector [0.5, 0.6] and this transformation coefficient tensor [0.368]. Combine each element of the input sign vector with the transformation coefficient tensor and calculate through the Sigmoid function to obtain the first-order sign vector of the "headache" symptom entity, assumed to be [0.65, 0.7]. This first-order sign vector combines the characteristics of headache itself and the initial influence of blood pressure on it. For "neck stiffness", its input sign vector is [0.4, 0.7] (representing the quantitative characteristics of the degree and range of neck stiffness respectively), and the weight tensor for sign aggregation processing is assumed to be [0.3, 0.7]. The first-order interaction aggregation vector [0.42] is linearly transformed with the weight tensor to obtain [0.42 * 0.3 + 0.42 * 0.7] = [0.42]. After being processed by the non-linear mapping function, the first-order sign vector of the "neck stiffness" symptom entity is obtained, such as [0.55, 0.8], which combines the characteristics of neck stiffness itself and the initial influence of the frequency of headache attacks on it.Taking "headache" as an example again, the previous - stage sign vector here is the first - stage sign vector [0.65, 0.7] just obtained. The associated sign entity is still "blood pressure", and the associated intensity vector remains [0.8]. The target sign interaction network adopts a more complex algorithm. For example, it performs a weighted convolution operation on the previous - stage sign vector and the associated intensity vector. Suppose the result is [0.52, 0.56], which is the target - stage interaction aggregation vector of the "headache" symptom entity. This vector further explores the deeper associated features between blood pressure and headache on the previous basis. For "neck stiffness", the previous - stage sign vector is [0.55, 0.8], the associated sign entity is "headache attack frequency", and the associated intensity vector is [0.7]. After similar complex processing by the target sign interaction network, suppose the result is [0.44, 0.56], which is used as the target - stage interaction aggregation vector of the "neck stiffness" symptom entity, reflecting the further associated characteristics between headache attack frequency and neck stiffness on the previous basis. For "headache", the previous - stage sign vector [0.65, 0.7] and the target - stage interaction aggregation vector [0.52, 0.56] are subjected to sign aggregation processing. The server obtains a suitable weight tensor again, suppose it is [0.5, 0.5]. Through linear transformation and non - linear mapping processing on the two vectors and the weight tensor (similar to the previous steps), the target - stage sign vector of the "headache" symptom entity is obtained, suppose it is [0.6, 0.63]. This target - stage sign vector more comprehensively integrates the characteristics of headache itself and the multi - level influence of blood pressure on it. For "neck stiffness", the previous - stage sign vector [0.55, 0.8] and the target - stage interaction aggregation vector [0.44, 0.56], under the action of the corresponding weight tensor (suppose it is [0.4, 0.6]), through linear transformation and non - linear mapping, the target - stage sign vector of the "neck stiffness" symptom entity is obtained, such as [0.5, 0.7], which comprehensively reflects the characteristics of neck stiffness itself and the multi - aspect influence of headache attack frequency on it. Finally, the feature vector of the "headache" symptom entity is the target - stage sign vector [0.6, 0.63] obtained previously, and the feature vector of the "neck stiffness" symptom entity is [0.5, 0.7]. These feature vectors comprehensively and deeply reflect the complex relationships between each symptom entity and its associated sign entities, providing key feature data for the subsequent comprehensive analysis of the entire target first - chronic - disease symptom instance topology.

[0105] In the embodiment of the present invention, the interaction aggregation process of the basic sign vector of the associated sign entity and the associated intensity vector based on the first sign interaction network to obtain the first - stage interaction aggregation vector of each symptom entity can be implemented through the following example.

[0106] Fuse the basic sign vector of the associated sign entity and the associated intensity vector to obtain a fusion vector;

[0107] Obtain the first weight tensor of the first feature aggregation module for interactive aggregation processing, perform a linear transformation operation on the first weight tensor and the fusion vector to obtain a first transformation coefficient tensor;

[0108] Process the first transformation coefficient tensor based on a preset first non-linear mapping function to obtain a first-order interactive aggregation vector of symptom entities.

[0109] In an embodiment of the present invention, by way of example, assume that the server is processing the chronic disease symptom data of a patient. The symptom description of this patient is "recently experiencing joint pain symptoms, the degree of pain is closely related to the amount of activity, and at the same time accompanied by slight swelling, and the degree of swelling is related to the pain frequency". For the "joint pain" symptom entity, its associated sign entity is the "amount of activity". Assume that the basic sign vector of the "amount of activity" obtained in the previous step is [0.6, 0.4], which respectively represent the quantization values of daily activity time and activity intensity. The association intensity vector between joint pain and the amount of activity is [0.8], indicating the tightness of the association between the degree of pain and the amount of activity. The server performs feature fusion on the association intensity vector [0.8] and the basic sign vector [0.6, 0.4] of the "amount of activity". A simple fusion method can be to multiply the association intensity vector by each element of the basic sign vector, obtaining the fusion vector [0.48, 0.32]. This fusion vector preliminarily reflects the influence characteristics of the amount of activity on joint pain based on the association intensity. For the "swelling" symptom entity, its associated sign entity is the "pain frequency". Assume that the basic sign vector of the "pain frequency" is [0.7], which is simply quantified as the frequency value of pain attacks. The association intensity vector between swelling and pain frequency is [0.7]. The server multiplies the two in the same way to obtain the fusion vector [0.49], and this vector reflects the preliminary influence of pain frequency on swelling based on the association intensity. Taking "joint pain" as an example, the server obtains the first weight tensor of the first feature aggregation module for interactive aggregation processing, assumed to be [0.5, 0.5]. Perform a linear transformation operation on the first weight tensor [0.5, 0.5] and the fusion vector [0.48, 0.32]. The linear transformation operation can be to multiply the corresponding elements and then add them, that is, (0.5 * 0.48 + 0.5 * 0.32) = 0.4. The obtained first transformation coefficient tensor is [0.4]. This coefficient tensor comprehensively considers the information of the weight and the fusion vector, preparing for subsequent non-linear processing. For the "swelling" symptom entity, assume the first weight tensor is [0.6]. Perform a linear transformation on it and the fusion vector [0.49], that is, 0.6 * 0.49 = 0.294. The obtained first transformation coefficient tensor is [0.294], and this tensor combines the information of the weight and the information after the fusion of swelling and pain frequency. For "joint pain", the server is based on a pre-set first non-linear mapping function, such as the Sigmoid function. Substitute the first transformation coefficient tensor [0.4] into the Sigmoid function for processing. For example, it is calculated that f(0.4) ≈ 0.6. The first-order interactive aggregation vector of the "joint pain" symptom entity is obtained as [0.6]. This first-order interactive aggregation vector further adjusts the coefficient obtained from the previous linear transformation through non-linear mapping, making it more capable of reflecting the complex association characteristics between joint pain and the amount of activity. For "swelling", the Sigmoid function is also used to process the first transformation coefficient tensor [0.294].Calculate f(0.294) ≈ 0.573. The first-order interaction aggregation vector of the "swelling" symptom entity is [0.573]. This vector comprehensively reflects the non-linearly adjusted correlation characteristics between swelling and pain frequency, providing a key intermediate result for subsequent more in-depth sign aggregation processing. Through such a series of operations, the server generates first-order interaction aggregation vectors for each symptom entity, which play a connecting role in the entire sign analysis process, gradually mining and integrating the complex relationships between symptoms and related signs, and providing strong support for the accurate analysis of chronic disease symptoms.

[0110] In the embodiment of the present invention, the sign aggregation processing of the input sign vector and the first-order interaction aggregation vector of each symptom entity to obtain the first-order sign vector of each symptom entity can be implemented through the following example.

[0111] Obtain the second weight tensor of the second feature aggregation module for sign aggregation processing;

[0112] Perform a linear transformation operation on the second weight tensor and the first-order interaction aggregation vector to obtain a second transformation coefficient tensor;

[0113] Process the input sign vector of each symptom entity and the second transformation coefficient tensor based on a pre-set second non-linear mapping function to obtain the first-order sign vector of each symptom entity.

[0114] In an embodiment of the present invention, by way of example, continuing with the above case of "the patient recently presented joint pain symptoms, the degree of pain is closely related to the amount of activity, and at the same time there is slight swelling, and the degree of swelling is related to the frequency of pain", the server then performs physical sign aggregation processing on the input physical sign vector and the first-order interaction aggregation vector of each symptom entity to obtain the first-order physical sign vector of each symptom entity. For the "joint pain" symptom entity, the server obtains the second weight tensor of the second feature aggregation module for performing physical sign aggregation processing according to parameters preset in the system or trained based on a large amount of data. Suppose this second weight tensor is [0.3, 0.7], and this weight tensor reflects the importance assignment of different feature information during the physical sign aggregation process. For example, the weight setting here means that during the aggregation process, certain features related to pain may be more emphasized (the part corresponding to the weight 0.7), while another part of the features is considered relatively less (the part corresponding to the weight 0.3), which is based on the experience and algorithm setting of the analysis of the joint pain symptom. For the "swelling" symptom entity, suppose the obtained second weight tensor is [0.4, 0.6]. Similarly, this weight assignment reflects the difference in the degree of emphasis on different related features when processing the swelling physical sign aggregation, and may focus more on the features related to the degree of swelling (weight 0.6), while paying relatively less attention to other related features (weight 0.4). Taking "joint pain" as an example, the previously obtained first-order interaction aggregation vector of "joint pain" is [0.6] (suppose it is obtained after the previous steps of processing). Perform a linear transformation operation on the second weight tensor [0.3, 0.7] and the first-order interaction aggregation vector [0.6]. Since the first-order interaction aggregation vector has only one value, it can be multiplied by each element of the weight tensor and then added together (similar to a simplified form of matrix multiplication), that is, (0.3 * 0.6 + 0.7 * 0.6) = 0.6. In this way, the second transformation coefficient tensor is obtained as [0.6]. This second transformation coefficient tensor integrates the importance information represented by the first-order interaction aggregation vector and the weight tensor, providing a basis for subsequent further processing. For the "swelling" symptom entity, its first-order interaction aggregation vector is supposed to be [0.573]. Perform a linear transformation operation on it with the second weight tensor [0.4, 0.6], that is, (0.4 * 0.573 + 0.6 * 0.573) = 0.573. The obtained second transformation coefficient tensor is [0.573], and this tensor combines the first-order interaction aggregation vector of swelling and the feature importance information represented by the weight tensor. For "joint pain", suppose its input physical sign vector is [0.5, 0.8], which respectively represent the quantified values of the degree and duration of joint pain. The server is based on a pre-set second non-linear mapping function, such as the ReLU function (Rectified Linear Unit, f(x) = max(0, x)). Combine each element of the input physical sign vector with the second transformation coefficient tensor [0.6].First, perform an operation (such as addition) on the first element 0.5 and 0.6 of the input sign vector, getting 0.5 + 0.6 = 1.1. Then, after processing by the ReLU function, f(1.1) = 1.1. Perform the same operation on the second element 0.8, 0.8 + 0.6 = 1.4, f(1.4) = 1.4. Thus, the first-order sign vector of the "joint pain" symptom entity is [1.1, 1.4]. This first-order sign vector not only contains the input sign information of joint pain itself but also integrates the first-order interaction aggregation information related to the activity level and is adjusted through non-linear mapping, more comprehensively and accurately reflecting the comprehensive sign characteristics of joint pain. For the "swelling" symptom entity, its input sign vector is assumed to be [0.4, 0.7], representing the quantification values of the swelling range and hardness respectively. Combine the input sign vector with the second transformation coefficient tensor [0.573]. For the first element, 0.4 + 0.573 = 0.973, after processing by the ReLU function, f(0.973) = 0.973. For the second element, 0.7 + 0.573 = 1.273, f(1.273) = 1.273. The first-order sign vector of the "swelling" symptom entity is [0.973, 1.273]. This first-order sign vector synthesizes the original signs of swelling, the first-order interaction aggregation information related to the pain frequency, and the adjustment after non-linear mapping, providing a more representative feature vector for the subsequent further analysis of the swelling symptom. Through this series of operations, the server generates first-order sign vectors for each symptom entity, and these vectors further refine and integrate sign information during the chronic disease symptom analysis process, helping to more accurately analyze the patient's condition.

[0115] In the embodiment of the present invention, the obtaining of the sign feature vector of each symptom entity, the associated sign entity of each symptom entity, and the association strength vector of the pathological association relationship between each symptom entity and the associated sign entity in the target first chronic disease symptom instance topology according to the disease description information of the target chronic disease symptom instance can be implemented through the following example.

[0116] According to the disease description information of the target first chronic disease symptom instance, obtain the clinical index data of each sub-symptom instance in the target first chronic disease symptom instance;

[0117] Perform feature coupling processing on the clinical index data of each sub-symptom instance to obtain the sign feature vector of each symptom entity in the target first chronic disease symptom instance topology;

[0118] Use the control sub-symptom having a clinical association path with each sub-symptom instance as the associated sub-symptom of each sub-symptom instance;

[0119] Obtain the clinical index data of the clinical association path between each sub-symptom instance and the associated sub-symptom;

[0120] Perform feature coupling processing on the clinical index data of the clinical association path to obtain the association strength vector of the pathological association relationship between each symptom entity and the associated sign entity.

[0121] In an embodiment of the present invention, by way of example, assume that the server is processing the description of a patient's chronic disease symptoms: "The patient has had diabetes for many years. Recently, symptoms such as excessive drinking, excessive eating have occurred. At the same time, blood sugar fluctuates greatly, and the weight has decreased. There is a certain correlation between the weight loss and the blood sugar fluctuation." For the sub-symptom instance of "excessive drinking", the server obtains clinical index data from the disease description information and relevant medical records. For example, it is found that the patient's daily water intake reaches 3000 ml, and this specific value is an important clinical index data for the sub-symptom instance of "excessive drinking". For the sub-symptom instance of "excessive eating", it is obtained that the patient's food intake per meal has increased by 50% recently, and this food intake change ratio is the clinical index data of "excessive eating". For "large blood sugar fluctuations", the server obtains that the patient's fasting blood sugar values have fluctuated between 6-10 mmol / L in the recent week, and this range of blood sugar values is the clinical index data of this sub-symptom instance. Regarding "weight loss", it is known that the patient's weight has decreased by 5 kg in the recent month, and this 5 kg weight change amount is the clinical index data of the sub-symptom instance of "weight loss". For "excessive drinking", the server converts the daily water intake of 3000 ml into a feature vector according to a preset feature coupling algorithm. Assume that the algorithm compares the water intake with the normal range and combines the common water intake characteristics of diabetic patients to generate a physical sign feature vector [0.8, 0.2]. Among them, 0.8 may represent the degree of exceeding the normal water intake, and 0.2 represents the degree of conformity with the typical excessive drinking characteristics of diabetes. For "excessive eating", the data of a 50% increase in food intake per meal is converted into a physical sign feature vector through the feature coupling algorithm, such as [0.7, 0.3]. 0.7 reflects the amplitude of the increase in food intake, and 0.3 reflects the degree of closeness of the association with the excessive eating symptoms of diabetes. The blood sugar value range of 6-10 mmol / L for "large blood sugar fluctuations" may obtain a physical sign feature vector [0.9, 0.1] after feature coupling processing. 0.9 highlights the degree of exceeding the normal range of blood sugar fluctuations, and 0.1 represents the consistency with the blood sugar fluctuation characteristics of diabetes. The 5 kg weight change amount for "weight loss" generates a physical sign feature vector [0.6, 0.4] through the feature coupling algorithm. 0.6 represents the amplitude of weight loss, and 0.4 represents the degree of fit with the weight loss symptoms of diabetes. For the sub-symptom instance of "excessive drinking", based on medical knowledge and a large amount of case data, the server identifies that "large blood sugar fluctuations" has a clinical association path with it, because an increase in blood sugar in diabetic patients stimulates the thirst center, leading to excessive drinking. Therefore, "large blood sugar fluctuations" is the associated sub-symptom of "excessive drinking". Similarly, "excessive eating" also has a clinical association path with "large blood sugar fluctuations". In a hyperglycemic state, body cells do not get enough energy, which stimulates the appetite center and causes excessive eating. Therefore, "large blood sugar fluctuations" is the associated sub-symptom of "excessive eating".There is also a clinical association path between "weight loss" and "significant blood glucose fluctuations". Large blood glucose fluctuations lead to metabolic disorders in the body, increased breakdown of fat and protein, thus causing weight loss. Therefore, "significant blood glucose fluctuations" is an associated sub-symptom of "weight loss". Regarding the clinical association path between "polydipsia" and "significant blood glucose fluctuations", the server obtains data showing that the water intake of the patient increases significantly when the blood glucose level rises. For example, when the blood glucose level rises from 7 mmol / L to 9 mmol / L, the water intake increases from 2500 ml to 3500 ml. This set of corresponding change data of blood glucose values and water intake is the clinical index data of their clinical association path. In the clinical association path between "polyphagia" and "significant blood glucose fluctuations", the server obtains data showing that the food intake of the patient increases when the blood glucose level rises. For example, when the blood glucose level rises from 6 mmol / L to 8 mmol / L, the food intake per meal increases from 200 g to 300 g. This set of data of blood glucose values and food intake changes is the clinical index data of this clinical association path. Regarding the clinical association path between "weight loss" and "significant blood glucose fluctuations", data on the weight loss rate during blood glucose fluctuations is obtained. For example, within half a month when the blood glucose fluctuates between 8 - 10 mmol / L, the weight drops by 3 kg. This set of data of blood glucose fluctuation range and weight loss amount is the clinical index data of this clinical association path. For the clinical association path data between "polydipsia" and "significant blood glucose fluctuations", the server processes it through a feature coupling algorithm. Considering comprehensively the relationship between the increase amplitude of blood glucose and the increase amplitude of water intake, an association intensity vector [0.7] is generated. This value indicates a relatively strong association degree between polydipsia and blood glucose fluctuations. The clinical association path data between "polyphagia" and "significant blood glucose fluctuations", after feature coupling processing, generates an association intensity vector [0.6], reflecting a certain intensity of association between polyphagia and blood glucose fluctuations. The clinical association path data between "weight loss" and "significant blood glucose fluctuations", after feature coupling processing, obtains an association intensity vector [0.8], indicating a relatively high degree of association between weight loss and blood glucose fluctuations. Through these steps, the server comprehensively and meticulously obtains the physical sign feature vectors of each symptom entity in the target first chronic disease symptom instance topology, the associated physical sign entities, and the association intensity vectors of the pathological association relationships between them, providing a key data basis for subsequent analysis and processing based on the pre-set basic model.

[0122] In an embodiment of the present invention, the initial analysis model that has been pre-trained is optimized based on the second chronic disease instance to obtain the completed training chronic disease recognition model corresponding to the disease analysis index, which can be implemented through the following examples.

[0123] Obtain the second chronic disease symptom instance topology of each second chronic disease symptom instance in the second chronic disease instance;

[0124] Performing identification processing on each second chronic disease symptom instance topology based on the pre-trained initial analysis model to obtain a second feature vector instance corresponding to each second chronic disease symptom instance topology;

[0125] Performing identification on each second feature vector instance based on the trained feature identification component to obtain a second symptom inference result corresponding to each second chronic disease symptom instance;

[0126] Performing integrated optimization training on the pre-trained initial analysis model according to the chronic disease target value of each second chronic disease symptom instance and the second symptom inference result to obtain a trained chronic disease identification model corresponding to the disease analysis index.

[0127] In an embodiment of the present invention, by way of example, it is assumed that the server is processing the analysis of chronic disease data related to diabetes and has obtained a batch of second chronic disease instances, which have been marked with chronic disease target values by professional doctors according to clinical diagnostic criteria and experience. Taking a specific second chronic disease instance as an example, the instance is described as "the patient has a history of diabetes for many years, recently has symptoms of blurred vision, and the blood glucose value has been continuously between 10 - 12 mmol / L, the glycated hemoglobin index is 8%, and the feeling of physical fatigue is obvious". The server constructs a topology of the second chronic disease symptom instance according to the disease description information. For the symptom of "blurred vision", the server analyzes related factors, such as the frequency and degree of blurred vision, and whether there is a temporal association with the change in blood glucose value, etc., and constructs these information into a topological structure. For "the blood glucose value has been continuously between 10 - 12 mmol / L", the blood glucose value range, fluctuation situation, and the relationship with other symptoms are recorded to form a topological structure. Similarly, similar processing is also carried out for "the glycated hemoglobin index is 8%" and "the feeling of physical fatigue is obvious", respectively constructing the topological structures of each second chronic disease symptom instance. These topological structures show the connections between various symptoms and their related factors, forming a topology of the second chronic disease symptom instance. The server uses the initially trained initial analysis model to perform recognition processing on each topology of the second chronic disease symptom instance constructed above. For example, for the topology of the symptom instance of "blurred vision", the initial analysis model analyzes each factor in the topology according to its internal algorithms and parameters. For example, the degree of blurred vision is quantified as a numerical value, and combined with information such as its association with the change in blood glucose value, through a series of calculations and conversions, a corresponding feature vector is generated, assumed to be [0.7, 0.3, 0.1]. This vector comprehensively reflects the characteristics of the symptom of "blurred vision" and its related factors, and is the second feature vector instance corresponding to the topology of the "blurred vision" symptom instance. Similarly, for the topology of the symptom instance of "the blood glucose value has been continuously between 10 - 12 mmol / L", the model analyzes characteristics such as the blood glucose value range and fluctuation, and generates a corresponding second feature vector instance, such as [0.8, 0.2]. Similar processing is also carried out for the topology of the symptom instances of "the glycated hemoglobin index is 8%" and "the feeling of physical fatigue is obvious", respectively obtaining their corresponding second feature vector instances. The server uses the trained feature recognition component to recognize each second feature vector instance. For example, for the second feature vector instance [0.7, 0.3, 0.1] of "blurred vision", the feature recognition component judges the degree of association between the symptom represented by this feature vector and diabetic retinopathy according to the knowledge and patterns learned during its training, and gives a second symptom inference result, such as "this symptom of blurred vision is moderately related to diabetic retinopathy".For the second eigenvector instance [0.8, 0.2] of "blood glucose level continuously between 10 - 12 mmol / L", the feature recognition component determines the degree of influence of the current blood glucose level on the control of diabetes condition, and obtains the second symptom inference result of "poor blood glucose control". The second eigenvector instances corresponding to "glycated hemoglobin index is 8%" and "obvious physical fatigue" are also recognized, and the corresponding second symptom inference results are given respectively. It is known that each second chronic disease symptom instance has a chronic disease target value annotated by a professional doctor. For example, for the symptom of "blurred vision" in the above instance, the chronic disease target value annotated by the doctor can be "high risk of diabetic retinopathy". The second symptom inference result given by the model is "this blurred vision symptom is moderately related to diabetic retinopathy", and there is a difference between the two. The server adjusts the parameters of the initial analysis model according to this difference. For example, in the algorithm modules related to blurred vision and diabetic retinopathy in the model, the relevant weight parameters are adjusted so that when the model encounters a similar symptom instance topology next time, it can more accurately infer a result consistent with the doctor's annotation. For other symptoms in this second chronic disease instance, such as "blood glucose level continuously between 10 - 12 mmol / L", "glycated hemoglobin index is 8%", "obvious physical fatigue", etc., similar parameter adjustments are made to the initial analysis model according to the difference between the chronic disease target value and the second symptom inference result. Through such an optimization training process for a large number of second chronic disease instances, the server gradually optimizes the initial analysis model, and finally obtains a completed training chronic disease recognition model for diabetes disease analysis indicators. This model becomes more accurate and reliable in identifying diabetes-related chronic disease symptoms and analyzing the condition.

[0128] An embodiment of the present invention provides a computer device 100. The computer device 100 includes a processor and a non-volatile memory storing computer instructions. When the computer instructions are executed by the processor, the computer device 100 executes the aforementioned chronic disease data analysis method based on natural language processing and integrated training. As Figure 2 shown, Figure 2 is a structural block diagram of the computer device 100 provided by the embodiment of the present invention. The computer device 100 includes a memory 111, a processor 112, and a communication unit 113. To achieve data transmission or interaction, the elements of the memory 111, the processor 112, and the communication unit 113 are electrically connected to each other directly or indirectly. For example, these elements can be electrically connected to each other through one or more communication buses or signal lines.

[0129] For purposes of illustration, the foregoing description has been made with reference to specific embodiments. However, the foregoing illustrative discussion is not intended to be exhaustive or to limit the disclosure to the precise forms disclosed. Numerous modifications and variations are possible in light of the above teachings. The embodiments were chosen and described in order to best illustrate the principles of the disclosure and its practical application, to thereby enable others skilled in the art to best utilize the disclosure and various embodiments with various modifications as are suited to the particular application contemplated.

Claims

1. A chronic disease data analysis method based on natural language processing and integrated training, characterized in that Including: Performing entity recognition and relationship extraction on unstructured medical texts through natural language processing technology to generate a first chronic disease instance without labeled chronic disease target values; Obtaining a pre-trained initial analysis model, where the pre-trained initial analysis model is obtained through contrastive learning based on the first chronic disease instance by an integrated training framework, and the integrated training framework integrates the feature representations output by multiple heterogeneous models; Obtaining a second chronic disease instance corresponding to a disease analysis index, where the second chronic disease instance is a chronic disease instance with labeled chronic disease target values; Performing multi-model integration optimization on the pre-trained initial analysis model based on the second chronic disease instance to obtain a trained chronic disease recognition model corresponding to the disease analysis index; Obtaining chronic disease symptoms to be analyzed, and performing recognition processing on the chronic disease symptoms to be analyzed based on the trained chronic disease recognition model to obtain a chronic disease data analysis result of the chronic disease symptoms to be analyzed.

2. The method according to claim 1, wherein The method further includes: Obtaining a first chronic disease instance, where the first chronic disease instance includes multiple first chronic disease symptom instances; Determining a first chronic disease symptom instance topology for each first chronic disease symptom instance according to the disease description information of each first chronic disease symptom instance, and determining a contrastive learning feature encoding for each first chronic disease symptom instance according to each first chronic disease symptom instance topology; Performing recognition processing on each first chronic disease symptom instance topology based on a preset basic model to obtain a first feature vector instance corresponding to each first chronic disease symptom instance topology; Performing recognition on each first feature vector instance based on a trained feature recognition component to obtain a first symptom inference result corresponding to each first chronic disease symptom instance; Performing integrated optimization training on the preset basic model according to the contrastive learning feature encoding and the first symptom inference result of each first chronic disease symptom instance to obtain a trained initial analysis model.

3. The method according to claim 2, wherein The determining the contrastive learning feature encoding for each first chronic disease symptom instance according to each first chronic disease symptom instance topology includes: Obtaining a plurality of pre-set symptom topology patterns, where each symptom topology pattern corresponds to a symptom combination structure with clinical diagnostic significance; Associating each first chronic disease symptom instance topology with the plurality of symptom topology patterns to obtain a correlation coefficient parameter for each first chronic disease symptom instance topology; Determining the contrastive learning feature encoding for each first chronic disease symptom instance according to the correlation coefficient parameter corresponding to each first chronic disease symptom instance topology.

4. The method according to claim 3, characterized in that, The associating each first chronic disease symptom instance topology with the plurality of symptom topology patterns to obtain a correlation coefficient parameter for each first chronic disease symptom instance topology includes: Associating a target first chronic disease symptom instance topology with each symptom topology pattern to obtain a correlation coefficient corresponding to each symptom topology pattern; Taking the symptom topology pattern with a correlation coefficient exceeding the correlation coefficient threshold as the target symptom topology pattern corresponding to the target first chronic disease symptom instance topology; Take the target symptom topology pattern corresponding to the target first chronic disease symptom instance topology as the correlation coefficient parameter of the target first chronic disease symptom instance topology.

5. The method according to claim 4, wherein The determining of the contrast learning feature encoding of each first chronic disease symptom instance according to the correlation coefficient parameter corresponding to each first chronic disease symptom instance topology includes: Obtain the core feature encoding of the target symptom topology pattern corresponding to the target first chronic disease symptom instance topology; Determine the control feature encoding that is not associated with the target chronic disease symptom instance topology among the multiple symptom topology patterns according to the core feature encoding; Perform feature coupling processing on the core feature encoding and the control feature encoding to obtain the contrast learning feature encoding of the target first chronic disease symptom instance topology.

6. The method according to claim 2, characterized in that, The preset basic model includes multiple physical sign interaction networks and at least one physical sign fusion network. The performing of the recognition processing on each first chronic disease symptom instance topology based on the preset basic model to obtain the first feature vector instance corresponding to each first chronic disease symptom instance topology includes: According to the disease description information of the target chronic disease symptom instance, obtain the physical sign feature vector of each symptom entity in the target first chronic disease symptom instance topology, the associated physical sign entity of each symptom entity, and the association intensity vector of the pathological association relationship between each symptom entity and the associated physical sign entity; Perform a feature transformation operation on the physical sign feature vector of each symptom entity to obtain the basic physical sign vector of each symptom entity; Perform feature fusion on the basic physical sign vector of the associated physical sign entity and the association intensity vector to obtain a fusion vector; Obtain the first weight tensor of the first feature aggregation module for interactive aggregation processing, and perform a linear transformation operation on the first weight tensor and the fusion vector to obtain a first transformation coefficient tensor; Process the first transformation coefficient tensor based on a preset first non-linear mapping function to obtain the first-order interactive aggregation vector of the symptom entity; Perform physical sign aggregation processing on the input physical sign vector and the first-order interactive aggregation vector of each symptom entity to obtain the first-order physical sign vector of each symptom entity; Perform interactive aggregation processing on the previous-order physical sign vector of the associated physical sign entity and the association intensity vector based on the target physical sign interaction network to obtain the target-order interactive aggregation vector of each symptom entity; Perform physical sign aggregation processing on the previous-order physical sign vector and the target-order interactive aggregation vector of each symptom entity to obtain the target-order physical sign vector of each symptom entity; Take the target-order physical sign vector of each symptom entity as the feature vector of each symptom entity; Process the feature vector of each symptom entity based on the physical sign fusion network to obtain the first feature vector instance corresponding to the target first chronic disease symptom instance topology.

7. The method according to claim 6, characterized in that The performing of the physical sign aggregation processing on the input physical sign vector and the first-order interactive aggregation vector of each symptom entity to obtain the first-order physical sign vector of each symptom entity includes: Obtain the second weight tensor of the second feature aggregation module for performing physical sign aggregation processing; Perform a linear transformation operation on the second weight tensor and the first-order interaction aggregation vector to obtain a second transformation coefficient tensor; Based on a preset second non-linear mapping function, process the input physical sign vectors of each symptom entity and the second transformation coefficient tensor to obtain the first-order physical sign vectors of each symptom entity.

8. The method according to claim 6, characterized in that The obtaining of the physical sign feature vectors of each symptom entity, the associated physical sign entities of each symptom entity, and the association strength vectors of the pathological association relationships between each symptom entity and the associated physical sign entities in the target first chronic disease symptom instance topology according to the disease description information of the target chronic disease symptom instance includes: According to the disease description information of the target first chronic disease symptom instance, obtain the clinical index data of each sub-symptom instance in the target first chronic disease symptom instance; Perform feature coupling processing on the clinical index data of each sub-symptom instance to obtain the physical sign feature vectors of each symptom entity in the target first chronic disease symptom instance topology; Use the control sub-symptoms that have a clinical association path with each sub-symptom instance as the associated sub-symptoms of each sub-symptom instance; Obtain the clinical index data of the clinical association path between each sub-symptom instance and the associated sub-symptom; Perform feature coupling processing on the clinical index data of the clinical association path to obtain the association strength vectors of the pathological association relationships between each symptom entity and the associated physical sign entities.

9. The method according to claim 1, wherein The model optimization of the initially trained initial analysis model based on the second chronic disease instance to obtain the trained chronic disease recognition model corresponding to the disease analysis index includes: Obtain the second chronic disease symptom instance topologies of each second chronic disease symptom instance in the second chronic disease instance; Based on the initially trained initial analysis model, perform recognition processing on each second chronic disease symptom instance topology to obtain the second feature vector instances corresponding to each second chronic disease symptom instance topology; Based on the trained feature recognition component, perform recognition on each second feature vector instance to obtain the second symptom inference results corresponding to each second chronic disease symptom instance; Perform integrated optimization training on the initially trained initial analysis model according to the chronic disease target values of each second chronic disease symptom instance and the second symptom inference results to obtain the trained chronic disease recognition model corresponding to the disease analysis index.

10. A server system, characterized in that, It includes a server, and the server is used to execute the method according to any one of claims 1-9.

Images