Historical-real-time data fused health data analysis method, system and processor

Abstract

The invention relates to the technical field of medical health, in particular to a health data analysis method and system fusing historical-real-time data and a processor, and the method comprises the steps: obtaining one or more different types of real-time data uploaded by a patient, and carrying out the preprocessing of the real-time data to obtain real-time health features; performing multi-modal processing on the real-time health features; performing feature engineering on the real-time health features to remove noise in the real-time health features; acquiring a currently stored historical data learning health mode, and constructing a personalized health model containing personalized health suggestions in combination with the health mode and the real-time health features; and evaluating the personalized health model, judging whether real-time health features needing early warning exist in the personalized health model or not in combination with a preset early warning algorithm, and generating corresponding health early warning information based on a judgment result. The method and the device have the effects of intelligently analyzing health problems and generating health suggestions in combination with multi-source data.

Description

Technical Field

[0001] The present application relates to the technical field of medical health, and in particular to a health data analysis method, system and processor that integrates historical and real-time data. Background Art

[0002] In modern medicine, online medical procedures are constantly developing, allowing patients to gradually shift from offline medical treatment to a model of online consultation plus offline medical treatment. More and more Internet software connecting patients and doctors have emerged.

[0003] Currently, many online medical apps can upload patients' medical records, health information from examination reports, and can also match patients with the department doctors they need. Department doctors can answer patients' questions online. They provide users with online consultations and online diagnoses based on the doctor's subjective judgment and large amounts of medical data on the Internet.

[0004] However, the above-mentioned existing technologies have the following disadvantages: due to the large gap between the number of online doctors and the number of patients, a doctor often needs to face dozens or even hundreds of online medical inquiries. At the same time, online doctors cannot act as health managers for patients. The current online medical program only provides an online communication platform between patients and doctors. It can only give manual replies based on patient descriptions when patients actively ask about their own health problems, and the reply process generally has time differences, making it impossible to monitor the patient's health status in real time and make intelligent diagnosis. In addition, when answering health questions, doctors cannot integrate the patient's multi-source data well and give comprehensive health advice due to time constraints, but can only give advice based on the patient's current description and their own experience, and the intelligent diagnosis effect is poor. Summary of the invention

[0005] In order to achieve intelligent analysis of health problems and generation of health advice by combining multi-source data, the present application provides a health data analysis method, system and processor that integrate historical and real-time data.

[0006] In the first aspect, the present application provides a health data analysis method integrating historical and real-time data, which adopts the following technical solutions: A health data analysis method integrating historical and real-time data comprises the following steps: Acquire one or more different types of real-time data uploaded by the patient, and pre-process the real-time data to obtain real-time health characteristics; Performing multimodal processing on the real-time health features to perform data fusion, data learning and data alignment on a plurality of data; Performing feature engineering on the real-time health feature to remove noise in the real-time health feature, wherein the noise is characterized as short-term fluctuating health data; Acquire currently stored historical data including historical health features and historical suggestions and use it as a model basis, learn a health pattern based on the historical data, and build a personalized health model by combining the health pattern and the real-time health features, wherein the personalized health model includes personalized health suggestions corresponding to the real-time health features; Evaluate the personalized health model, determine whether there are real-time health features that require warning in the personalized health model in combination with a preset warning algorithm, and generate corresponding health warning information based on the judgment result.

[0007] In some embodiments, the real-time data includes text data, language data, image data, sensor data and behavior data, and preprocessing the real-time data to obtain real-time health features includes the following steps: If the real-time data is the voice data, the voice data is recognized to be converted into text data, and the voice data is analyzed to obtain object recognition information, emotion information, and acoustic feature information; If the real-time data is the text data, then obtaining text understanding information based on a natural language processing algorithm, obtaining sentiment information based on sentiment analysis, and obtaining subject and noun information corresponding to the text data based on a recognition algorithm; If the real-time data is the image data, identifying the corresponding object recognition information, and analyzing the facial diagnosis information and the emotion information corresponding to the image data; If the real-time data is the sensor data, then the corresponding heart rate data, blood pressure data, body temperature data, blood sugar information, and sleep information are obtained, and the physiological indicator information corresponding to the sensor data is analyzed, and the physiological indicator information includes change trends, abnormal values, and repeated patterns; If the real-time data is the behavior data, the daily behavior pattern corresponding to the behavior data is analyzed and the correlation information between the behavior data and the health status is mined.

[0008] In some embodiments, multimodal processing is performed on the real-time health features to perform data fusion, data learning and data alignment on a plurality of data, including the following steps: Data fusion: merging several processed output contents of several different types of real-time data that have been independently pre-processed; Data learning: different types of real-time data are used as learners and learnees respectively, and the learnee derives and verifies its own processed output content through the processed output content corresponding to the learner to achieve understanding gain; Data alignment: acquiring time information and / or spatial information of several different types of the real-time data, and synchronizing the several time information and / or spatial information.

[0009] In some embodiments, feature engineering is performed on the real-time health feature to remove noise in the real-time health feature, comprising the following steps: Creating a number of time windows of different lengths, the time windows including short-term time windows and seasonal time windows; Performing trend analysis based on the real-time data to obtain prediction standard trends corresponding to the time windows of different lengths; It is determined whether there is data fluctuation in the time window according to the prediction standard trend. If there is data fluctuation, the real-time data in the time window is treated as noise and removed.

[0010] In some embodiments, building a personalized health model in combination with the health pattern and the real-time health feature includes the following steps: Acquire each item of the historical data, group several items of the historical data having the same historical health characteristics to generate a first training set, and group several items of the historical data having the same historical suggestions to generate a second training set; Bidirectionally training a model through the first training set and the second training set respectively to learn the corresponding health pattern; The real-time health features are put into the model to obtain a plurality of health suggestions, and the plurality of health suggestions are sorted to obtain the personalized health suggestions.

[0011] In some of the embodiments, after constructing the personalized health model, the following steps are further included: Collecting the data fluctuations, and defining the data fluctuations as short-term fluctuations and seasonal fluctuations based on the time window corresponding to the data fluctuations; Classifying the short-term fluctuations based on preset screening features to obtain irrelevant fluctuations and special fluctuations, and screening out the irrelevant fluctuations; Generating a real-time adjustment suggestion for the personalized health model based on the special fluctuation and the seasonal fluctuation, wherein the real-time adjustment suggestion includes characteristic parameters and adjustment contents related to the fluctuation; Based on the real-time adjustment suggestion, corresponding characteristic parameters in the personalized health model are selected and the corresponding characteristic parameters are updated according to the adjustment content.

[0012] In some embodiments, the short-term fluctuations are classified based on preset screening features to obtain irrelevant fluctuations and special fluctuations, including the following steps: Acquire a high-density time period based on each of the time windows corresponding to the short-term fluctuations, and use the high-density time period as the first screening feature; Classify the time type corresponding to the time window into a daily time period and a holiday time period, and use the daily time period as a second screening feature; Acquire a high degree of volatility based on the volatility corresponding to the short-term volatility, and use the high degree of volatility as a third screening feature; The short-term fluctuation is classified based on the first screening feature, the second screening feature, and the third screening feature.

[0013] In some of the embodiments, after generating the personalized health model, the following steps are further included: generating feed-forward information based on the historical data in the personalized health model, wherein the feed-forward information is characterized by health content automatically pushed to the user based on the historical data; generating feedback information based on the real-time data in the personalized health model, wherein the feedback information is characterized as health content pushed to the user through the user's active behavior; generating an incentive mechanism, under which the amount of the feedforward information and the feedback information is determined; If the feedforward information is less than a first preset value, generating a feedback incentive instruction based on the incentive mechanism to obtain the latest real-time data and generate the latest feedback information, and retrieving and generating the corresponding feedforward information based on the real-time health feature in the latest feedback information; If the feedback information is less than a second preset value, a feedforward incentive instruction is generated based on the incentive mechanism to obtain the historical health characteristics corresponding to several of the personalized health recommendations and / or the historical health characteristics that have been warned and integrate them into an expected range, and the feedforward information that meets the expected range is retrieved and generated.

[0014] In a second aspect, the present application provides a health data analysis system integrating historical and real-time data, which adopts the following technical solutions: A health data analysis system integrating historical and real-time data, comprising: A real-time data acquisition module is used to acquire one or more different types of real-time data uploaded by patients; A preprocessing module, used for preprocessing the real-time data to obtain real-time health characteristics; A multimodal processing module, used for performing multimodal processing on the real-time health features to perform data fusion, data learning and data alignment on a plurality of data; a denoising module, configured to perform feature engineering on the real-time health feature to remove noise in the real-time health feature, wherein the noise is characterized as health data with short-term fluctuations; A health data analysis module, used to obtain currently stored historical data containing historical health features and historical suggestions and use it as a model basis, learn a health model based on the historical data, and build a personalized health model in combination with the health model and the real-time health features, wherein the personalized health model contains personalized health suggestions corresponding to the real-time health features; The early warning module is used to evaluate the personalized health model, determine whether there are real-time health features that require early warning in the personalized health model in combination with a preset early warning algorithm, and generate corresponding health early warning information based on the judgment result.

[0015] In a third aspect, the present application provides a processor, which adopts the following technical solution: A processor, the processor is used to run a program, and the program executes the method according to any one of claims 1 to 8 when running.

[0016] The technical solution provided by the embodiments of the present application has the following technical effects: By using historical data as the basis for model training and combining it with corresponding learning algorithms, the personalized health model can obtain the association between health data and health recommendations. By acquiring real-time data and putting it into the personalized health model after preprocessing and multimodal processing to obtain personalized health recommendations, the system can combine historical data and real-time data to achieve a comprehensive diagnosis of health problems and quickly make health recommendations. The health recommendations are made more intelligently and accurately. At the same time, real-time monitoring of users' health data is achieved, and rapid response warnings are provided when health risks arise. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flowchart of the steps of the health data analysis method integrating historical and real-time data provided in an embodiment of the present application.

[0018] Figure 2 This is a schematic diagram of the module connections of a health data analysis system that integrates historical and real-time data provided in an embodiment of the present application. DETAILED DESCRIPTION

[0019] To more clearly understand the purpose, technical solutions and advantages of the present application, the present application is described and illustrated below in conjunction with the accompanying drawings and embodiments. However, it should be understood by those of ordinary skill in the art that the present application can be implemented without these details. In some cases, in order to avoid unnecessary descriptions that make various aspects of the present application obscure, well-known methods, processes, systems, components and / or circuits that have been described at a higher level will not be described in detail. For those of ordinary skill in the art, it is obvious that various changes can be made to the embodiments disclosed in the present application, and without departing from the principles and scope of the present application, the general principles defined in the present application can be applied to other embodiments and application scenarios. Therefore, the present application is not limited to the embodiments shown, but conforms to the broadest scope consistent with the scope claimed for protection of the present application.

[0020] It should be noted that the description of these embodiments is used to help understand the present invention, but does not constitute a limitation of the present invention. In addition, the technical features involved in each embodiment of the present invention described below can be combined with each other as long as there is no conflict between them.

[0021] In the description of this application, "several" means one or more, "more" means more than two, "greater than", "less than", "exceed", etc. are understood to exclude the number itself, and "above", "below", "within", etc. are understood to include the number itself. If there is a description of "first" or "second", it is only used to distinguish the technical features, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features or implicitly indicating the order of the indicated technical features.

[0022] In the description of the present application, the description with reference to the terms "one embodiment", "some embodiments", "illustrative embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials, or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any one or more embodiments or examples.

[0023] like Figure 1 As shown, the embodiment of the present application discloses a health data analysis method integrating historical and real-time data, comprising the following steps: S100, obtaining one or more different types of real-time data uploaded by the patient, and preprocessing the real-time data to obtain real-time health characteristics.

[0024] The real-time data uploaded by patients include active upload and passive upload. Active upload refers to the user manually uploading his or her own health data through mobile terminals, electronic watches, electronic detectors and other devices. Passive upload refers to the uploading of detected health data without the user's subjective instruction when the mobile terminals, electronic watches, electronic detectors and other devices are worn on the user.

[0025] Real-time data refers to the latest, real-time health data. Real-time data is pre-processed to convert health values ​​or health content into corresponding indicators or content that can be quantified in subsequent health data analysis.

[0026] S200, performing multimodal processing on real-time health features to perform data fusion, data learning and data alignment on a plurality of data.

[0027] Due to the different categories of real-time data, in order to convert different types of real-time health features into unified analysis data, and in order to assist each other in data analysis accuracy through certain data associations between different types of real-time health features, it is necessary to perform multimodal processing on different types of real-time health features.

[0028] S300, performing feature engineering on the real-time health features to remove noise in the real-time health features.

[0029] Noise is characterized by short-term fluctuations in health data. This type of data is caused by short-term changes in lifestyle, such as temporary increase in work pressure due to overtime, holiday travel, etc. This type of data often cannot represent a person's long-term daily health status. In order to make subsequent health models applicable to a wider range of daily time, it is necessary to remove these data that will cause parameter deviations in model training and model generation.

[0030] Feature engineering is algorithmic engineering obtained by the framework of one or more algorithms for denoising.

[0031] S400, obtaining the currently stored historical data including historical health characteristics and historical suggestions and using it as a model basis, learning health patterns based on the historical data, and building a personalized health model by combining the health patterns and real-time health characteristics.

[0032] First, historical data is obtained. Historical data is the real-time data that the user has uploaded in the past. These historical data, as effective objective data that has been recorded and used for health advice, can be used as the basis of the model, so that the model can be trained and learned from the user's historical data to obtain the correlation between data and health advice. Combined with the real-time health features received subsequently, a personalized health model for the individual can be constructed. The personalized health model contains personalized health advice corresponding to the real-time health features. Users can obtain personalized health advice for real-time data based on the personalized health model through real-time data.

[0033] S500, evaluating the personalized health model, combining the preset early warning algorithm to determine whether there are real-time health features that require early warning in the personalized health model, and generating corresponding health early warning information based on the determination result.

[0034] The early warning algorithm includes time series analysis and anomaly detection algorithm. The early warning system is developed using long short-term memory networks and autoencoders. Historical data and real-time data are combined to identify potential health risks in advance. When abnormalities in health indicators are identified, early warnings are issued in a timely manner to remind users to undergo health checks and treatments.

[0035] Through the above technical solution, by using historical data as the basis for model training and combining it with the corresponding learning algorithm, the personalized health model can obtain the association between health data and health recommendations. By acquiring real-time data and putting it into the personalized health model after preprocessing and multimodal processing to obtain personalized health recommendations, the system can combine historical data and real-time data to achieve a comprehensive diagnosis of health problems and quickly make health recommendations. The health recommendations are made more intelligently and accurately. At the same time, real-time monitoring of the user's health data is achieved, and a quick response warning is given when there is a risk of health problems.

[0036] In another embodiment, the real-time data includes text data, language data, image data, sensor data and behavior data, and preprocessing the real-time data to obtain real-time health characteristics includes the following steps: S110, if the real-time data is voice data, the voice data is recognized to be converted into text data, and the voice data is analyzed to obtain object recognition information, emotion information and acoustic feature information.

[0037] Voice data includes user voice recordings, voice waveform data, voice spectrograms, etc. Voice is converted into text through voice recognition algorithms. At the same time, speaker recognition is used to identify and verify whether the speaker is the user bound to the system. Emotional features are extracted from voice information through sentiment analysis. Emotional features can effectively reflect the user's current state, such as weakness, energy, doubt, etc. The extracted acoustic feature information includes pitch, volume, speech speed, etc.

[0038] S120, if the real-time data is text data, then obtain text understanding information based on a natural language processing algorithm, obtain sentiment information based on sentiment analysis, and obtain subject and noun information corresponding to the text data based on a recognition algorithm.

[0039] Text data includes text entered by users, text converted from voice, and text such as physical examination reports and medical history reports uploaded by some users.

[0040] The content of the text is understood and analyzed through natural language processing algorithms combined with a keyword library. Sentiment analysis is used to obtain emotional information in the text, such as urgency, doubt, and calmness. Topic modeling is used to identify the main topics in the text, such as the type of illness being asked about, whether it is a consultation or a report. Named entity recognition is used to identify medical terms, symptoms, etc. in the text.

[0041] S130: If the real-time data is image data, the corresponding object recognition information is identified, and the facial diagnosis information and emotion information corresponding to the image data are analyzed.

[0042] Image data includes facial images and skin images taken and uploaded by users.

[0043] The user's identity is identified through facial feature point recognition to determine whether the image corresponds to the user bound to the system. The facial expressions in the image are analyzed by combining facial feature points with the emotion library to obtain emotional information.

[0044] S140, if the real-time data is sensor data, then the corresponding heart rate data, blood pressure data, body temperature data, blood sugar information, and sleep information are obtained, and the physiological indicator information corresponding to the sensor data is analyzed.

[0045] Physiological indicator information includes changing trends, abnormal values, and repetitive patterns.

[0046] Combining several types of sensor data, we use time series analysis algorithms to analyze the short-term and long-term change trends of various types of data, use anomaly detection algorithms to identify outliers in various types of data, and use pattern recognition algorithms to identify repeated patterns in various types of data. Repeated patterns can effectively characterize the credibility of data.

[0047] S150: If the real-time data is behavior data, then the daily behavior pattern corresponding to the behavior data is analyzed and the correlation information between the behavior data and the health status is mined.

[0048] Behavioral data includes the type and duration of a user's behavioral activities, the user's dietary records, and the user's medication use records.

[0049] Behavior pattern analysis algorithms are used to identify users' daily behaviors, such as the frequency of behaviors, time period repetition, etc. Association rule mining is used to discover the relationship between behaviors and health status, such as finding that regular medication is associated with stable health data, regular running is associated with stable health data, and staying up late is associated with unstable health data.

[0050] The processing of real-time data also includes stream data processing technologies such as Apache Kafka or Apache Flink, which are capable of processing high-throughput data streams and supporting real-time data analysis and response, allowing the system to instantly receive and analyze user health data.

[0051] In some other embodiments, multimodal processing is performed on real-time health features to perform data fusion, data learning, and data alignment on a plurality of data, including the following steps: S210, data fusion: merging several processed output contents of several different types of real-time data that are independently pre-processed.

[0052] Multimodal fusion is the integration of health data of different modalities. At this stage, the real-time data of each modality (type) is first pre-processed independently, and then the output of each type of data (such as classification results, feature representation, etc.) is merged in the decision-making stage.

[0053] The advantage of this strategy is that it allows the processing pipeline of each modality to be optimized independently and can more easily handle the heterogeneity of data from different modalities.

[0054] For example, in sentiment analysis, the sentiment information of speech data and the sentiment information of text data may be analyzed independently first, and then the output results of the two may be combined to make a final sentiment judgment.

[0055] S220, data learning: different types of real-time data are used as the learner and the learned party respectively, and the learned party derives and verifies its own processed output content through the corresponding processed output content of the learner to achieve understanding gain.

[0056] Cross-modal learning is the use of information from one modality to enhance the understanding of another modality to achieve a gain in understanding.

[0057] For example, when a user describes symptoms through voice data, the description may be broad or difficult to understand due to weakness or lack of familiarity with medical terminology. In this case, text data such as the user's medical records or test results can be used to improve understanding of the voice data.

[0058] S230, data alignment: obtaining time information and / or spatial information of several different types of real-time data, and synchronizing the several time information and / or spatial information.

[0059] Multimodal alignment ensures that data from different modalities are synchronized in time or space, such as aligning a user's voice data recording with a facial expression video at the same time point to more accurately analyze the user's emotions.

[0060] In some other embodiments, feature engineering is performed on the real-time health features to remove noise in the real-time health features, comprising the following steps: S310, creating several time windows of different time lengths.

[0061] Time windows include short-term time windows and seasonal time windows. Time windows are used to define the duration of analysis. Short-term time windows are generally 12-24 hours long, while seasonal time windows are generally one month long.

[0062] S320, performing trend analysis based on real-time data to obtain prediction standard trends corresponding to time windows of different lengths.

[0063] The time series analysis technique of trend analysis is used to predict and reveal the short-term trend of a data in a short time window through moving average or exponential smoothing methods, and to identify and predict the seasonal fluctuation trend of the data through seasonal decomposition, such as STL decomposition.

[0064] S330, judging whether there is data fluctuation in the time window according to the prediction standard trend, if there is data fluctuation, the real-time data in the time window is treated as noise and removed.

[0065] The calculated prediction standard trend is used to determine whether the data will fluctuate in the future prediction time length, such as changes in health data caused by short-term holidays or work pressure, or fluctuations in certain health data caused by different seasons. If so, such data will be excluded as noise.

[0066] In addition, data noise can be suppressed through anomaly detection methods. The Isolation Forest algorithm is used to construct a random forest to isolate outliers, and the One-Class SVM algorithm is used to identify outliers by maximizing the gap between data points and decision boundaries. The above algorithms can identify and exclude abnormal data points that do not represent the user's actual daily health status.

[0067] Finally, the effectiveness of noise suppression can be improved by integrating the prediction results of multiple models through ensemble learning techniques, such as Bagging and Boosting. This can reduce the bias and noise that may be introduced by a single model. At the same time, regular model evaluation and updating ensure that the model can adapt to new data and changes and maintain its predictive performance.

[0068] In some other embodiments, building a personalized health model by combining health patterns and real-time health features includes the following steps: S410, obtaining various historical data, grouping several historical data with the same historical health characteristics to generate a first training set, and grouping several historical data with the same historical suggestions to generate a second training set.

[0069] Historical data consists of historical health features and corresponding historical recommendations. In order to make the correlation between data features and recommendations more stable and accurate, a two-way training method is used when training the model. First, the historical data is grouped into two groups, with the same historical health features as one group and the same historical recommendations as one group, so that two different numbers of training sets can be obtained.

[0070] S420, bidirectionally training the model through the first training set and the second training set to learn the corresponding health pattern.

[0071] The model is trained using the first training set and the second training set respectively, and the training results are finally merged synchronously, so that the model can achieve dual verification training of verifying suggestions through features and verifying features through suggestions, thereby improving the correlation between features and suggestions.

[0072] S430, putting the real-time health features into the model to obtain a number of health suggestions, and organizing the several health suggestions to obtain personalized health suggestions.

[0073] By putting the acquired real-time health features into the model, corresponding health recommendations can be output, and several health recommendations for different real-time health features can be sorted out to obtain personalized health recommendations.

[0074] In some other embodiments, after building the personalized health model, the following steps are further included: S440, collecting data fluctuations, and defining the data fluctuations as short-term fluctuations and seasonal fluctuations based on a time window corresponding to the data fluctuations.

[0075] When there are data fluctuations, each data fluctuation is collected and defined as short-term fluctuation and seasonal fluctuation according to the length of the time window corresponding to the data fluctuation. Short-term fluctuation corresponds to the fluctuation corresponding to the short time window, while seasonal fluctuation is characterized as the fluctuation corresponding to the seasonal time window.

[0076] S441, classify short-term fluctuations based on preset screening features to obtain irrelevant fluctuations and special fluctuations, and screen out irrelevant fluctuations.

[0077] When real-time data fluctuates, it is characterized by the real-time data being displayed as abnormal data with fluctuations under the current personalized health model. In some cases, the fluctuation of the data may be due to certain changes in the user's living conditions and the changes will appear as long-term changes in the future, or the fluctuation of the data will be maintained for a long time in the future.

[0078] For example, if a user moves to a different city for work or study, the differences in climate, diet, and lifestyle in different cities may cause fluctuations in health data, and this fluctuation may persist for a longer period of time as the user lives there for a longer period of time; or for example, when the season enters autumn and winter, as the temperature drops, the user may suffer from colds, fevers, and other health data fluctuations, which may also persist over a longer season.

[0079] At this time, the health data does have volatility at the current time, but this volatility will remain for a long time in the future. Therefore, in order to prevent long-term health data changes from being repeatedly defined as data fluctuations and frequently screened out due to the non-updating of the health model, resulting in a decrease in the accuracy of health data analysis, short-term fluctuations are further classified into irrelevant fluctuations and special fluctuations.

[0080] Among them, irrelevant fluctuations are characterized by abnormal fluctuations caused by short-term changes determined through analysis of data fluctuations, which have nothing to do with the user's real long-term health status; special fluctuations are characterized by the tendency or possibility that data fluctuations will change into long-term health changes in the future due to analysis of data fluctuations, which requires updates to the current personalized health model.

[0081] The current personalized health model can be updated through special fluctuations, so that the personalized health model can be updated in accordance with the current changes of the user and will become normalized in the future. In this way, the subsequently updated model will be retrained based on the latest historical data to avoid more health data being misjudged as data fluctuations, while irrelevant fluctuations will continue to be screened out.

[0082] S442, generating real-time adjustment suggestions for the personalized health model based on special fluctuations and seasonal fluctuations, wherein the real-time adjustment suggestions include characteristic parameters and adjustment contents related to the fluctuations.

[0083] The personalized health model is updated through fluctuation data such as special fluctuations and seasonal fluctuations that will affect the user's long-term health status in the future. The characteristic parameters and adjustment contents related to the above fluctuations are obtained through the generated real-time adjustment suggestions, and brought into the personalized health model for updating.

[0084] S443, selecting corresponding characteristic parameters in the personalized health model based on the real-time adjustment suggestion and updating the corresponding characteristic parameters according to the adjustment content.

[0085] Combined with online learning or incremental learning algorithms, the system has adaptive learning capabilities. Through dynamically generated real-time adjustment suggestions, when the model receives feature parameters and adjustment content corresponding to new data, it only updates the corresponding original feature parameters, without the need to retrain the entire model.

[0086] In some other embodiments, classifying short-term fluctuations based on preset screening features to obtain irrelevant fluctuations and special fluctuations includes the following steps: S4411, obtaining high-density time periods based on each time window corresponding to the short-term fluctuation, and using the high-density time periods as the first screening feature.

[0087] S4412: Classify the time type corresponding to the time window into a daily time period and a holiday time period, and use the daily time period as the second screening feature.

[0088] S4413, obtaining a high degree of volatility based on the volatility corresponding to the short-term volatility, and using the high degree of volatility as the third screening feature.

[0089] S4414, classify the short-term fluctuations based on the first screening feature, the second screening feature and the third screening feature.

[0090] A series of screening criteria are developed to classify short-term fluctuations.

[0091] First, it is based on the overlapping frequency of time windows. When a user has continuous short-term fluctuations, and the short-term fluctuations that occur every day are concentrated in a certain time length, for example, fluctuations occur between 8 and 9 o'clock every day, then it may be because the user starts exercising during this time period. When the overlapping frequency of time windows is high, the corresponding time period is relatively dense, which is a high-density time period. At this time, the short-term fluctuations corresponding to this time period are defined as special fluctuations.

[0092] Secondly, determine whether the time corresponding to the time window is a holiday period. If it is a holiday period, it means that the changes during this period are due to changes in status during the holiday, such as changes in health data caused by travel and long-term traffic. Because holidays do not belong to daily time, fluctuations in holiday time will not affect the normal personalized health model. At this time, it is considered to be irrelevant fluctuations. If the time is daily time, the short-term fluctuations at this time may be the time of long-term status changes that will occur in the user in the future.

[0093] Finally, it is necessary to determine the volatility corresponding to short-term fluctuations. When the volatility is higher than the preset value, the short-term fluctuation corresponds to a high degree of fluctuation. When the degree of fluctuation of the physical health data is high, it is considered that the instantaneous change of the physical health data is large. If the physical health data is in a state of high change for a long time, it will cause a greater burden on the normal health of the individual. Then, such fluctuations need to be recorded in the personalized health model to prevent the occurrence of health risks. Therefore, short-term fluctuations with high volatility need to be classified as special fluctuations.

[0094] Combined with the above-mentioned screening features, each short-term fluctuation is comprehensively considered to determine the specific type of short-term fluctuation, and the reference data required for updating the personalized health model is obtained through different dimensions.

[0095] In some other embodiments, after generating the personalized health model, the following steps are further included: S450, generating feedforward information based on historical data in the personalized health model.

[0096] Feedforward information is characterized by health content that is automatically pushed to users based on historical data. The system can push health content such as health articles, medical policies, physician lectures, and patient sharing to users based on historical data as the basis of the current personalized health model. Generate user preferences and personalized recommendation services based on historical data.

[0097] S451, generating feedback information based on real-time data in the personalized health model.

[0098] Feedback information refers to health content pushed to users through their active behaviors. For example, after users actively upload their health data or make inquiries, the health content that users want needs to be fed back to users, including but not limited to: diagnosis content, personalized health advice, health articles, medical policies, health warnings, etc. The corresponding services are fed back through the real-time data sent by users.

[0099] Good feedback and feedforward can help the system better understand user needs and encourage users to actively participate in health management. In order to increase users' enthusiasm for using personalized health models to upload data, obtain health advice, and conduct health management, it is also necessary to use incentive mechanisms to positively increase user participation.

[0100] S452, generating an incentive mechanism, and determining the amount of feedforward information and feedback information under the incentive mechanism.

[0101] First, determine the amount of feedforward information and feedback information. Ideally, there needs to be an appropriate amount of feedback information and feedforward information to ensure the user's active use of the system. Good human-computer interaction should have a good mixture of a certain degree of assimilative feedforward and a certain degree of compliant feedback.

[0102] S453, if the feedforward information is less than the first preset value, a feedback incentive instruction is generated based on the incentive mechanism to obtain the latest real-time data and generate the latest feedback information, and the corresponding feedforward information is retrieved and generated based on the real-time health characteristics in the latest feedback information.

[0103] If there is less feedforward information, more targeted feedforward information can be provided by adding feedback information. Specifically, feedback incentive instructions are generated to enable the system to automatically obtain or issue requests to enable users to provide the latest real-time data, so as to analyze the latest problems in the user's current health status based on the latest real-time health characteristics, and obtain more feedforward information in combination with the latest personalized health advice, such as pushing relevant medical articles, physician lectures, etc. This type of information can be retrieved by crawling software on popular medical information websites.

[0104] S454, if the feedback information is less than the second preset value, a feedforward incentive instruction is generated based on the reward mechanism to obtain historical health characteristics corresponding to a number of personalized health suggestions and / or historical health characteristics that have been warned and integrated into an expected range, and the feedforward information that meets the expected range is retrieved and generated.

[0105] If there is less feedback information, users can be guided to take more actions targeting their health status by increasing feedforward information, such as health data query, obtaining personalized health advice, etc. Specifically, feedforward incentives are generated to obtain the latest feedforward information, such as medical articles, physician lectures, patient sharing, etc., and the feedforward information is sent to the user end. When the feedforward information on the user end increases, the user will remember to upload real-time health data or be guided to upload real-time health data to increase the amount of feedback information after seeing a large amount of feedforward information. Through this method, the simple way of issuing prompt information is transformed into a way of guiding users to take feedback actions by increasing the amount of feedback information. Because when simple reminder information increases, users will gradually lose interest in their own health management. However, if the reminder information is converted into feedforward information, the user can subjectively increase his interest in his own health management while learning certain medical and health content.

[0106] At the same time, in order to increase the possibility of users actively participating in their own health management after seeing the added feedforward information, the added feedforward information needs to be consistent with the user's current health problems or health risks corresponding to the warning. Therefore, the generation of feedforward information needs to be based on the historical health characteristics corresponding to the user's past personalized health advice and / or historical health characteristics that have been warned. For example, if a user has pneumonia, if the health content of the added feedforward information is all about fracture rehabilitation, uremia, etc., the user may ignore the feedforward information because it does not meet his or her needs. However, if the pushed feedforward information is all about pneumonia, it will increase the user's vigilance against his or her own health problems, thereby increasing the generation of feedback information.

[0107] The present application also discloses a health data analysis system integrating historical and real-time data, including: The real-time data acquisition module is used to acquire one or more different types of real-time data uploaded by patients.

[0108] The preprocessing module is used to preprocess the real-time data to obtain real-time health features.

[0109] The multimodal processing module is used to perform multimodal processing on real-time health features to perform data fusion, data learning and data alignment on multiple data.

[0110] The denoising module is used to perform feature engineering on the real-time health features to remove noise in the real-time health features, wherein the noise is characterized as short-term fluctuating health data.

[0111] The health data analysis module is used to obtain the currently stored historical data containing historical health characteristics and historical suggestions and use it as the model basis, learn health patterns based on historical data, and build a personalized health model based on health patterns and real-time health characteristics. The personalized health model contains personalized health suggestions corresponding to the real-time health characteristics.

[0112] The early warning module is used to evaluate the personalized health model, combine the preset early warning algorithm to determine whether there are real-time health features in the personalized health model that require early warning, and generate corresponding health warning information based on the judgment results.

[0113] The present application also discloses a processor, which is used to run a program, and executes the above method when the program is running.

[0114] The implementation principle is: By using historical data as the basis for model training and combining it with corresponding learning algorithms, the personalized health model can obtain the association between health data and health recommendations. By acquiring real-time data and putting it into the personalized health model after preprocessing and multimodal processing to obtain personalized health recommendations, the system can combine historical data and real-time data to achieve a comprehensive diagnosis of health problems and quickly make health recommendations. The health recommendations are made more intelligently and accurately. At the same time, real-time monitoring of users' health data is achieved, and rapid response warnings are provided when health risks arise.

[0115] It should be understood that although the steps in the flowchart of the accompanying drawings are shown in sequence according to the instructions of the arrows, these steps are not necessarily performed in the order indicated by the arrows. Unless otherwise clearly stated in this document, the execution of these steps is not strictly limited in order and can be performed in other orders.

[0116] The above are all preferred embodiments of the present application, and the protection scope of the present application is not limited thereto. Therefore, any equivalent changes made according to the structure, shape, and principle of the present application should be included in the protection scope of the present application.

Claims

1. A health data analysis method integrating historical and real-time data, characterized in that: The following steps are involved: Acquire one or more different types of real-time data uploaded by the patient, and pre-process the real-time data to obtain real-time health characteristics; Performing multimodal processing on the real-time health features to perform data fusion, data learning and data alignment on a plurality of data; Performing feature engineering on the real-time health feature to remove noise in the real-time health feature, wherein the noise is characterized as short-term fluctuating health data; Acquire currently stored historical data including historical health features and historical suggestions and use it as a model basis, learn a health pattern based on the historical data, and build a personalized health model by combining the health pattern and the real-time health features, wherein the personalized health model includes personalized health suggestions corresponding to the real-time health features; Evaluate the personalized health model, determine whether there are real-time health features that require warning in the personalized health model in combination with a preset warning algorithm, and generate corresponding health warning information based on the judgment result.

2. The health data analysis method integrating historical and real-time data according to claim 1, characterized in that: The real-time data includes text data, language data, image data, sensor data and behavior data. Preprocessing the real-time data to obtain real-time health features includes the following steps: If the real-time data is the voice data, the voice data is recognized to be converted into text data, and the voice data is analyzed to obtain object recognition information, emotion information, and acoustic feature information; If the real-time data is the text data, then obtaining text understanding information based on a natural language processing algorithm, obtaining sentiment information based on sentiment analysis, and obtaining subject and noun information corresponding to the text data based on a recognition algorithm; If the real-time data is the image data, identifying the corresponding object recognition information, and analyzing the facial diagnosis information and the emotion information corresponding to the image data; If the real-time data is the sensor data, then the corresponding heart rate data, blood pressure data, body temperature data, blood sugar information, and sleep information are obtained, and the physiological indicator information corresponding to the sensor data is analyzed, and the physiological indicator information includes change trends, abnormal values, and repeated patterns; If the real-time data is the behavior data, the daily behavior pattern corresponding to the behavior data is analyzed and the correlation information between the behavior data and the health status is mined.

3. The health data analysis method integrating historical and real-time data according to claim 1, characterized in that: The real-time health features are subjected to multimodal processing to perform data fusion, data learning and data alignment on a plurality of data, including the following steps: Data fusion: merging several processed output contents of several different types of real-time data that have been independently pre-processed; Data learning: different types of real-time data are used as learners and learnees respectively, and the learnee derives and verifies its own processed output content through the processed output content corresponding to the learner to achieve understanding gain; Data alignment: acquiring time information and / or spatial information of several different types of the real-time data, and synchronizing the several time information and / or spatial information.

4. The health data analysis method integrating historical and real-time data according to claim 1, characterized in that: Performing feature engineering on the real-time health feature to remove noise in the real-time health feature comprises the following steps: Creating a number of time windows of different lengths, the time windows including short-term time windows and seasonal time windows; Performing trend analysis based on the real-time data to obtain prediction standard trends corresponding to the time windows of different lengths; It is determined whether there is data fluctuation in the time window according to the prediction standard trend. If there is data fluctuation, the real-time data in the time window is treated as noise and removed.

5. The health data analysis method integrating historical and real-time data according to claim 1, characterized in that: Combining the health mode and the real-time health features to build a personalized health model includes the following steps: Acquire each item of the historical data, group several items of the historical data having the same historical health characteristics to generate a first training set, and group several items of the historical data having the same historical suggestions to generate a second training set; Bidirectionally training a model through the first training set and the second training set respectively to learn the corresponding health pattern; The real-time health features are put into the model to obtain a plurality of health suggestions, and the plurality of health suggestions are sorted to obtain the personalized health suggestions.

6. The health data analysis method integrating historical and real-time data according to claim 4, characterized in that: After constructing the personalized health model, the following steps are also included: Collecting the data fluctuations, and defining the data fluctuations as short-term fluctuations and seasonal fluctuations based on the time window corresponding to the data fluctuations; Classifying the short-term fluctuations based on preset screening features to obtain irrelevant fluctuations and special fluctuations, and screening out the irrelevant fluctuations; Generating a real-time adjustment suggestion for the personalized health model based on the special fluctuation and the seasonal fluctuation, wherein the real-time adjustment suggestion includes characteristic parameters and adjustment contents related to the fluctuation; Based on the real-time adjustment suggestion, corresponding characteristic parameters in the personalized health model are selected and the corresponding characteristic parameters are updated according to the adjustment content.

7. The health data analysis method integrating historical and real-time data according to claim 6, characterized in that: Classifying the short-term fluctuations based on preset screening features to obtain irrelevant fluctuations and special fluctuations includes the following steps: Acquire a high-density time period based on each of the time windows corresponding to the short-term fluctuations, and use the high-density time period as the first screening feature; Classify the time type corresponding to the time window into a daily time period and a holiday time period, and use the daily time period as a second screening feature; Acquire a high degree of volatility based on the volatility corresponding to the short-term volatility, and use the high degree of volatility as a third screening feature; The short-term fluctuation is classified based on the first screening feature, the second screening feature, and the third screening feature.

8. The health data analysis method integrating historical and real-time data according to claim 1, characterized in that: After the personalized health model is generated, the following steps are also included: generating feed-forward information based on the historical data in the personalized health model, wherein the feed-forward information is characterized by health content automatically pushed to the user based on the historical data; generating feedback information based on the real-time data in the personalized health model, wherein the feedback information is characterized as health content pushed to the user through the user's active behavior; generating an incentive mechanism, under which the amount of the feedforward information and the feedback information is determined; If the feedforward information is less than a first preset value, generating a feedback incentive instruction based on the incentive mechanism to obtain the latest real-time data and generate the latest feedback information, and retrieving and generating the corresponding feedforward information based on the real-time health feature in the latest feedback information; If the feedback information is less than a second preset value, a feedforward incentive instruction is generated based on the incentive mechanism to obtain the historical health characteristics corresponding to several of the personalized health recommendations and / or the historical health characteristics that have been warned and integrate them into an expected range, and the feedforward information that meets the expected range is retrieved and generated.

9. A health data analysis system integrating historical and real-time data, characterized in that: include: A real-time data acquisition module is used to acquire one or more different types of real-time data uploaded by patients; A preprocessing module, used for preprocessing the real-time data to obtain real-time health characteristics; A multimodal processing module, used for performing multimodal processing on the real-time health features to perform data fusion, data learning and data alignment on a plurality of data; a denoising module, configured to perform feature engineering on the real-time health feature to remove noise in the real-time health feature, wherein the noise is characterized as health data with short-term fluctuations; A health data analysis module, configured to obtain currently stored historical data containing historical health features and historical recommendations and use it as a model basis, learn a health model based on the historical data, and build a personalized health model in combination with the health model and the real-time health features, wherein the personalized health model includes personalized health recommendations corresponding to the real-time health features; The early warning module is used to evaluate the personalized health model, determine whether there are real-time health features that require early warning in the personalized health model in combination with a preset early warning algorithm, and generate corresponding health early warning information based on the judgment result.

10. A processor, characterized in that: The processor is used to run a program, and the program executes the method according to any one of claims 1 to 8 when running.

Images