Big data analysis-based obesity risk early warning method and system

Abstract

The invention provides an obesity risk early warning method and system based on big data analysis, and the method comprises the steps: carrying out the mining of detail implicit representation and overall implicit representation of obesity feature data in obesity statistical big data of a to-be-analyzed user and reference obesity big data in a reference obesity big data set in a big data processing process; and generating a detail-guided topological structure through the mined implicit representation, so that the mined detail implicit representation generates a positive synergistic effect on the overall implicit representation, and a more accurate overall implicit representation topological structure is obtained. And then, indexing is performed through the overall implicit representation after iteration, and a more accurate obesity big data indexing result of obesity feature data to be identified and referenced is obtained, so that an accurate obesity early warning level is obtained based on the result.

Description

Technical Field

[0001] The present application relates to the field of data processing, and specifically, to an obesity risk warning method and system based on big data analysis. Background Art

[0002] With the change of modern lifestyle, obesity has become a global public health challenge, which has a significant impact on individual health and social economy. Obesity not only increases the risk of multiple chronic diseases such as diabetes and cardiovascular disease, but also significantly reduces the quality of life of patients. Therefore, early warning and intervention of obesity risk are particularly important. Traditional obesity risk assessment methods mainly rely on subjective assessment methods such as questionnaires and physical examinations. Although these methods can reflect the obesity status of individuals to a certain extent, they often have problems such as strong subjectivity, low efficiency, and limited coverage. With the rapid development of big data and artificial intelligence technology, obesity risk assessment using big data analysis and machine learning models has become an emerging and highly promising research direction. The current obesity risk assessment methods based on big data still have some shortcomings. Most methods only focus on the statistical analysis of overall obesity indicators, ignoring the complex relationship and detailed information between obesity characteristics, resulting in inaccurate assessment results. On the other hand, when dealing with large-scale obesity data, existing methods often lack effective data representation and feature extraction technology, making it difficult to fully explore the implicit information and association patterns in the data, thus affecting the accuracy and efficiency of the assessment. Summary of the invention

[0003] The purpose of the present invention is to provide an obesity risk warning method and system based on big data analysis. The embodiment of the present application is implemented as follows:

[0004] In a first aspect, the present application provides an obesity risk warning method based on big data analysis, which is applied to a computer system, and the method comprises: obtaining user obesity statistical big data to be analyzed and a reference obesity big data set, wherein the user obesity statistical big data to be analyzed includes target obesity feature data to be matched; performing obesity information implicit representation mining on the user obesity statistical big data to be analyzed and the reference obesity big data in the reference obesity big data set through a machine learning network to obtain obesity information implicit representations of multiple obesity feature data, wherein the multiple obesity feature data include the target obesity feature data, and the obesity information implicit representations include overall implicit representations and detailed implicit representations; using the detailed implicit representations of the multiple obesity feature data as topology data; The method comprises the following steps: generating a first topological structure corresponding to the detailed implicit representation of the vertices, and generating a second topological structure corresponding to the overall implicit representation of the topological vertices using the overall implicit representation of the plurality of obesity feature data as the topological vertices, wherein the first topological structure and the second topological structure both have a link state between the topological vertices, and the link state includes being connected or not connected, wherein the topological vertices greater than the commonality threshold are connected by connecting lines, and the topological vertices less than the commonality threshold are not connected; iterating the second topological structure according to the link state between the topological vertices in the first topological structure to obtain a target topological structure; determining the target obesity big data in the reference obesity big data set according to the commonality measurement result between the topological vertices in the target topological structure;

[0005] The warning level corresponding to the target obesity big data is used as the target warning level, and obesity warning is performed based on the target warning level.

[0006] In a second aspect, the present application provides a computer system comprising: one or more processors; a memory; one or more computer programs; wherein the one or more computer programs are stored in the memory and configured to be executed by the one or more processors, and when the one or more computer programs are executed by the processors, the method as described above is implemented.

[0007] Beneficial effects of the present application: The obesity risk warning method and system based on big data analysis provided by the present application mines the detailed implicit representation and overall implicit representation of the obesity feature data in the user obesity statistical big data to be analyzed and the reference obesity big data in the reference obesity big data set, and generates a detail-guided topological structure through the mined implicit representation, so that the mined detailed implicit representation has a positive synergistic effect on the overall implicit representation, and obtains a more accurate overall implicit representation topological structure. Then, through the iterative overall implicit representation, indexing is performed to obtain an obesity big data index result that is more accurate than the obesity feature data to be identified and referenced, so as to obtain an accurate obesity warning level based on this. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 This is a flow chart of an obesity risk warning method based on big data analysis provided in an embodiment of the present application.

[0009] Figure 2 It is a schematic diagram of the composition of a computer system provided in an embodiment of the present application.

[0010] Reference numerals: 100 - computer system; 101 - processor; 102 - bus; 103 - memory; 104 - transceiver. DETAILED DESCRIPTION

[0011] The execution subject of the obesity risk warning method based on big data analysis in the embodiment of the present application is a computer system, including but not limited to a server, a personal computer, a laptop, a tablet computer, a smart phone, etc. The network where the computer system is located includes but is not limited to the Internet, a wide area network, a metropolitan area network, a local area network, a VPN network, etc. The embodiment of the present application provides an obesity risk warning method based on big data analysis, which is applied to a computer system, such as Figure 1 As shown, the method includes:

[0012] Step S10: Obtaining the user obesity statistical big data to be analyzed and the reference obesity big data set, wherein the user obesity statistical big data to be analyzed includes the target obesity feature data to be matched.

[0013] In step S10, the big data of obesity statistics of the user to be analyzed is obtained. Specifically, for the user to be analyzed, the big data of obesity statistics can be obtained through various channels (such as health apps, physical examination reports, wearable devices, etc.). These data usually include but are not limited to the user's height, weight, waist circumference, body fat percentage, BMI (body mass index), daily dietary intake, exercise volume and other indicators related to health and obesity. These data constitute the user's obesity feature data set, each feature is a data unit, and they together describe the user's current health status and obesity risk.

[0014] For example, suppose that for a user, the computer system obtains the following data from his health app: height 175cm (data unit 1), weight 85kg (data unit 2), waist circumference 90cm (data unit 3), body fat percentage 25% (data unit 4). These data units together constitute the user's obesity feature data set, among which weight and body fat percentage are important indicators for assessing obesity risk.

[0015] At the same time, the computer system will also build a reference obesity big data set. This set contains a large number of statistical data of users with known obesity conditions. The obesity characteristic data of each user is also composed of multiple data units, and each user has a corresponding obesity assessment result or health status label. These data can come from medical institutions, scientific research institutions, health management institutions, etc. After cleaning and sorting, a structured data set is formed for subsequent data matching and analysis.

[0016] For example, the reference obesity big data set may contain tens of thousands of records, each of which corresponds to a user's obesity feature data set, such as user A's feature data set including: height 170cm, weight 70kg, waist circumference 80cm, body fat percentage 20%; user B's feature data set including: height 165cm, weight 65kg, waist circumference 75cm, body fat percentage 18%, etc. At the same time, each record may also contain the user's obesity assessment results, such as "normal", "overweight", "obese" and other labels.

[0017] It should be noted that when it comes to collecting user privacy data, it must be done in compliance with the provisions of laws and regulations and with the user's permission.

[0018] After collecting the big data of obesity statistics of the users to be analyzed, the computer system identifies the target obesity feature data to be matched. These target data are, for example, some of the most critical indicators for assessing obesity risk, such as weight, body fat percentage, waist circumference, etc. In the subsequent data processing and analysis process, the computer system will mainly operate around these target obesity feature data.

[0019] Step S20: Obesity information implicit representation mining is performed on the user obesity statistical big data to be analyzed and the reference obesity big data in the reference obesity big data set through a machine learning network to obtain obesity information implicit representations of multiple obesity feature data, wherein the multiple obesity feature data include the target obesity feature data, and the obesity information implicit representation includes an overall implicit representation and a detailed implicit representation.

[0020] In step S20, the computer system uses a machine learning network to conduct an in-depth analysis of the collected user obesity statistical big data to be analyzed and the reference obesity big data set, aiming to mine the implicit representation of obesity information and transform the original data into higher-level abstract features, so as to provide strong support for subsequent data processing and risk assessment.

[0021] First, the computer system selects or designs a suitable machine learning model to mine implicit representations of obesity information. This model may be a neural network based on deep learning, such as a convolutional neural network (CNN), a recurrent neural network (RNN) or its variants (such as a long short-term memory network LSTM), etc. These models perform well in processing complex data relationships. After the model is selected, the computer system uses a portion of the data (usually labeled data) in the reference obesity big data set to train the model. During the training process, the model learns how to extract implicit representations from the original obesity feature data that are useful for assessing obesity risk. These implicit representations may exist in the form of vectors, each of which captures certain specific patterns or features in the data that are critical to identifying obesity risk.

[0022] Once the model is trained, it is applied to the obesity statistical big data of the user to be analyzed and the remaining unlabeled data in the reference obesity big data set. For each set of obesity feature data (including the target obesity feature data), the model will output its corresponding implicit representation of obesity information. These implicit representations can be divided into two categories: overall implicit representation and detailed implicit representation. The overall implicit representation captures the characteristics of the user's overall obesity status. For example, it can be a vector in which each element represents the comprehensive evaluation result of the user on different obesity-related indicators. For example, a possible overall implicit representation vector can be [O.3, 0.7, 0.5], which represents the user's comprehensive obesity risk score in terms of body fat percentage, BMI, and waist circumference. Unlike the overall implicit representation, the detailed implicit representation focuses more on specific obesity features. It may contain multiple vectors, each of which corresponds to an implicit description of an obesity feature data. For example, for the feature of weight, its detailed implicit representation may be a vector containing information such as the trend of weight change and the ratio of weight to other indicators (such as height).

[0023] Step S30: Using the detailed implicit representations of the multiple obesity feature data as topological vertices to generate a first topological structure corresponding to the detailed implicit representations, and using the overall implicit representations of the multiple obesity feature data as topological vertices to generate a second topological structure corresponding to the overall implicit representation, wherein the first topological structure and the second topological structure both have a link state between topological vertices, and the link state includes connected or disconnected, wherein topological vertices greater than a commonality threshold are connected by connecting lines, and topological vertices lower than the commonality threshold are not connected.

[0024] In step S30, the computer system further constructs two topological structures using the detailed implicit representation and the overall implicit representation of the obesity feature data obtained in step S20: a first topological structure based on the detailed implicit representation and a second topological structure based on the overall implicit representation. These two topological structures not only reflect the intrinsic connection between the obesity feature data, but also reveal the similarities and differences between the features through the link status (connected or unconnected) between vertices.

[0025] Implicit representation of details as vertices of the first topological structure: Each implicit representation of details (such as the trend of weight change, the implicit feature vector of a specific indicator) is regarded as an independent vertex in the first topological structure. For example, if the weight details of user A are implicitly represented as [85kg, ↑] (indicating that the current weight is 85kg and is on an upward trend), and the weight details of user B are implicitly represented as [70kg, →] (indicating that the current weight is 70kg and remains stable), then these two representations will appear as two vertices in the first topological structure respectively.

[0026] The overall implicit representation is used as a vertex in the second topological structure: Similarly, each overall implicit representation (such as a vector that integrates multiple obesity indicators) is used as a vertex in the second topological structure. For example, the overall implicit representation of user A may be [0.6, 0.8, 0.7] (corresponding to the comprehensive scores of body fat percentage, BMI, and waist circumference, respectively), while the overall implicit representation of user B may be [0.4, 0.6, 0.5].

[0027] In order to determine the link status between topological vertices, the computer system calculates the commonality measurement results between the vertices, which usually involves calculating the similarity between the vertex representations. Commonly used similarity calculation methods include cosine similarity, Euclidean distance, etc. Before executing the link, the computer system presets a commonality threshold (such as 0.7, indicating the similarity threshold). Only when the similarity between two vertices is greater than this threshold will they be connected by a connecting line, indicating that the two vertices (i.e., the corresponding obesity feature data) are similar or related to some extent.

[0028] For any two detail implicit representation vertices in the first topological structure, the computer system calculates the similarity between them. If the similarity is greater than the commonality threshold, a connecting line is drawn between the two vertices, indicating that there is a significant correlation between the two detail features. For example, if the weight increase trend of user A is statistically significantly correlated with the waist circumference increase trend of user B, their corresponding vertices will be connected. Similar to the first topological structure, the computer system also calculates the similarity between any two overall implicit representation vertices in the second topological structure, and determines whether to connect the two vertices based on the comparison result of the similarity and the commonality threshold.

[0029] For example, suppose that when constructing the first topological structure, the similarity between the vertex of user A's weight increase trend and the vertex of user C's body fat percentage increase trend reaches 0.8 (higher than the commonality threshold of 0.7), so the two vertices are connected by a connecting line. This implies that weight gain may be directly related to the increase in body fat percentage, which is of great significance for assessing the user's obesity risk. In the second topological structure, if the overall implicit representation similarity of user A and user B is high (for example, their comprehensive score vectors are close in multiple dimensions), their vertices will also be connected. This means that overall, the obesity conditions of the two users are similar and they may face similar health risks. By constructing such a topological structure, the computer system can display the complex relationship between obesity feature data in an intuitive and efficient way, providing strong support for subsequent risk assessment and early warning.

[0030] Step S40: iterating the second topological structure according to the link status between the topological vertices in the first topological structure to obtain a target topological structure.

[0031] In step S40, the computer system guides the iterative update of the second topological structure by utilizing the link status between topological vertices in the first topological structure, aiming to optimize the structure of the overall implicit representation through the guidance of the detailed implicit representation, thereby obtaining a more accurate and meaningful target topological structure.

[0032] Specifically, the computer system reviews the first topological structure, i.e., the topological structure based on the implicit representation of details. In this structure, the link status between vertices reflects the correlation and similarity between different detail features. For example, if the trend of weight gain is statistically significantly correlated with the trend of waist circumference growth, they will be connected by a connecting line in the first topological structure. Next, the computer system adjusts the second topological structure, i.e., the topological structure based on the overall implicit representation, according to the link status in the first topological structure. The iterative process, for example, involves the following steps:

[0033] Identify key links: The computer system identifies strongly connected vertex pairs in the first topological structure (i.e., vertex pairs whose link status is connected and whose similarity is higher than a certain threshold). These vertex pairs represent highly related detail features.

[0034] Update the second topology structure: Based on these key links, the computer system adjusts the position or connection mode of the corresponding global implicit representation vertices in the second topology structure. Specifically, if there is a strong connection between two global implicit representation vertices and the corresponding detail implicit representation vertices, then the two global implicit representation vertices should also be more closely connected in the second topology structure, or their relative positions should be adjusted to reflect this association.

[0035] Iterative optimization: This process can be repeated many times, and the computer system gradually optimizes the second topology through multiple iterations. Each iteration will fine-tune the second topology according to the latest link status in the first topology until a stable target topology that accurately reflects the internal relationship of the data is reached.

[0036] For example, suppose that in the first topology, there is a strong connection between the weight gain trend vertex of user A and the body fat percentage increase trend vertex of user B, indicating that the changes of the two are highly correlated. In the second topology, the overall implicit representation vertices of user A and user B may not be tightly connected initially, or the connection strength is not strong enough to reflect their high correlation in detailed features.

[0037] During the iteration of step S40, the computer system identifies this key link and adjusts the second topological structure accordingly. Specifically, it may add a stronger connection line between the overall implicit representation vertices of user A and user B, or move the two vertices to a closer position to intuitively show their similarities and associations in their overall obesity status.

[0038] After multiple iterations, the computer system will obtain an optimized target topology structure, which not only reflects the overall obesity status of each user, but also accurately reveals the similarities and differences in obesity characteristics between different users through the connection status between vertices. This is of great significance for subsequent obesity risk assessment and early warning.

[0039] Step S50: determining target obesity big data from the reference obesity big data in the reference obesity big data set according to the commonality measurement result between the topological vertices in the target topological structure.

[0040] In step S50, the computer system uses the commonality measurement results between topological vertices in the target topological structure constructed in step S40 to guide the search and determination of reference obesity big data that best matches the obesity characteristics of the target user in the reference obesity big data set, namely, the target obesity big data.

[0041] First, the computer system traverses all topological vertices in the target topological structure, which represent the overall implicit representation after optimization and iteration. For each topological vertex, the computer system calculates the commonality measurement results between it and other vertices, which reflect the similarity and correlation between the overall obesity characteristics represented by the vertices. Then, the computer system uses these commonality measurement results to search in the reference obesity big data set. The reference obesity big data set is a collection of a large amount of user data with known obesity conditions. Each user's data contains a detailed description of the obesity characteristics and the corresponding health status label (such as "normal", "overweight", "obese", etc.).

[0042] During the search process, attention is paid to reference data with high commonality measurement results with specific vertices in the target topological structure (especially vertices that are highly correlated with the obesity characteristics of the target user). These high commonality measurement results mean that the obesity characteristics in the reference data have a high overall similarity with the obesity characteristics of the target user. In order to determine the final target obesity big data, the computer system can adopt a variety of strategies, such as weighted summation, voting mechanism, or probability-based sorting algorithm. For example, it can assign a similarity score to each reference data, which is calculated based on the commonality measurement results between the reference data and each vertex in the target topological structure. The computer system then sorts the reference data according to the similarity score and selects the data with the highest score as the target obesity big data.

[0043] Finally, the computer system determines the target obesity big data based on the search and matching results, and this data will be used for subsequent obesity risk assessment and early warning. Through this process, the computer system can more accurately predict the obesity risk of the target user and provide more personalized health advice for them.

[0044] Step S60: taking the warning level corresponding to the target obesity big data as the target warning level, and performing obesity warning based on the target warning level.

[0045] In step S60, the computer system extracts warning level information from the target obesity big data determined in step S50. In the reference obesity big data set, each reference data is usually accompanied by one or more tags or fields to indicate the user's obesity risk level, such as "low risk", "medium risk" or "high risk". These level labels are based on a comprehensive evaluation of a large amount of historical data and professional knowledge and have high reliability. Then, the computer system uses the extracted warning level as the target warning level. This level directly reflects the current obesity risk status of the target user and is the basis for subsequent warning actions.

[0046] Finally, the computer system generates obesity warning information based on the target warning level and sends it to the user through appropriate channels (such as mobile phone APP notifications, emails, text messages, etc.). The warning information should clearly inform the user of their obesity risk level and provide corresponding health advice or guidance measures to help users take positive actions to reduce obesity risks.

[0047] For example, suppose that in step S50, the target obesity big data determined by the computer system highly matches the reference data of user D, and the reference data of user D clearly marks his obesity risk level as "high risk". In step S60, the computer system uses "high risk" as the target warning level and generates obesity warning information accordingly.

[0048] As an implementation mode, the step S20, performing obesity information implicit representation mining on the user obesity statistical big data to be analyzed and the reference obesity big data in the reference obesity big data set through a machine learning network to obtain obesity information implicit representations of multiple obesity feature data, wherein the multiple obesity feature data include the target obesity feature data, and the obesity information implicit representation includes an overall implicit representation and a detailed implicit representation. It may also include a training process of the machine learning network, specifically including:

[0049] Step S201: obtaining a debugging learning sample, wherein the debugging learning sample includes a plurality of obesity big data samples and a warning priori mark of each obesity big data sample, wherein the obesity big data samples include a positive obesity big data sample and a negative obesity big data sample, and the warning priori mark includes a detail priori mark, an overall priori mark, and a warning level priori mark;

[0050] Step S202: obtaining a machine learning network to be debugged, wherein the machine learning network includes a data implicit representation mining component, an obesity feature data detail recognition component, an obesity feature data overall recognition component, and a warning level analysis component;

[0051] Step S203: Debugging the machine learning network through the debugging learning examples.

[0052] In the embodiment of the present application, the training of the machine learning network is an important process to ensure that the model can accurately identify and predict the user's obesity risk. This process not only involves the preparation of data, but also covers the selection, debugging and optimization of the model. In step S201, the computer system collects and organizes a set of debugging learning samples from existing data resources. These samples are a collection of obesity big data, and each sample contains multiple obesity feature data and corresponding warning prior markers. The warning prior marker is the risk level pre-judged by the expert based on the user's obesity feature data, which provides a supervision signal for model training.

[0053] For example, suppose a medical institution cooperates with a scientific research institution to carry out an obesity risk warning project. The medical institution provides a large number of patient health records, including height, weight, waist circumference, body fat percentage, eating habits, exercise volume and other obesity-related characteristic data. At the same time, each record is also accompanied by an obesity risk level (such as "low risk", "medium risk", "high risk") assessed by a professional doctor as a warning priori mark.

[0054] The detailed priori mark is, for example, a specific evaluation result for each obesity feature, such as a "high" mark for body fat percentage. The overall priori mark is, for example, a comprehensive evaluation result based on all feature data, such as "medium risk".

[0055] The warning level priori mark is, for example, a directly given obesity risk level, such as "2" (assuming a level division of 1-3, 1 being low risk and 3 being high risk).

[0056] In step S202, after having the debugging learning samples, the computer system selects or designs a suitable machine learning network for training. This network needs to include multiple components to process the implicit representation of data, the detailed recognition of obesity feature data, the overall recognition and the final warning level analysis.

[0057] The data implicit representation mining component is responsible for extracting information from the raw data that protects user privacy and effectively represents obesity characteristics. This can be achieved through differential privacy, federated learning and other technologies. The obesity feature data detail recognition component focuses on identifying specific obesity characteristics from the data, such as waist circumference, body fat percentage, etc., and understanding how these characteristics individually affect obesity risk. The obesity feature data overall recognition component integrates multiple detail features to form an understanding of the user's overall obesity status. This is a network structure that can process multi-dimensional inputs and output a comprehensive score. The warning level analysis component outputs the user's obesity risk warning level based on the details and overall recognition results. This may require a classifier or regression model to achieve this. Based on the convolutional neural network (CNN), combined with the long short-term memory network (LSTM) and the fully connected layer (FC), a composite network model is designed. Among them, CNN is used to extract spatial features from image data (such as body fat distribution maps), LSTM is used to process time series data (such as long-term eating habits records), and the FC layer is responsible for fusing the extracted features and making the final warning level judgment.

[0058] In step S203, the computer system uses the debugging learning examples obtained in step S201 to train and debug the machine learning network. During the training process, the model will continuously try to learn from the sample data how to accurately identify obesity characteristics, understand the overall obesity status, and predict the warning level. Debugging involves adjusting model parameters, optimizing algorithm selection, adding regularization terms, etc. to improve the generalization ability and prediction accuracy of the model.

[0059] During the training process, the debugging learning samples are normalized and standardized to ensure that the scales of different features are consistent, which is convenient for model learning. Then, the weights and bias parameters of the model are randomly initialized. The debugging samples are input into the model and processed by each component in turn to obtain the prediction results. Then, the difference between the prediction results and the warning prior marks is compared to calculate the loss function value. Commonly used loss functions include cross entropy loss (for classification problems) and mean square error loss.

[0060] For example, the cross entropy loss function is:

[0061]

[0062] Where N is the number of samples, M is the number of categories, and yi c is the true label (0 or 1), indicating whether the i-th sample belongs to the c-th category, is the probability that the model predicts that it belongs to class c.

[0063] After that, according to the loss function value, use optimization algorithms such as gradient descent to back propagate the error and update the model parameters. Repeat the above steps until the loss function value converges or reaches the preset number of iterations.

[0064] Finally, an independent validation set can be used to evaluate model performance, such as accuracy, recall, F1 score, etc. Based on the evaluation results, the model structure, hyperparameters, or regularization terms can be adjusted to improve model performance.

[0065] In addition, to prevent overfitting, the computer system can also add regularization terms (such as L1 regularization or L2 regularization) to the loss function, and use early stopping during training to avoid continuing training when performance on the validation set decreases. Through this series of complex training and debugging processes, the computer system can build an efficient and accurate obesity risk warning model. This model can not only understand the user's overall obesity status, but also capture key information in detailed features, providing users with personalized health risk prompts and suggestions.

[0066] As an implementation manner, the step S203, debugging the machine learning network through the debugging learning example, includes:

[0067] Step S2031: Obesity information implicit representation mining is performed on the obesity big data samples through the data implicit representation mining component of the machine learning network to obtain a feature vector set of the obesity big data samples.

[0068] Step S2031 utilizes the data implicit representation mining component in the machine learning network to mine useful obesity features from the obesity big data sample, while ensuring that these features do not leak the user's personal privacy. To achieve this goal, the computer system uses a combination of multiple privacy protection technologies and machine learning algorithms.

[0069] The computer system can receive obesity big data samples from different sources. These data samples may include direct indicators such as the user's height, weight, waist circumference, body fat percentage, etc., and may also include factors that indirectly affect obesity such as the user's eating habits and exercise volume. Since these data often contain the user's personal privacy information, there are risks in using them directly for model training, so they need to be processed through the data implicit representation mining component.

[0070] The data implicit representation mining component contains, for example, autoencoders and differential privacy technologies. Autoencoders can perform feature extraction, and differential privacy interferes with the real data value by adding an appropriate amount of random noise to the original data, while ensuring that the data after adding noise still maintains a high accuracy in statistical analysis. This mechanism allows researchers or models to extract useful information from the data while protecting user privacy.

[0071] In actual applications, the data implicit representation mining component will process each obesity big data sample one by one. Taking weight data as an example, assuming that a user's actual weight is 75kg, the data implicit representation mining component will generate a random noise value (for example, noise that conforms to the Laplace distribution) for the data according to the algorithm requirements of differential privacy. Then, the actual weight is added to the noise value to obtain a perturbed weight value, such as 75.03kg (the noise value here is only an example, and the actual value will depend on the specific parameters of the differential privacy algorithm).

[0072] In addition to weight data, other obesity feature data (such as waist circumference, body fat percentage, etc.) will also undergo a similar processing process. After each feature data is processed with privacy protection, it will be converted into a feature vector. These feature vectors together constitute a feature vector set, that is, a feature graph structure, in which each feature vector represents a privacy-protected obesity feature representation in the original data sample. The noise value is randomly generated and may be different each time it is processed. Therefore, in practical applications, the specific values ​​in the feature vector set will change, but overall it will maintain statistical properties similar to the original data.

[0073] Through the processing of step S2031, the computer system successfully mines useful obesity feature representations from the obesity big data sample containing sensitive information, while ensuring the security of user privacy. These privacy-protected feature vector sets provide a solid foundation for subsequent obesity feature data detail recognition, overall recognition, and warning level analysis.

[0074] Step S2032: performing detail recognition on the feature vector set through the obesity feature data detail recognition component to obtain detail recognition results of multiple sample obesity feature data.

[0075] In step S2032, the computer system uses the obesity feature data detail recognition component to conduct an in-depth analysis of the feature vector set obtained in step S2031 to identify the specific details of each obesity feature data. In step S2031, the computer system has processed the obesity big data sample through the data implicit representation mining component to obtain a feature vector set containing multiple feature vectors. These feature vectors have been processed with privacy protection, which not only protects user privacy but also retains the useful information of the data. Next, in step S2032, the computer system will use the obesity feature data detail recognition component to perform detail recognition on these feature vectors. The obesity feature data detail recognition component may be a complex system containing multiple submodules, each of which is specifically responsible for identifying the details of a certain type of obesity feature data. For example, the component may include a weight recognition submodule, a waist circumference recognition submodule, a body fat percentage recognition submodule, etc. Each submodule will process and analyze the corresponding feature vector according to a preset rule or algorithm.

[0076] Taking the weight recognition submodule as an example, a series of corresponding relationships between weight ranges and obesity levels can be set. For example, for adults, body mass index (BMI) is a commonly used evaluation indicator. The weight recognition submodule may calculate the weight of the person according to the BMI calculation formula (weight (kg) / height 2 (m2)) and the preset BMI threshold to determine whether the weight is overweight. Specifically, if a user's BMI value exceeds 25kg / m 2 (This is a commonly used overweight threshold), the weight recognition submodule will recognize the user's weight as "overweight". Similarly, the waist recognition submodule and body fat recognition submodule will also process the corresponding feature vectors according to their respective evaluation criteria and algorithms to obtain detailed recognition results of waist circumference and body fat percentage. For example, the waist recognition submodule may determine whether the waist circumference is too thick based on the ratio of waist circumference to height; the body fat percentage recognition submodule may compare the body fat percentage with the preset healthy range to conclude whether the body fat percentage is too high.

[0077] Through the processing of step S2032, the computer system successfully realizes the accurate recognition of the details of the obesity feature data. These detail recognition results lay a solid foundation for the overall obesity status assessment and obesity risk warning in the subsequent steps. At the same time, since these detail recognition results are obtained based on the feature vectors that have been processed with privacy protection, the security of user privacy is also ensured.

[0078] Step S2033: the obesity feature data overall recognition component is used to perform overall recognition on the feature vector set to obtain overall recognition results of multiple sample obesity feature data.

[0079] In step S2033, the computer system uses the obesity feature data overall recognition component to perform overall recognition on the feature vector set obtained in step S2031, aiming to obtain the overall status of the obesity feature data of each sample. In step S2032, the computer system has performed detail recognition on each feature vector in the feature vector set through the obesity feature data detail recognition component, and obtained detail recognition results of multiple obesity feature data. These results provide specific information about the user's performance in different obesity characteristics, but have not yet formed a comprehensive overall evaluation. The goal of step S2033 is to fill this gap and conduct a more in-depth analysis of the feature vector set through the overall recognition component to obtain the overall obesity status of each sample.

[0080] The obesity feature data holistic recognition component is a machine learning model that can process multi-dimensional input (i.e., a set of feature vectors) and output a comprehensive evaluation result. This model may be built based on deep learning, ensemble learning, or other advanced machine learning techniques, and can automatically learn the complex relationships between features and capture hidden patterns in the data.

[0081] In practical applications, the overall recognition component can first weight each feature vector to reflect the importance of different features in the overall assessment. For example, some features (such as BMI, body fat percentage) may be more important for predicting obesity risk, so they should be given higher weights in the model. After the weighting process is completed, the overall recognition component will use these weighted feature vectors as input and obtain the overall recognition result through a series of calculations and transformations.

[0082] The overall recognition result is, for example, a numerical score, a classification label, or a structured output containing multiple dimensions. For example, it can be a numerical score between 0 and 1, indicating the user's obesity risk level (the higher the value, the greater the risk); it can also be a classification label, such as "low risk", "medium risk" or "high risk"; it can also be a structured output containing multiple obesity-related indicators, such as weight status, body fat distribution, metabolic health, etc.

[0083] Through the processing of step S2033, not only can the obesity characteristic data of each sample be evaluated as a whole, but also specific risk levels and recommended measures can be provided. These overall recognition results provide an important reference for subsequent warning level analysis and user intervention. At the same time, since the overall recognition component is built based on a complex machine learning model, it can more accurately capture hidden patterns and associations in the data, thereby improving the accuracy and reliability of obesity risk warnings.

[0084] Step S2034: performing obesity warning level analysis on the sample obesity feature data in the feature vector set through the warning level analysis component to obtain an obesity warning level analysis result.

[0085] Step S2034 marks the transformation from data analysis to the actual application of risk warning. In this step, the computer system uses the warning level analysis component to perform in-depth analysis on the sample obesity feature data in the feature vector set to evaluate and determine the user's obesity warning level.

[0086] Specifically, after completing the construction of the feature vector set, detail identification and overall identification, the core task of step S2034 is to convert these analysis results into obesity warning levels with practical significance. As the executor of this step, the warning level analysis component integrates a variety of algorithms and models, which can comprehensively consider the user's overall obesity status, detailed feature performance, and other possible risk factors, so as to give a comprehensive and accurate warning level assessment. The warning level analysis component may use a variety of machine learning techniques, including but not limited to logistic regression, decision trees, random forests, gradient boosting trees (GBDT) or deep neural networks (DNN). These technologies have their own advantages and can be flexibly selected according to the different characteristics of the data and analysis requirements. For example, if there are a large number of nonlinear relationships in the data set, a deep neural network may be a better choice; if the data dimension is high and decisions need to be made quickly, a random forest or a gradient boosting tree may be more appropriate.

[0087] In actual applications, the warning level analysis component will first receive the comprehensive evaluation results (such as numerical scores, classification labels or structured outputs) from the overall recognition component and other possible risk factor information (such as user age, gender, family medical history, etc.). Then, it will weight this information and perform classification or regression prediction based on the preset warning level classification criteria (such as low risk, medium risk, high risk, etc.).

[0088] During the prediction process, the warning level analysis component may use cross-validation and other technical means to ensure the generalization ability of the model and avoid overfitting. At the same time, it will also record the results of each prediction and the corresponding confidence level to facilitate subsequent evaluation and optimization of model performance.

[0089] Through the processing of step S2034, the computer system converts complex obesity characteristic data into easy-to-understand warning level information, providing strong support for the prevention, control and management of obesity risks.

[0090] Step S2035: Optimizing the network configuration parameters of the machine learning network according to the detail recognition result, the overall recognition result, and the error between the obesity warning level analysis result and the warning priori mark.

[0091] Step S2035 involves optimizing the network configuration parameters according to the error between the model prediction result and the actual result (ie, the warning priori mark), thereby improving the prediction accuracy and generalization ability of the model.

[0092] Specifically, after completing detail recognition, overall recognition and obesity warning level analysis, the computer system obtains prediction results of multiple samples, including detail recognition results, overall recognition results and obesity warning level analysis results. At the same time, these samples are also accompanied by warning prior marks, that is, the actual obesity risk level determined by experts or historical data. The purpose of step S2035 is to compare the differences between these prediction results and warning prior marks, that is, errors, and adjust the configuration parameters of the machine learning network accordingly to reduce errors and improve model performance.

[0093] To achieve this goal, the computer system first calculates the error between the predicted result and the warning prior mark. The way the error is calculated depends on the specific task type and the choice of loss function. For classification tasks (such as obesity warning level classification), commonly used loss functions include cross-entropy loss.

[0094] In step S2035, since the detailed recognition results, the overall recognition results and the obesity warning level analysis results are considered at the same time, a composite loss function can be designed to comprehensively consider the errors of all prediction results. This usually involves assigning different weights to each prediction result and combining them into a total loss value.

[0095] After obtaining the total loss value, the computer system uses the backpropagation algorithm to update the configuration parameters (such as weights and biases) of the machine learning network. The backpropagation algorithm calculates the gradient of the loss function with respect to each network parameter and updates the parameter value in the opposite direction of the gradient, thereby gradually reducing the loss value and optimizing the model performance. During the optimization process, the computer system can use a variety of optimization algorithms, such as stochastic gradient descent (SGD), Adam, RMSprop, etc. These algorithms control the step size and direction of parameter updates by adjusting hyperparameters such as learning rate and momentum to achieve faster convergence speed and better optimization effect.

[0096] In addition, in order to avoid the occurrence of overfitting (i.e., the model performs well on the training data but performs poorly on unseen data), the computer system may also introduce regularization terms (such as L1 regularization, L2 regularization) or use early stopping methods and other technical means to limit the complexity of the model during the training process. For example, suppose that in the initial stage, the classification accuracy of the obesity warning level of the machine learning network on the test set is only 70%. Through the optimization process of step S2035, the computer system gradually adjusts the configuration parameters such as the weight and bias of the network. After many iterations, the accuracy on the test set increased to more than 90%. This shows that the model can more accurately predict the user's obesity risk level, thereby providing strong support for obesity prevention and control work.

[0097] Through the optimization process of step S2035, the computer system not only improves the prediction accuracy of the model but also enhances its generalization ability so that the model can maintain good performance on a wider range of data sets.

[0098] As an implementation manner, the step S2035, optimizing the network configuration parameters of the machine learning network according to the detail recognition result, the overall recognition result, and the error between the obesity warning level analysis result and the warning priori mark, may specifically include:

[0099] Step S20351: Generate a first evaluation function value through the error between the detail recognition result and the detail priori mark corresponding to the obesity big data sample.

[0100] In the embodiment of the present application, the detail recognition component is responsible for extracting and identifying specific obesity feature details, such as weight, waist circumference, body fat percentage, etc., from the feature vector set of the obesity big data sample. These detail recognition results are obtained after processing and analyzing the input data based on a machine learning model (such as a neural network). At the same time, each obesity big data sample is accompanied by a detail prior tag determined by experts or historical data, which represents the true state or value of the sample on a specific obesity feature.

[0101] The core task of step S20351 is to compare the difference between the detail recognition result and the detail prior mark, and quantify the difference as a loss value. The size of the loss value reflects the degree of deviation between the model prediction result and the actual situation, and is one of the important indicators for evaluating model performance.

[0102] In order to calculate the loss value, the computer system selects an appropriate loss function. For classification problems (such as determining whether the weight is overweight, whether the waist circumference is too thick, and other binary classification problems), commonly used loss functions include cross-entropy loss (Cross-EntropyLoss). For regression problems (such as predicting a specific body fat percentage value), mean squared error loss (MeanSquared Error Loss) or other suitable regression loss functions may be used.

[0103] Taking the classification problem as an example, assuming that the result of detail recognition is a binary classification prediction of a certain obesity feature (such as predicting whether the weight is overweight), the detail prior label is the true classification label of the feature (0 or 1). For the jth detail feature of the i-th sample, the cross entropy loss calculation formula is as follows:

[0104]

[0105] Among them, y detail,ij represents the true classification label (0 or 1) of the jth detail feature of the i-th sample, It indicates the probability that the model predicts that the feature is a positive class (in a binary classification problem, it is usually also predicted as a negative class, but the cross entropy loss function only cares about the predicted probability of the positive class, because the predicted probability of the negative class can be calculated by 1-). In practical applications, there may be more than one detail feature, and each sample may contain multiple detail features. Therefore, for the i-th sample, the computer system sums the losses of all its detail features to obtain the total loss of detail recognition for the sample:

[0106]

[0107] Among them, M represents the number of detail features of the i-th sample.

[0108] Finally, in order to obtain the first evaluation function value (ie, total loss) on the entire training set or validation set, the computer system sums up the total loss of detail recognition of all samples.

[0109] If the sample contains other detailed features (such as waist circumference, body fat percentage, etc.), similar loss calculations need to be performed on these features, and the results are summed to obtain the total loss of detail recognition for the sample. Finally, the total loss of detail recognition for all samples will be further summed to obtain the first evaluation function value (total loss) used to optimize the machine learning network.

[0110] Step S20352: Generate a second evaluation function value through the error between the overall recognition result and the overall priori label corresponding to the obesity big data sample.

[0111] Step S20352 focuses on evaluating the difference between the overall recognition result and the overall prior label, and generates a second evaluation function value accordingly. This process is crucial for guiding the adjustment of network parameters and improving model performance.

[0112] In the embodiment of the present application, the overall recognition component is responsible for comprehensively analyzing the feature vector set of the obesity big data sample to obtain the overall obesity status assessment of the user. This process may involve the integration of multiple obesity features, the modeling of complex relationships, and the prediction of potential risks. The overall recognition result is obtained after high-level abstraction and reasoning of the input data based on a machine learning model (such as a deep neural network, an integrated learning model, etc.).

[0113] At the same time, each obesity big data sample is accompanied by an overall prior label, which is usually given by experts based on the user's comprehensive obesity status, or based on historical data and statistical models. The overall prior label represents the true evaluation result of the sample in terms of overall obesity status, such as classification labels such as "low risk", "medium risk" or "high risk".

[0114] Step S20352 compares the difference between the overall recognition result and the overall prior mark, and quantifies the difference as a loss value. The size of the loss value directly reflects the degree of deviation between the model prediction result and the actual situation, and is an important indicator for evaluating the performance of the model in evaluating the overall obesity status. In order to calculate the loss value, the computer system selects a suitable loss function. For classification problems (such as obesity risk level classification), commonly used loss functions include cross-entropy loss. The cross-entropy loss function can effectively measure the difference between the predicted probability distribution and the true probability distribution, and is one of the widely used loss functions in classification problems.

[0115] Assume that the overall recognition result is a probability distribution vector, which indicates the probability that the model predicts that the sample belongs to each obesity risk level; and the overall prior label is a one-hot encoding vector, which indicates the actual obesity risk level of the sample. For the entire training set or validation set, the second evaluation function value (total loss) is the average or sum of the loss values ​​of all samples, depending on the definition of the loss function and the optimization goal. By calculating the overall recognition loss of each sample and summing or averaging them, the computer system can obtain the second evaluation function value for optimizing the machine learning network, which will guide the subsequent network parameter adjustment process to improve the accuracy and robustness of the model in the overall obesity status assessment.

[0116] Step S20353: Generate a third evaluation function value through the error between the obesity warning level analysis result and the warning level priori mark corresponding to the obesity big data sample.

[0117] In the embodiment of the present application, after multiple steps such as detail recognition and overall recognition, the warning level analysis component will output the obesity warning level analysis results of each obesity big data sample based on the previous processing results. These results are usually given in the form of classification labels or probability distributions, indicating the obesity risk level of the model-predicted sample. At the same time, each sample is accompanied by a warning level prior mark, which is the real obesity risk level of the sample determined by experts or historical data.

[0118] Step S20353 compares the difference between the warning level analysis result and the warning level priori mark, and calculates the loss value generated by the difference. The size of the loss value directly reflects the degree of deviation between the model prediction result and the actual situation, and is an important indicator for evaluating model performance.

[0119] To calculate the loss value, the computer system selects an appropriate loss function. For classification problems, especially when the output is a classification label, commonly used loss functions include cross-entropy loss. The cross-entropy loss function can effectively measure the difference between the predicted probability distribution and the true probability distribution (or one-hot encoded label).

[0120] Through the calculation of step S20353, the computer system obtains a quantitative index (the third evaluation function value) for evaluating the performance of the model in obesity warning level analysis and provides a clear direction for subsequent optimization.

[0121] Step S20354: obtaining a target evaluation function value according to the first evaluation function value, the second evaluation function value and the third evaluation function value, and obtaining a network change vector according to the target evaluation function value.

[0122] In step S20354, the computer system comprehensively considers the losses in three aspects, namely, detail recognition, overall recognition and obesity warning level analysis (i.e., the first evaluation function value, the second evaluation function value and the third evaluation function value), to generate a comprehensive target evaluation function value, and then calculates the network change vector based on the value to guide the adjustment of network parameters.

[0123] In steps S20351, S20352 and S20353, the computer system has calculated the detail recognition loss, the overall recognition loss and the obesity warning level analysis loss respectively. These loss values ​​quantify the difference between the performance of the model in detail recognition, overall recognition and warning level analysis and the actual situation.

[0124] In order to obtain a comprehensive evaluation index, the computer system integrates these three loss values ​​into a target evaluation function value. For example, it is achieved by weighted summation, and the weight of each loss value reflects the importance of the performance of this aspect to the overall model performance. The selection of weights may be based on prior knowledge, expert opinions, or determined through experimental methods such as cross-validation.

[0125] After obtaining the target evaluation function value, the computer system then calculates the network change vector based on this value. This step is usually implemented through the backpropagation algorithm, which uses the chain rule to calculate the gradient of the loss function with respect to the network parameters, that is, the network change vector. The network change vector indicates how to adjust the network parameters to reduce the loss value and thus improve the model performance.

[0126] During the back-propagation process, the computer system traverses the network layer by layer, starting from the output layer, and calculates the gradient of the previous layer based on the gradient of the current layer until it reaches the input layer. For each layer, the computer system updates the parameters of that layer (such as weights and biases) based on the gradient of that layer, usually using some form of gradient descent (such as stochastic gradient descent SGD, Adam optimizer, etc.) for updating.

[0127] After obtaining Ltotal, the computer system starts the back propagation process, calculates the gradient layer by layer and updates the network parameters. Taking a certain layer as an example, assuming that the weight of the layer is W, the bias is b, the activation function is σ, the input is x, and the output is y = σ (Wx + b), then the gradient of the layer with respect to the loss function can be expressed as and According to these gradient values, the computer system updates the weights and biases of the layer in the hope of reducing the loss value in the next iteration. By repeatedly executing the process of steps S20351 to S20355, the computer system can gradually optimize the performance of the machine learning network and improve the accuracy and reliability of obesity risk warning.

[0128] Step S20355: Iterate the configuration parameters of the machine learning network according to the network change vector.

[0129] Step S20355 adjusts and optimizes the configuration parameters of the machine learning network according to the network change vector calculated in the previous step. This process is carried out in an iterative manner, aiming to gradually reduce the prediction error and improve the accuracy and generalization ability of the model.

[0130] In step S20354, the computer system has calculated the target evaluation function value (total loss) based on the first evaluation function value (detail recognition loss), the second evaluation function value (overall recognition loss) and the third evaluation function value (obesity warning level analysis loss), and further obtained the network change vector through the back propagation algorithm. This vector contains specific information on how to adjust the network configuration parameters (such as weights and biases) to reduce the loss.

[0131] In step S20355, the computer system will use these network change vectors to iteratively update the configuration parameters of the machine learning network. The iterative update process usually follows an optimization algorithm, such as stochastic gradient descent (SGD), Adam, RMSprop, etc. These algorithms determine the direction and step size of parameter updates by calculating the sign and size of the gradient (i.e., the component of the network change vector). Taking the SGD algorithm as an example, its update rule can be simplified as:

[0132] θ t+1 =θ t -η·g t ;

[0133] Among them, θ t represents the network parameter value at the tth iteration, θ t+1 represents the updated parameter value, η is the learning rate (a hyperparameter that controls the update step size), g t is the network change vector calculated at the tth iteration. In practical applications, since machine learning networks usually contain a large number of parameters, the above update process will be performed separately for each parameter. This means that for each weight and bias in the network, the computer system will update it according to its corresponding gradient value.

[0134] By continuously repeating the process of steps S20351 to S20355, the computer system can gradually adjust and optimize the configuration parameters of the machine learning network so that it can better adapt to the characteristics and distribution patterns of obesity big data, thereby improving the accuracy and reliability of obesity risk warning.

[0135] As an implementation manner, the step S30, using the detailed implicit representation of the plurality of obesity feature data as topological vertices to generate a first topological structure corresponding to the detailed implicit representation, and using the overall implicit representation of the plurality of obesity feature data as topological vertices to generate a second topological structure corresponding to the overall implicit representation, may specifically include:

[0136] Step S31: Obtain a first commonality measurement result between multiple mined detail implicit representations, and obtain a second commonality measurement result between multiple overall implicit representations.

[0137] In step S31, the computer system first focuses on the relationship between multiple implicit representations of details. These implicit representations of details are obtained by deep learning or other machine learning techniques on obesity big data, and each captures the specific performance of the user in different obesity characteristics (such as weight, waist circumference, body fat percentage, etc.). In order to understand the intrinsic connection between these characteristics, the computer system calculates the commonality measurement results between them. The commonality measurement results reflect the similarity or correlation between the implicit representations of details. This similarity may be based on a variety of factors, such as the proximity of feature values, the consistency of change trends, or their importance in the overall obesity status assessment. In order to quantify this commonality, the computer system can use a variety of measurement methods, such as cosine similarity, Pearson correlation coefficient, Euclidean distance, etc.

[0138] Taking cosine similarity as an example, suppose there are two detail implicit representation vectors A and B, which represent the implicit representation of the weight and waist circumference features of two users respectively. Vector A can be expressed as [weight A, waist circumference A], and vector B can be expressed as [weight B, waist circumference B]. The calculation formula of cosine similarity is as follows:

[0139]

[0140] By calculating the cosine similarity between vectors A and B, the computer system can obtain a measure of the commonality between them. If the similarity value is close to 1, it means that the two details are very close in direction and size, that is, they have a high degree of similarity in weight and waist circumference; if the similarity value is close to 0, it means that there is almost no commonality between them.

[0141] Similarly, the computer system will also measure the commonalities between multiple overall implicit representations. The overall implicit representation is a high-level abstraction that integrates multiple detailed features and reflects the user's overall obesity status. Calculating the commonality measurement results between the overall implicit representations helps to understand the similarities and differences between different users in their overall obesity status. This process can also use measurement methods such as cosine similarity and Pearson correlation coefficient. Through the processing of step S31, the computer system not only obtains the first commonality measurement result between the detailed implicit representations, but also obtains the second commonality measurement result between the overall implicit representations. These measurement results provide key information for the subsequent construction of the topological structure, allowing the computer system to more accurately grasp the intrinsic connection and overall distribution law between obesity feature data.

[0142] Step S32: determining a first link state between the plurality of detail implicit representations according to the first commonality measurement result, and determining a second link state between the plurality of overall implicit representations according to the second commonality measurement result.

[0143] Step S32 determines the link status between different implicit representations based on the commonality measurement results, which is a key step in forming a meaningful topological structure.

[0144] In step S31, the computer system has calculated the first commonality measurement results between multiple detail implicit representations and the second commonality measurement results between multiple overall implicit representations. These commonality measurement results quantify the similarity or correlation between the implicit representations and provide a basis for determining the link status between them.

[0145] In step S32, the computer system sets the link status according to the commonality measurement results. There are usually two types of link status: connected (indicating a strong association between two implicit representations) and disconnected (indicating a weak association or no obvious association between two implicit representations).

[0146] For the detail implicit representation, the computer system compares the first commonality measurement result with the preset link threshold. If the commonality measurement result between a pair of detail implicit representations is higher than the link threshold, the computer system believes that there is a strong correlation between the two implicit representations, so they are connected by a connecting line in the topological structure to form a first link state in the first topological structure. On the contrary, if the commonality measurement result is lower than the link threshold, the two implicit representations are not connected in the topological structure. Similarly, for the overall implicit representation, the computer system will also compare the second commonality measurement result with the preset link threshold to determine the second link state between them. This process is the same as the processing method of the detail implicit representation, except that the data objects are different.

[0147] Assume that there are two detail implicit representation vectors A and B, which represent the implicit representation of the weight and waist circumference features of two users respectively. In step S31, the computer system has calculated the cosine similarity between them (as a commonality measurement result), assuming that the similarity value is 0.9. If the preset link threshold is 0.8, then since the similarity value is higher than the link threshold, the computer system connects vectors A and B through a connecting line in the first topological structure, indicating that they have a high degree of similarity in weight and waist circumference features.

[0148] Similarly, suppose there are two overall implicit representation vectors C and D, representing the overall obesity status of the two users respectively. After calculating the commonality measurement results between them (assuming that cosine similarity is also used as the measurement method), if the similarity value is also higher than the preset link threshold, vectors C and D are connected in the second topology structure, indicating that the two users have similarities in overall obesity status.

[0149] Through the processing of step S32, the computer system successfully converts the commonality measurement results between implicit representations into link states in the topological structure, providing an intuitive and insightful data representation for subsequent analysis and early warning.

[0150] Step S33: taking the plurality of detail implicit representations as topological vertices, and generating a first topological structure according to a first link state between the plurality of detail implicit representations.

[0151] Step S33 takes multiple implicit representations of details as topological vertices and generates a first topological structure according to the first link state between them. Specifically, in step S33, the computer system regards multiple implicit representations of details as vertices of the topological structure. These implicit representations of details are mined from obesity big data through machine learning models (such as deep learning networks), and each of them represents the specific performance of the user in different obesity characteristics, such as weight, waist circumference, body fat percentage, etc. Each implicit representation of details can be regarded as a feature vector or feature map, which contains rich detail information.

[0152] Next, the computer system constructs a connection relationship between vertices according to the first link state determined in step S32. The first link state is obtained based on the commonality measurement result between the detail implicit representations, which reflects the similarity and correlation between different detail features. If the commonality measurement result between two detail implicit representations is higher than a preset link threshold, the link state between them is connected; otherwise, it is disconnected.

[0153] In practical applications, a computer system can use graphics processing software or an algorithm library to generate a topology graph. It first creates an empty graphics object, and then adds each detail implicit representation as a vertex to the graph. Subsequently, according to the first link state, connecting lines are drawn in the graph to represent the connection relationship between the vertices. If two vertices are connected, a straight line or curve is drawn between them; if they are not connected, no connecting line is drawn.

[0154] Step S34: taking the plurality of overall implicit representations as topological vertices, and generating a second topological structure according to a second link state between the plurality of overall implicit representations.

[0155] Step S34 is the last step of constructing the topological structure of obesity feature data, which focuses on integrating multiple overall implicit representations into topological vertices and generating a second topological structure based on the second link state between these vertices. Specifically, in step S34, the computer system first regards multiple overall implicit representations as vertices of the second topological structure. The overall implicit representation is obtained by comprehensively analyzing multiple detailed implicit representations, and they represent the overall obesity status of the user. These overall implicit representations may be obtained by abstracting and fusing detailed features through a machine learning model (such as a deep learning network), and each representation contains rich comprehensive information. Next, the computer system constructs the connection relationship between the vertices based on the second link state determined in step S32. The second link state is obtained based on the commonality measurement results between the overall implicit representations, which reflects the similarities and differences between different users in the overall obesity status. If the commonality measurement result between two overall implicit representations is higher than the preset link threshold, the link state between them is connected; otherwise, it is disconnected.

[0156] In actual operation, the computer system generates the second topology structure graph using a method similar to step S33. It first creates a new graph object and adds each overall implicit representation as a vertex to the graph. Then, according to the second link state, a connection line is drawn in the graph to represent the connection relationship between the vertices. If two vertices are connected, a straight line or curve is drawn between them; if they are not connected, no connection line is drawn.

[0157] As an implementation manner, the step S40 of iterating the second topological structure according to the link status between the topological vertices in the first topological structure to obtain the target topological structure may specifically include:

[0158] Step S41: iterating the link states between topological vertices in the second topological structure through the link states between topological vertices in the first topological structure;

[0159] Step S42: iterating the overall implicit representation represented by the topological vertices in the second topological structure according to the iterated link state to obtain a target topological structure.

[0160] In an embodiment of the present application, step S40 optimizes and iterates the second topological structure (the topological structure based on the overall implicit representation) by analyzing the first topological structure (the topological structure based on the implicit representation of the details), thereby obtaining a more accurate and meaningful target topological structure. In step S41, the computer system first deeply analyzes the link status in the first topological structure. The first topological structure is constructed based on the implicit representation of the details, which reveals the intrinsic connections and similarities between different detail features. These link states reflect the importance of detail features in obesity risk assessment and their mutual influence.

[0161] Next, the computer system guides the iteration of the link states in the second topology according to the link states in the first topology. The second topology is constructed based on the overall implicit representation, which shows the similarities and differences in the overall obesity status of different users. However, the initially constructed second topology may not fully capture all important correlations and differences, so it needs to be optimized through an iterative process.

[0162] During the iteration process, the computer system considers highly connected detail implicit representation vertex pairs in the first topological structure. If two detail implicit representation vertices are closely connected in the first topological structure, the features they represent may have similar importance or influence each other in obesity risk assessment. Therefore, the computer system checks the link status between the overall implicit representation vertices (i.e., vertices in the second topological structure) corresponding to these two detail implicit representation vertices. If the two overall implicit representation vertices are already connected in the second topological structure, the computer system may strengthen the connection weight between them to reflect the stronger similarity.

[0163] If the two overall implicit representation vertices are not connected in the second topological structure, but there should be a strong association between them according to the link state in the first topological structure, then the computer system adds a new connection line for them in the second topological structure, thereby updating the link state.

[0164] In step S41, the link state in the second topological structure has been iteratively updated according to the link state in the first topological structure. Next, in step S42, the computer system will further optimize the overall implicit representation represented by the topological vertices in the second topological structure according to the iterated link state. The core of this step is to use the iterated link state to guide the update of the overall implicit representation. Since the link state reflects the similarities and differences in the overall obesity status between different users, these relationships can be better reflected by adjusting the overall implicit representation.

[0165] Specifically, the computer system can adopt the following strategies to iteratively update the overall implicit representation:

[0166] Weighted average: For each overall implicit representation vertex, the computer system can perform a weighted average based on the overall implicit representation of its neighbor vertices. The weight distribution can be determined based on the strength of the link state, that is, the neighbor vertices with higher connection weights contribute more to the overall implicit representation of the target vertex.

[0167] Feature fusion: In addition to weighted averaging, the computer system can also consider the feature fusion of highly similar overall implicit representation vertices. For example, if two users show high similarity in multiple detail features, their overall implicit representations can be fused to form a more representative overall implicit representation.

[0168] Nonlinear transformation: In some cases, simple weighted average or feature fusion may not be sufficient to capture the complex relationship between the overall implicit representation. In this case, the computer system can use nonlinear transformations (such as neural network layers) to further process the overall implicit representation to generate more discriminative and expressive feature vectors.

[0169] For example, assuming that in the second topological structure, after the iterative update in step S41, a connection is established between the overall implicit representation vertices of user A and user B. In step S42, the computer system will further optimize the overall implicit representation of these two vertices.

[0170] First, the computer system may consider the similarity between user A and user B in multiple detail features. For example, their weight, waist circumference, and body fat percentage all show high similarity in the first topological structure. Based on this information, the computer system may adopt a weighted average strategy to update the overall implicit representation of user A and user B. Specifically, it will merge part of the features of the overall implicit representation of user B into the overall implicit representation of user A (or vice versa) according to the connection weight to reflect the similarity between them. In addition, if the system detects that the similarity between user A and user B in the overall obesity status exceeds the range that can be described by a simple weighted average, it may adopt a more complex nonlinear transformation to further process the overall implicit representation of the two vertices. For example, it can pass the overall implicit representation of the two vertices as input to a neural network layer and output a new, more representative overall implicit representation vector. Through the processing of step S42, the computer system is able to generate a more accurate and meaningful target topological structure. This target topological structure not only reflects the similarities and differences between different users in the overall obesity status, but also improves the accuracy and expression ability of the overall implicit representation through an iterative update process. This is of great significance for subsequent obesity risk assessment and early warning.

[0171] As an implementation manner, the step S41, iterating the link states between the topological vertices in the second topological structure through the link states between the topological vertices in the first topological structure, may specifically include:

[0172] Step S411: obtaining the association between each topological vertex in the first topological structure and each topological vertex in the second topological structure;

[0173] Step S412: verifying the matching of the link status of any topological vertex tuple in the first topological structure and the second topological structure through the association situation;

[0174] Step S413: If it is identified that the link state of the target topology vertex tuple in the first topology structure is different from the link state in the second topology structure, iterate the link state of the target topology vertex tuple in the second topology structure through the link state of the target topology vertex tuple in the first topology structure.

[0175] Step S41 involves iteratively updating the link status in the second topological structure (the topological structure based on the overall implicit representation) through the link status in the first topological structure (the topological structure based on the detailed implicit representation). In step S411, the computer system clarifies the association between the vertices in the first topological structure and the second topological structure. Since the first topological structure is constructed based on the detailed implicit representation, its vertices represent different detailed features (such as weight, waist circumference, body fat percentage, etc.); and the second topological structure is constructed based on the overall implicit representation, and its vertices represent the performance of different users or user groups in terms of overall obesity. Therefore, to obtain the association between the two, the computer system establishes a mapping relationship to correspond the detailed features in the first topological structure with the users or user groups in the second topological structure.

[0176] In practical applications, this mapping relationship may be established through the obesity feature data set of users or user groups. For example, the obesity feature data set of each user or user group contains multiple detailed feature values ​​(these values ​​are processed to form detailed implicit representations), and also contains the overall obesity status assessment of the user or user group (which forms the basis of the overall implicit representation). The computer system can analyze these data sets to find the corresponding relationship between the detailed features and the overall obesity status, thereby determining the association between the vertices in the first topological structure and the second topological structure.

[0177] In order to obtain the association between the vertices in the first topological structure and the second topological structure, the computer system traverses the obesity feature data set and associates the detailed feature value of each user with the corresponding overall obesity level assessment. In this way, when the computer system finds a vertex representing weight in the first topological structure, it can know which vertices in the second topological structure (i.e., which users or user groups) this vertex is associated with.

[0178] In step S412, after determining the association between each vertex in the first topological structure and the second topological structure, the computer system will then verify the matching of the link status of any topological vertex tuple (i.e., a pair of two vertices) in the first topological structure and the second topological structure. The purpose of this step is to check whether the link status between the same or associated vertex pairs in the two topological structures is consistent. If not, it may be necessary to perform iterative updates.

[0179] The matching verification process usually involves comparing the link states of vertex pairs in two topological structures. In the first topological structure, the link state reflects the similarity and correlation between the detailed features; while in the second topological structure, the link state reflects the similarity and difference of the overall obesity status of users or user groups. The computer system ensures that the two link states are logically consistent, that is, if two detailed features are closely connected in the first topological structure (indicating that there is a strong correlation between them), then the corresponding users or user groups should also show strong similarities in the second topological structure (that is, the link state between them should also be strong).

[0180] Continuing with the previous example, assume that in the first topology, the vertices representing weight and waist circumference are closely connected (because weight and waist circumference are usually considered highly correlated features in obesity risk assessment). At the same time, in the second topology, there are two users A and B, whose overall obesity level assessments are similar. However, in the initially constructed second topology, the vertices of user A and user B are not directly connected (i.e., the link state between them is weak).

[0181] In this case, the computer system verifies the weight-waistline vertex pair (vertex dyad in the first topology) and the user A-user B vertex pair (vertex dyad in the second topology) for matching. Since weight and waistline are closely linked in the first topology, and user A and user B are similar in overall obesity level assessment, the computer system expects that there should also be a strong link state between the vertices of user A and user B in the second topology. If this is not the case (i.e., the link state between the vertices of user A and user B is weak), the computer system performs an iterative update to correct this.

[0182] In step S412, if the computer system identifies that the link state of a target topological vertex tuple in the first topological structure is inconsistent with the link state in the second topological structure, then it needs to correct the inconsistency by iteratively updating the link state in the second topological structure through the link state in the first topological structure.

[0183] Specifically, the computer system determines the link state that the corresponding vertex pair in the second topological structure should have based on the link state of the target topological vertex tuple in the first topological structure. If the link state in the first topological structure is strong (indicating a strong correlation between the two detail features), then the corresponding vertex pair in the second topological structure should also have a strong link state (indicating a strong similarity between the two users or user groups in terms of overall obesity status).

[0184] To achieve this, a computer system can employ one or a combination of the following strategies:

[0185] Strengthening connections: If the link state of the target topology vertex pair in the second topology is weak or non-existent, the computer system adds a new connection line between them and assigns an initial connection weight. This connection weight can be determined based on the link strength of the corresponding vertex pair in the first topology.

[0186] Adjusting connection weights: If a link already exists for the target topology vertex pair in the second topology, but the connection weight does not match the link strength of the corresponding vertex pair in the first topology, the computer system adjusts the connection weight to make it closer to the link strength in the first topology.

[0187] Fusion features: In some cases, if the link state of two vertices in the second topology is too different from that in the first topology, simply strengthening the connection or adjusting the connection weight may not be enough to correct the inconsistency. At this time, we can consider fusing the overall implicit representations represented by the two vertices to generate a new, more representative overall implicit representation.

[0188] Continuing with the previous example, assume that the computer system has identified that the link state between user A and user B in the second topology is weak, which is inconsistent with the strong link state between their corresponding detailed features (weight and waist circumference) in the first topology. In order to correct this inconsistency, the computer system takes the following steps to iterate and update:

[0189] Analyze the first topological structure: The computer system will first review the strong link status between the weight and waist circumference vertices in the first topological structure to understand the importance of these two features in obesity risk assessment and their mutual influence.

[0190] Strengthening the connection: Based on the analysis results in the first topological structure, the computer system adds a new connection line between the vertices of user A and user B in the second topological structure (if there is no connection before), and assigns an initial connection weight. This connection weight can be determined according to the link strength of weight and waist circumference in the first topological structure.

[0191] Adjusting the overall implicit representation: If simply strengthening the connection is not enough to fully reflect the similarity between user A and user B in terms of their overall obesity status, the computer system may also consider further adjusting their overall implicit representation. For example, it may fuse the overall implicit representations of user A and user B to generate a new feature vector that more accurately reflects their overall obesity status.

[0192] Through the iterative update process of step S41, the computer system can ensure that the link states in the first topology structure and the second topology structure are logically consistent. This consistency is crucial for subsequent obesity risk assessment and early warning, because it ensures that the analysis results based on detailed features and overall obesity status are mutually consistent and complementary.

[0193] As an implementation manner, the step S50, determining the target obesity big data from the reference obesity big data in the reference obesity big data set according to the commonality measurement result between the topological vertices in the target topological structure, may specifically include:

[0194] Step S51: determining a first target topological vertex corresponding to the target obesity characteristic data in the target topological structure;

[0195] Step S52: Determine target obesity big data in the reference obesity big data set based on the third commonality measurement result between the first target topological vertex and the remaining topological vertices in the target topological structure, and the remaining topological vertices in the target topological structure are topological vertices in the target topological structure that have a link state with the first target topological vertex.

[0196] In step S51, the computer system locates a topological vertex directly associated with the target obesity feature data in the target topological structure, which is called the first target topological vertex. This target vertex represents the user's performance on a specific obesity feature and is the starting point for subsequent screening of the target obesity big data.

[0197] In order to determine the first target topological vertex, the computer system first needs to clarify the specific content of the target obesity feature data. These feature data may include key indicators such as the user's weight, waist circumference, body fat percentage, etc. After implicit representation mining and topological structure construction, these indicators will exist in the target topological structure in the form of specific topological vertices. Next, the computer system traverses all topological vertices in the target topological structure and finds the vertex that best matches it by comparing the similarity of each vertex's attributes (such as labels, feature vectors, etc.) with the target obesity feature data. This matching process may involve a variety of methods such as cosine similarity calculation of feature vectors and Euclidean distance measurement, depending on the representation form of the topological vertex and the commonality metric.

[0198] For example, suppose the target obesity feature data is the weight information of user A. After implicit representation mining and topological structure construction, the weight feature corresponds to a specific topological vertex V1 in the target topological structure. In this scenario, the task of the computer system is to find vertex V1 in the target topological structure and determine it as the first target topological vertex. To achieve this goal, the computer system may traverse all vertices in the target topological structure and calculate the similarity of each vertex with the target weight information. For example, if the attributes of vertex V1 contain a weight feature vector [75kg, 0.05 (noise)] (where 75kg is the actual weight of user A and 0.05 is the added noise value), then the computer system will compare this feature vector with the target weight information and determine whether vertex V1 is the first target topological vertex by calculating the similarity.

[0199] In step S52, after the first target topological vertex is determined, the next step of the computer system is to select the target obesity big data that best matches the target obesity feature data from the reference obesity big data set based on the commonality measurement results between the vertex and the remaining vertices in the target topological structure. This process involves an in-depth analysis of the correlation between topological vertices and an efficient retrieval of the reference obesity big data set.

[0200] The commonality measurement result is an important indicator reflecting the similarity and correlation between topological vertices. In the target topological structure, the commonality measurement result between the first target topological vertex and the remaining vertices reveals the comprehensive performance of the user in different obesity characteristics and their mutual relationship. By analyzing these commonality measurement results, the computer system can identify reference obesity big data that shows high consistency with the target obesity feature data in multiple dimensions. To achieve this goal, the computer system first calculates the third commonality measurement result between the first target topological vertex and all other vertices with link status in the target topological structure. This commonality measurement process may involve multiple measurement methods and threshold settings, depending on the characteristics of the topological structure and the needs of commonality measurement.

[0201] The computer system then compares these commonality measurement results with the data in the reference obesity big data set. This comparison process may include entering the commonality measurement results as search conditions into a database query statement, or using a machine learning model to classify and sort the reference obesity big data. Through this series of operations, the computer system can ultimately determine the target obesity big data that best matches the target obesity feature data.

[0202] Continuing with the previous example, assume that vertex V1 (representing the weight feature of user A) has been determined as the first target topology vertex. Next, the computer system analyzes the commonality measurement results between vertex V1 and the remaining vertices in the target topology structure (such as vertices representing waist circumference, body fat percentage, etc.).

[0203] These commonality measurement results reflect the comprehensive performance of user A in different obesity characteristics and their interrelationships. For example, if the commonality measurement result between vertex V1 and vertex V2 representing waist circumference is high (indicating that user A's weight and waist circumference are both at a high level), then the computer system will consider these two characteristics to be of high importance in assessing user A's obesity risk.

[0204] Based on these commonality measurement results, the computer system retrieves data records in the reference obesity big data set that are highly consistent with user A in terms of weight, waist circumference and other obesity characteristics. These data records may contain obesity characteristic data sets similar to user A and corresponding obesity risk assessment results.

[0205] By comparing and screening these data records, the computer system can eventually determine the target obesity big data that best matches the obesity characteristic data of user A. These data will provide key data support and analysis basis for subsequent obesity risk warnings.

[0206] As an implementation mode, the method may further include: determining a second target topological vertex corresponding to the target obesity feature data in the first topological structure. Based on this, the step S52: determining the target obesity big data in the reference obesity big data set according to the third commonality measurement result between the first target topological vertex and the remaining topological vertices in the target topological structure may specifically include:

[0207] Step S521: acquiring a target commonality measurement result according to a third commonality measurement result between the first target topology vertex and the remaining topology vertices in the target topology structure, and a fourth commonality measurement result between the second target topology vertex and the remaining topology vertices in the second topology structure, wherein the remaining topology vertices in the second topology structure are topology vertices in the second topology structure that are in a linked state with the first target topology vertex;

[0208] Step S522: determining target obesity big data in the reference obesity big data set according to the target commonality measurement result.

[0209] In step S521, the computer system first identifies two topological vertices in the first topological structure that are directly related to the target obesity characteristic data: a first target topological vertex (determined directly based on the target obesity characteristic data) and a second target topological vertex (as a supplement or verification of the first target vertex). These two vertices together form the basis for the system to analyze the user's obesity characteristics.

[0210] Next, the system calculates two sets of commonality measures:

[0211] The third commonality measurement result: This is the commonality measurement between the first target topological vertex and the remaining topological vertices with link status in the target topological structure. These measurement results reflect the association strength and similarity between the target obesity feature and other obesity features.

[0212] Fourth commonality measurement result: This is the commonality measurement between the second target topology vertex and the remaining topology vertices in the second topology structure that have a link state with the first target topology vertex. Since the second topology structure is constructed based on the overall implicit representation, these measurement results provide further verification and supplementation of the target obesity characteristics from the perspective of overall obesity status.

[0213] To integrate the two sets of commonality metric results, the computer system uses a comprehensive strategy to calculate the target commonality metric result. This strategy may involve weighted averaging, maximum and minimum selection, or more complex fusion algorithms, depending on the requirements and design of the system.

[0214] For example, assuming that the target obesity feature data is the waist circumference information of user A, after implicit representation mining and topological structure construction, the waist circumference feature corresponds to two topological vertices in the first topological structure: V1 (as the first target topological vertex) and V2 (as the second target topological vertex, which may represent the trend or pattern of waist circumference changes). At the same time, there are also vertices associated with V1 and V2 in the second topological structure (based on the overall implicit representation).

[0215] For the first topological structure, the computer system calculates the third commonality measurement results between V1 and all vertices that are linked to V1 (such as vertices representing weight, body fat percentage, etc.). These results reflect the direct correlation between waist circumference and other obesity characteristics.

[0216] For the second topology, the system calculates the fourth commonality measure between V2 and all vertices that are linked to V1 (these vertices in the second topology represent users or user groups that are similar to user A in terms of overall obesity). These results provide additional verification of waist circumference features from the perspective of overall obesity.

[0217] To obtain the target commonality measurement results, the computer system can adopt one of the following strategies:

[0218] Weighted average method: Assign different weights to the third and fourth commonality measurement results, and calculate the weighted average as the target commonality measurement result. The weight allocation can be based on domain knowledge, expert opinions, or experience accumulated during system optimization.

[0219] Fusion algorithm: Use machine learning models (such as neural networks, decision trees, etc.) to fuse the third and fourth commonality measurement results to generate a comprehensive target commonality measurement result. This method can capture the nonlinear relationship between measurement results and improve the prediction accuracy of the system.

[0220] In step S522, the computer system uses the target commonality measurement results calculated in step S521 to select the target obesity big data that best matches the target obesity feature data in the reference obesity big data set. This process is a comprehensive application of all previous steps and is also one of the core outputs of the obesity risk warning system.

[0221] To achieve this goal, the computer system uses the target commonality measurement results as retrieval conditions or scoring basis to traverse and evaluate the data in the reference obesity big data set. The evaluation process may involve the following steps:

[0222] Data preprocessing: The data in the reference obesity big data set were cleaned, transformed, and standardized to ensure that they were comparable with the data in the target topology and the second topology.

[0223] Similarity calculation: For each data record in the collection, calculate its similarity with the target commonality measure result. This similarity may be based on the cosine similarity of the feature vectors, Euclidean distance, or other metrics.

[0224] Sorting and screening: According to the similarity calculation results, the data in the reference obesity big data set are sorted. Then, according to the preset threshold or ranking position, the data records that best match the target commonality measurement results are selected as the target obesity big data.

[0225] In step S522, the system starts to traverse the reference obesity big data set. This set contains a large number of users' obesity feature data and corresponding obesity risk assessment results. For each data record in the set (representing a user or user group), the system calculates the similarity between it and the target commonality measurement result.

[0226] Similarity calculations can be based on multiple dimensions: direct comparison of waist circumference features, correlation analysis of other relevant obesity features, and verification from the perspective of overall obesity status. The system uses a comprehensive similarity measurement method (such as weighted sum, Manhattan distance, etc.) to integrate the similarity results of these dimensions into a comprehensive similarity score. Then, the system sorts the data in the reference obesity big data set according to the similarity score. The data records at the top are most consistent with the target commonality measurement results, so they are more likely to represent the actual obesity status of user A. Finally, the system selects a certain number of data records as target obesity big data based on the preset threshold or ranking position and outputs them to the user or health manager for further analysis and intervention.

[0227] Through the processing of step S522, the computer system selects the target obesity big data that best matches the target obesity characteristic data from the reference obesity big data set. These data provide strong data support and analysis basis for subsequent obesity risk warning and health management. At the same time, because the system adopts a comprehensive commonality measurement method and similarity calculation method, it can accurately identify key risk factors in complex and changeable obesity characteristic data, providing strong guarantee for the formulation of personalized health management plans.

[0228] As an implementation manner, the step S522, determining the target obesity big data in the reference obesity big data set according to the target commonality measurement result, may specifically include:

[0229] Step S5221: determining, in the target topological structure, a target matching topological vertex whose target commonality measurement result is greater than a reference critical result;

[0230] Step S5222: determining matching obesity feature data corresponding to the target matching topological vertex;

[0231] Step S5223: Determine the obesity big data including the matching obesity feature data to obtain target obesity big data.

[0232] In step S5221, the computer system identifies target matching topological vertices whose commonality measurement results significantly exceed a certain reference critical value in the target topological structure. These vertices represent feature nodes that are highly correlated with the target obesity feature data and have significant importance in the topological structure.

[0233] The reference critical result is a pre-set threshold used to distinguish the significance and non-significance of the commonality measurement results. The setting of this threshold is usually based on domain knowledge, expert experience or statistical results in the system optimization process. In practical applications, the reference critical result may be adjusted according to the specific application scenario and data characteristics. In order to determine the target matching topological vertex, the computer system traverses all vertices in the target topological structure and calculates the difference between the commonality measurement result of each vertex and the reference critical result. If the commonality measurement result of a vertex is greater than the reference critical result, it is regarded as the target matching topological vertex.

[0234] For example, suppose that in the target topology, there are multiple topological vertices representing different obesity characteristics, such as weight vertex V1, waist vertex V2, body fat percentage vertex V3, etc. These vertices are connected to each other through link states to form a complex topological network. In the previous step, the computer system has calculated the commonality measurement results between these vertices and obtained a comprehensive target commonality measurement result.

[0235] Now, the computer system determines which vertices have a commonality measurement result exceeding a reference critical result. Assuming that the reference critical result is set to 0.7 (a hypothetical threshold), the computer system traverses all vertices in the target topology and checks whether their commonality measurement results are greater than 0.7.

[0236] For example, if the commonality measurement result of waist vertex V2 is 0.85, which is greater than the reference critical result 0.7, then V2 is considered a target matching topological vertex. On the contrary, if the commonality measurement result of weight vertex V1 is 0.65, which is less than the reference critical result 0.7, then V1 will not be considered a target matching topological vertex.

[0237] In step S5222, after determining the target matching topological vertices, the next step of the computer system is to find the matching obesity feature data represented by these vertices. These feature data are data points that are highly correlated with the user's specific obesity characteristics and have an important impact on obesity risk assessment.

[0238] In order to find matching obesity feature data, the computer system searches for corresponding feature data in the attribute table or data dictionary of the topological structure according to the identification or attribute of the target matching topological vertex. These feature data may exist in the form of feature vectors, attribute values ​​or data tables, depending on the data representation method of the system.

[0239] Continuing with the previous example, assume that waistline vertex V2 is determined as a target matching topological vertex. After determining the identity of V2, the computer system searches for feature data associated with V2 in the attribute table or data dictionary of the topological structure.

[0240] In step S5223, after the matching obesity feature data is determined, the final task of the computer system is to find the obesity big data records containing these feature data, i.e., the target obesity big data. These data records are usually stored in the reference obesity big data set and are the basic data source used by the system for obesity risk assessment and early warning.

[0241] In order to find the target obesity big data, the computer system uses the matching obesity feature data as the retrieval condition or filtering rule to traverse and filter the reference obesity big data set. The screening process may involve a variety of technical means such as data query, pattern matching, and similarity calculation, depending on the data storage format and the system's retrieval capabilities. During the screening process, the computer system searches for data records that are highly consistent or similar to the matching obesity feature data. These records may contain information such as the user's detailed obesity feature data set, obesity risk assessment results, and health management recommendations. By comparing and analyzing these data records, the computer system can determine the target obesity big data that best matches the user's specific obesity features.

[0242] After screening, the computer system obtains a set of data records that highly match the target obesity feature data. These records represent user groups or individual cases that are most similar to the user's specific obesity features. By analyzing the obesity feature data set and evaluation results in these data records, the computer system can provide users with personalized obesity risk assessments and health management recommendations.

[0243] As an implementation manner, in step S30, using the detailed implicit representations of the plurality of obesity feature data as topological vertices to generate a first topological structure corresponding to the detailed implicit representations may further include:

[0244] Step S30A: filtering the detail implicit representations of a plurality of obesity feature data to obtain a target detail implicit representation;

[0245] Step S30B: using the target detail implicit representation as a topological vertex to generate a first topological structure corresponding to the detail implicit representation.

[0246] In step S30A, the computer system filters multiple implicit representations of details extracted from the obesity big data. These implicit representations of details are obtained by mining the original obesity feature data through implicit representations using a machine learning model (such as a deep learning network), and each captures the specific performance of the user in different obesity characteristics. However, not all implicit representations of details are equally important for subsequent topological structure construction and analysis. Therefore, the computer system uses a filtering step to screen out the most representative and relevant implicit representations of details, namely, the target implicit representations of details. The filtering process may involve a variety of methods and strategies, depending on the characteristics of the data and the analysis requirements. The following are some possible filtering methods:

[0247] Filtering based on feature importance: Use feature selection algorithms (such as recursive feature elimination, model-based feature selection, etc.) to evaluate the importance of each detail implicit representation. These algorithms build prediction models to quantify the predictive power of each feature for the target variable (such as obesity risk level) to determine which features are more important. By setting an importance threshold, the computer system can filter out detail implicit representations above the threshold as the target detail implicit representation.

[0248] Similarity-based filtering: If there is a high degree of similarity between multiple implicit representations of details, it may be necessary to retain only one of them as a representative. The computer system can filter out redundant features by calculating the similarity between implicit representations of details (such as cosine similarity, Pearson correlation coefficient, etc.) and setting a similarity threshold.

[0249] Filtering based on domain knowledge: In some cases, domain knowledge can be used to guide the filtering process. For example, based on the experience of medical research and clinical practice, certain obesity features (such as BMI, waist circumference, etc.) have a higher priority in assessing obesity risk. Therefore, the computer system can give priority to retaining the detailed implicit representation of these features.

[0250] At the same time, the computer system also calculated the similarity between the remaining features and found that there was a high similarity between two of them (such as weight and BMI, weight and body fat percentage). To avoid redundancy, it set a similarity threshold and filtered out the implicit representation of details whose similarity with the selected features was higher than the threshold.

[0251] Ultimately, the computer system arrived at a collection of implicit representations of target details that were most representative and relevant in assessing obesity risk.

[0252] In step S30B, the computer system uses the implicit representation of the target details screened out in step S30A as the vertices of the topological structure to construct a first topological structure. This topological structure reflects the intrinsic connection and similarity between different detail features, and provides an intuitive and insightful data representation for subsequent analysis and early warning. In order to build the topological structure, the computer system first needs to create an empty graph object as a container for the topological structure. Then, it adds each target detail implicit representation as a vertex to the graph, and determines the connection relationship between them based on the commonality measurement results between these vertices. The commonality measurement results can be obtained by calculating the similarity, distance or other measurement indicators between the vertices.

[0253] If the commonality measurement result between two vertices is higher than the preset link threshold, the computer system believes that the details represented by the two vertices have a strong correlation, so a connection line is drawn between them in the topology. Attributes such as the thickness or color of the connection line can be adjusted according to the specific value of the commonality measurement result to reflect the strength of the association between different pairs of vertices.

[0254] By constructing such a topological structure, the computer system not only intuitively displays the intrinsic connections and similarities between different detailed features, but also provides strong data support for subsequent obesity risk assessment and early warning. For example, if a user's weight value and weight change trend both show high risk characteristics (such as continuous weight gain), the computer system can quickly identify this risk pattern based on the connection relationship in the topological structure and issue a corresponding early warning signal.

[0255] The present application embodiment provides a computer system, such as Figure 2 As shown, the computer system 100 includes: a processor 101 and a memory 103. The processor 101 and the memory 103 are connected, such as through a bus 102. Optionally, the computer system 100 may also include a transceiver 104. It should be noted that in actual applications, the transceiver 104 is not limited to one, and the structure of the computer system 100 does not constitute a limitation on the embodiments of the present application.

[0256] An embodiment of the present application provides a computer system. The computer system in the embodiment of the present application includes: one or more processors; a memory; and one or more computer programs, wherein the one or more computer programs are stored in the memory and configured to be executed by the one or more processors, and when the one or more programs are executed by the processor, the above method is implemented.

Claims

1. An obesity risk warning method based on big data analysis, characterized in that: Applied to a computer system, the method comprises: Obtaining user obesity statistical big data to be analyzed and a reference obesity big data set, wherein the user obesity statistical big data to be analyzed includes target obesity feature data to be matched; Obesity information implicit representation mining is performed on the user obesity statistical big data to be analyzed and the reference obesity big data in the reference obesity big data set through a machine learning network to obtain obesity information implicit representations of multiple obesity feature data, wherein the multiple obesity feature data include the target obesity feature data, and the obesity information implicit representation includes an overall implicit representation and a detailed implicit representation; The detailed implicit representations of the plurality of obesity feature data are used as topological vertices to generate a first topological structure corresponding to the detailed implicit representations, and the overall implicit representations of the plurality of obesity feature data are used as topological vertices to generate a second topological structure corresponding to the overall implicit representation, wherein both the first topological structure and the second topological structure have a link state between topological vertices, and the link state includes connected or disconnected, wherein the topological vertices greater than a commonality threshold are connected by connecting lines, and the topological vertices less than the commonality threshold are not connected; Iterating the second topological structure according to the link states between the topological vertices in the first topological structure to obtain a target topological structure; Determining target obesity big data from the reference obesity big data in the reference obesity big data set according to the commonality measurement result between the topological vertices in the target topological structure; The warning level corresponding to the target obesity big data is used as the target warning level, and obesity warning is performed based on the target warning level.

2. The method according to claim 1, characterized in that The method further comprises: performing obesity information implicit representation mining on the user obesity statistical big data to be analyzed and the reference obesity big data in the reference obesity big data set through a machine learning network to obtain obesity information implicit representations of a plurality of obesity feature data, wherein the plurality of obesity feature data include the target obesity feature data, and the obesity information implicit representation includes an overall implicit representation and a detailed implicit representation. Acquire debugging learning samples, wherein the debugging learning samples include multiple obesity big data samples and a warning priori mark of each obesity big data sample, wherein the obesity big data samples include positive obesity big data samples and negative obesity big data samples, and the warning priori mark includes a detail priori mark, an overall priori mark, and a warning level priori mark; Obtaining a machine learning network to be debugged, wherein the machine learning network includes a data implicit representation mining component, an obesity feature data detail recognition component, an obesity feature data overall recognition component, and a warning level analysis component; The machine learning network is debugged through the debugging learning examples.

3. The method according to claim 2, characterized in that The debugging of the machine learning network through the debugging learning examples includes: The implicit representation mining component of the data of the machine learning network is used to mine the implicit representation of obesity information on the obesity big data samples, and the feature vector set of the obesity big data samples is obtained; Performing detail recognition on the feature vector set by the obesity feature data detail recognition component to obtain detail recognition results of multiple sample obesity feature data; The obesity feature data overall recognition component performs overall recognition on the feature vector set to obtain overall recognition results of multiple sample obesity feature data; Performing obesity warning level analysis on the sample obesity feature data in the feature vector set by the warning level analysis component to obtain an obesity warning level analysis result; The network configuration parameters of the machine learning network are optimized according to the detail recognition result, the overall recognition result, and the error between the obesity warning level analysis result and the warning priori mark.

4. The method according to claim 3, characterized in that The optimizing of the network configuration parameters of the machine learning network according to the detail recognition result, the overall recognition result, and the error between the obesity warning level analysis result and the warning priori mark comprises: Generate a first evaluation function value by using an error between the detail recognition result and a detail priori mark corresponding to the obesity big data sample; Generate a second evaluation function value by using the error between the overall recognition result and the overall priori mark corresponding to the obesity big data sample; The third evaluation function value is generated by the error between the obesity warning level analysis result and the warning level priori mark corresponding to the obesity big data sample; Obtaining a target evaluation function value according to the first evaluation function value, the second evaluation function value, and the third evaluation function value, and obtaining a network change vector according to the target evaluation function value; The configuration parameters of the machine learning network are iterated according to the network change vector.

5. The method according to claim 1, characterized in that The step of using the detailed implicit representations of the plurality of obesity feature data as topological vertices to generate a first topological structure corresponding to the detailed implicit representations, and using the overall implicit representations of the plurality of obesity feature data as topological vertices to generate a second topological structure corresponding to the overall implicit representation, comprises: Obtaining a first commonality measurement result between multiple mined detail implicit representations, and obtaining a second commonality measurement result between multiple overall implicit representations; Determining a first link state between the plurality of detail implicit representations according to the first commonality measurement result, and determining a second link state between the plurality of overall implicit representations according to the second commonality measurement result; Using the plurality of detail implicit representations as topological vertices, and generating a first topological structure according to first link states between the plurality of detail implicit representations; Using the plurality of overall implicit representations as topological vertices, and generating a second topological structure according to a second link state between the plurality of overall implicit representations; The iterating the second topological structure according to the link status between the topological vertices in the first topological structure to obtain the target topological structure includes: Iterating the link states between topological vertices in the second topological structure through the link states between topological vertices in the first topological structure; The overall implicit representation represented by the topological vertices in the second topological structure is iterated according to the iterated link state to obtain a target topological structure.

6. The method according to claim 5, characterized in that The iterating the link states between topological vertices in the second topological structure by using the link states between topological vertices in the first topological structure includes: Obtaining the association between each topological vertex in the first topological structure and each topological vertex in the second topological structure; Verify the matching of the link status of any topological vertex tuple in the first topological structure and the second topological structure through the association situation; If it is identified that the link state of the target topology vertex tuple in the first topology structure is different from the link state in the second topology structure, the link state of the target topology vertex tuple in the second topology structure is iterated through the link state of the target topology vertex tuple in the first topology structure.

7. The method according to claim 1, characterized in that The determining target obesity big data from the reference obesity big data in the reference obesity big data set according to the commonality measurement result between the topological vertices in the target topological structure comprises: Determining a first target topological vertex corresponding to the target obesity characteristic data in the target topological structure; Based on the third commonality measurement result between the first target topological vertex and the remaining topological vertices in the target topological structure, target obesity big data is determined in the reference obesity big data set, and the remaining topological vertices in the target topological structure are topological vertices in the target topological structure that have a link state with the first target topological vertex.

8. The method according to claim 7, characterized in that The method further comprises: Determining a second target topological vertex corresponding to the target obesity feature data in the first topological structure; determining target obesity big data in the reference obesity big data set based on a third commonality measurement result between the first target topological vertex and the remaining topological vertices in the target topological structure, comprises: Obtain a target commonality measurement result according to a third commonality measurement result between the first target topology vertex and the remaining topology vertices in the target topology structure, and a fourth commonality measurement result between the second target topology vertex and the remaining topology vertices in the second topology structure, wherein the remaining topology vertices in the second topology structure are topology vertices in the second topology structure that have a link state with the first target topology vertex; Target obesity big data is determined in the reference obesity big data set according to the target commonality measurement result.

9. The method according to claim 8, characterized in that The step of determining target obesity big data from the reference obesity big data set according to the target commonality measurement result includes: Determining, in the target topological structure, a target matching topological vertex whose target commonality measurement result is greater than a reference critical result; Determine matching obesity feature data corresponding to the target matching topological vertex; Determine the obesity big data including the matching obesity feature data to obtain target obesity big data; The step of using the detailed implicit representations of the plurality of obesity feature data as topological vertices to generate a first topological structure corresponding to the detailed implicit representations includes: Filtering the detailed implicit representations of multiple obesity feature data to obtain the target detailed implicit representation; The target detail implicit representation is used as a topological vertex to generate a first topological structure corresponding to the detail implicit representation.

10. A computer system, characterized in that: include: one or more processors; Memory; one or more computer programs; The one or more computer programs are stored in the memory and configured to be executed by the one or more processors, and when the one or more computer programs are executed by the processors, the method according to any one of claims 1 to 9 is implemented.

Images