Disease risk early warning system and method based on digital behavior analysis

Abstract

The invention provides a disease risk early warning system and method based on digital behavior analysis, and relates to the technical field of disease risk early warning, the system comprises a data acquisition layer, a feature extraction layer, a risk analysis layer and an early warning output layer, the method comprises the following steps: collecting browsing behavior data, mouse behavior data, search behavior data and time pattern data when a user browses a page, extracting time dimension characteristics, content dimension characteristics and behavior dimension characteristics when the user browses the page, and normalizing the time dimension characteristics, the content dimension characteristics and the behavior dimension characteristics respectively; obtaining a time characteristic value, a content characteristic value and a behavior characteristic value, and inputting the time characteristic value, the content characteristic value and the behavior characteristic value into a preset risk assessment model to obtain a risk score; and finally determining disease risk early warning information of the user according to the risk score. According to the invention, continuous monitoring and early warning of the potential health risk of the user can be realized, and timeliness and accuracy of health risk early warning are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of disease risk early warning, and in particular to a disease risk early warning system and method based on digital behavior analysis. Background Art

[0002] In recent years, with the rapid development of Internet technology, people's daily life and work are increasingly dependent on the Internet, thus generating a large amount of digital behavior data. The digital behavior characteristics corresponding to these digital behavior data can often reflect the user's physical and mental health status, providing a new research direction for health risk warning.

[0003] At present, early warning of user health risks mainly relies on regular physical examinations, wearable device monitoring and other methods. These methods require users to actively cooperate or wear special detection equipment, which are inconvenient to use and poor compliance. Moreover, traditional health monitoring methods can only obtain intermittent data, and cannot achieve continuous and real-time monitoring of user health status, and are prone to miss important health risk signals. At the same time, the existing health risk warning system is mainly based on physiological indicators for analysis, and fails to make full use of users' daily digital behavior characteristics, but these digital footprints often contain important health risk information. Summary of the invention

[0004] The purpose of the present invention is to provide a disease risk warning system and method based on digital behavior analysis, which is used to solve the problems existing in the health risk warning methods in the prior art, such as the need for active cooperation of users, discontinuous data collection, and untimely warnings, so as to realize continuous monitoring and early warning of potential health risks of users, without the need for users to wear any equipment or change their daily behavior habits, and can improve the timeliness and accuracy of health risk warnings.

[0005] In order to achieve the above-mentioned objectives, in a first aspect, the present invention provides a disease risk warning system based on digital behavior analysis, comprising a data collection layer, a feature extraction layer, a risk analysis layer and a warning output layer, wherein the data collection layer is used to collect browsing behavior data, mouse behavior data, search behavior data and time pattern data when a user browses a page; the feature extraction layer is used to extract the time dimension characteristics, content dimension characteristics and behavior dimension characteristics when the user browses the page according to the browsing behavior data, mouse behavior data, search behavior data and time pattern data; the risk analysis layer is used to normalize the time dimension characteristics, content dimension characteristics and behavior dimension characteristics respectively to obtain time feature values, content feature values ​​and behavior feature values, and then input the time feature values, content feature values ​​and behavior feature values ​​into a preset risk assessment model to obtain a risk score; the warning output layer is used to determine the user's disease risk warning information according to the risk score.

[0006] According to a disease risk early warning system based on digital behavior analysis provided by the present invention, browsing behavior data includes URL address, page title, access timestamp, source page, dwell time, page content update and HTTP request response; mouse behavior data includes mouse movement trajectory, mouse movement speed, mouse clicks and mouse scrolling; search behavior data includes search keywords, search time, search frequency and search result clicks; time pattern data includes time distribution of visited pages.

[0007] A data collection layer provided according to the present invention is specifically used for: collecting and recording URL addresses, page titles, access timestamps, source pages, and dwell time through browser plug-ins, and obtaining page content updates and HTTP request responses through DOM change monitoring.

[0008] According to a disease risk warning system based on digital behavior analysis provided by the present invention, the time dimension characteristics include the time distribution pattern of visiting pages; the content dimension characteristics include the document health relevance of the page; and the behavior dimension characteristics include the switching behavior and attention distribution of visiting pages.

[0009] According to a disease risk warning system based on digital behavior analysis provided by the present invention, the feature extraction layer is specifically used to calculate the document health relevance of the page based on the deep semantic analysis method.

[0010] According to a disease risk early warning system based on digital behavior analysis provided by the present invention, the calculation formula of the document health relevance of a page is: H(d)=∑(wi·sim(ti,M)) Among them, sim(ti,M)=max{cos(v_ti,v_m)|m∈M}, wi=tf(ti)·log(N / df(ti)); Where H(d) is the health relevance of the document, wi is the weight, ti is the keyword item in the document, M is the medical dictionary set, m is the medical concept, v_ti is the word vector representation of the keyword item, v_m is the word vector representation of the medical concept, sim(ti,M) is the semantic similarity between the keyword item and the medical concept, tf(ti) is the frequency of the keyword item, df(ti) is the number of documents containing the keyword item, and N is the total number of documents.

[0011] According to a disease risk warning system based on digital behavior analysis provided by the present invention, the risk analysis layer is specifically used for: when the time distribution pattern of users' access to pages is manifested as the more frequent the users' access, the larger the time feature value output by the risk assessment model; when the document health relevance of the user's accessed page is higher, the larger the content feature value output by the risk assessment model; when the user switches between accessed pages more frequently and the user's attention is more focused, the larger the behavior feature value output by the risk assessment model.

[0012] According to a disease risk early warning system based on digital behavior analysis provided by the present invention, the calculation formula of the risk score is: Risk_base = w1×F1+w2×F2+w3×F3 In the formula, Risk_base is the risk score, w1 is the time feature weight, F1 is the time feature value, w2 is the content feature weight, F2 is the content feature value, w3 is the behavior feature weight, and F3 is the behavior feature value.

[0013] According to a disease risk warning system based on digital behavior analysis provided by the present invention, the disease risk warning information includes the disease risk level; the warning output layer is specifically used for: when the risk score is larger, the output disease risk level is higher, indicating that the user is more likely to suffer from the disease.

[0014] In a second aspect, the present invention provides a disease risk early warning method based on digital behavior analysis, using the disease risk early warning system based on digital behavior analysis of the first aspect, the method comprising: Step 1: The data collection layer collects browsing behavior data, mouse behavior data, search behavior data and time pattern data when users browse pages; Step 2: The feature extraction layer extracts the time dimension features, content dimension features and behavior dimension features of the user when browsing the page based on the browsing behavior data, mouse behavior data, search behavior data and time pattern data; Step 3: The risk analysis layer normalizes the time dimension features, content dimension features, and behavior dimension features respectively to obtain the time feature value, content feature value, and behavior feature value, and then inputs the time feature value, content feature value, and behavior feature value into the preset risk assessment model to obtain the risk score; Step 4: The warning output layer determines the user's disease risk warning information based on the risk score.

[0015] The technical solution of the present invention has at least the following technical effects: The present invention provides a disease risk early warning system and method based on digital behavior analysis, which relates to the field of disease risk early warning technology. The system includes a data collection layer, a feature extraction layer, a risk analysis layer and an early warning output layer. By collecting browsing behavior data, mouse behavior data, search behavior data and time pattern data when users browse pages, the time dimension features, content dimension features and behavior dimension features when users browse pages are extracted, and the time dimension features, content dimension features and behavior dimension features are normalized respectively to obtain time feature values, content feature values ​​and behavior feature values, and then the time feature values, content feature values ​​and behavior feature values ​​are input into a preset risk assessment model to obtain a risk score; finally, the disease risk early warning information of the user is determined according to the risk score. The present invention can realize continuous monitoring and early warning of potential health risks of users without the need for users to wear any equipment or change their daily behavior habits, thereby improving the timeliness and accuracy of health risk early warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0017] In the attached picture: Figure 1 It is a structural block diagram of the disease risk early warning system based on digital behavior analysis of the present invention; Figure 2 An early warning curve diagram for case verification of the present invention; Figure 3 This is a warning curve diagram for group verification of the present invention. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] Some embodiments of the present invention will be described in detail below in conjunction with the accompanying drawings. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0020] See also Figure 1An embodiment of the present invention provides a disease risk warning system based on digital behavior analysis, including a data collection layer, a feature extraction layer, a risk analysis layer and a warning output layer, wherein the data collection layer is used to collect browsing behavior data, mouse behavior data, search behavior data and time pattern data when a user browses a page; the feature extraction layer is used to extract the time dimension features, content dimension features and behavior dimension features when the user browses the page according to the browsing behavior data, mouse behavior data, search behavior data and time pattern data; the risk analysis layer is used to normalize the time dimension features, content dimension features and behavior dimension features respectively to obtain time feature values, content feature values ​​and behavior feature values, and then input the time feature values, content feature values ​​and behavior feature values ​​into a preset risk assessment model to obtain a risk score; the warning output layer is used to determine the disease risk warning information of the user according to the risk score.

[0021] Specifically, the data collection layer collects browsing behavior data of users when browsing pages based on a multi-source data collection mechanism. Browsing behavior data includes URL address, page title, access timestamp, source page, stay time, page content update, and HTTP request response.

[0022] The data collection layer collects the URL address, page title, access timestamp, source page, duration of stay, etc. of the user's browsing page through the browser plug-in. The browser plug-in adopts an event-driven collection architecture, and realizes continuous tracking of user behavior data by registering page life cycle event listeners. The data collection layer adopts a hierarchical data caching mechanism, establishes a lightweight database locally, and performs preliminary storage and preprocessing of the collected browsing behavior data. The data cache adopts a three-level architecture: memory cache is used to store real-time data, local storage is used to save short-term historical data, and cloud storage is used to maintain long-term data archives.

[0023] The data collection layer also records the user's browsing behavior data through browser plug-ins to build a complete page access archive, which not only records the basic URL address, page title, access timestamp, source page, length of stay and other metadata. For dynamically loaded content, it also obtains page content updates and HTTP request responses through DOM change monitoring to achieve real-time tracking, and records the time point and specific changes of each page content update. By analyzing the HTTP request header and response header information, it is possible to identify the page's cache status, redirection history, and server response characteristics, thereby building a complete page access link map.

[0024] The data collection layer also has an adaptive mouse behavior data collection mechanism to collect mouse movement trajectory, mouse movement speed, and interactive events such as mouse clicks and mouse scrolls. In terms of trajectory sampling, the data collection layer adopts a state-based multi-level sampling strategy to divide the mouse movement speed into five levels: static state (0~50 pixels / second), fine operation state (50~200 pixels / second), normal movement state (200~800 pixels / second), fast movement state (800~2000 pixels / second) and extreme movement state (greater than 2000 pixels / second). These speed thresholds are natural groupings obtained by clustering analysis of the mouse operation data of 10,000 users. The corresponding sampling intervals are: 2 milliseconds, 5 milliseconds, 10 milliseconds, 20 milliseconds and 40 milliseconds. The data collection layer predicts the mouse movement state in real time through the Kalman filter, adjusts the sampling parameters in advance, and ensures a balance between sampling accuracy and system performance.

[0025] Specifically, search behavior data includes search keywords, search time, search frequency, and search result clicks; time pattern data includes the time distribution of page visits.

[0026] The feature extraction layer is based on a multi-level page activity calculation method. It monitors the foreground / background status of the page through the page visibility API. Combined with the actual user interaction events, the feature extraction layer can accurately distinguish between effective browsing time and background stay time, and then extract time dimension features, including the time distribution law of page access. In the interaction event analysis, the feature extraction layer adopts an adaptive interaction threshold adjustment mechanism to balance the sensitivity and stability of detection. The initial value of the interaction threshold is set to 2 seconds. This initial value is the median obtained by analyzing large-scale user behavior data, which can better balance the sensitivity and stability of interaction detection. The feature extraction layer dynamically adjusts the interaction threshold by analyzing the user's operation frequency distribution in real time. When it is detected that the user is in a high-frequency interaction state (for example, the number of interactions per second exceeds 3 times), the interaction threshold will be reduced to 1 second; in a low-frequency interaction state (for example, the number of interactions per second is less than 0.5 times), the interaction threshold will be increased to 3 seconds. This adaptive mechanism has shown excellent performance in the experiment and can accurately capture the interaction patterns of different users. At the same time, the feature extraction layer establishes an interaction intensity calculation model based on a sliding time window. The size of the time window is automatically adjusted according to the user's operation density. For example, it can dynamically change between 10 seconds and 60 seconds.

[0027] In the session analysis mechanism, the feature extraction layer achieves accurate characterization of user multitasking characteristics through fine-grained behavior tracking. Based on the adaptive session segmentation algorithm, the feature extraction layer dynamically adjusts the session time window to adapt to the browsing habits of different users, thereby extracting the user's behavioral dimension characteristics, such as attention distribution, switching behavior of accessed pages, etc. The initial value of the session time window is set to 30 minutes. This initial value is determined based on large-scale user behavior research and covers 85% of common browsing session patterns. The feature extraction layer dynamically adjusts the size of the session time window by analyzing the user's tab switching behavior pattern. When the user shows a dense page switching pattern (for example, the average switching interval is less than 1 minute), the session time window will be reduced to 15 minutes; when the user shows a long-term focus mode (for example, the average switching interval is greater than 5 minutes), the session time window will be expanded to 45 minutes. This dynamic adjustment mechanism significantly improves the accuracy of session recognition in experiments, from 78% of the traditional fixed session time window to 91%.

[0028] In the document health content analysis, the feature extraction layer calculates the document health relevance of the page through deep semantic analysis, that is, extracts content dimension features. For document d, its health relevance H(d) is calculated by the following formula: H(d)=∑(wi·sim(ti,M)) Among them, ti is the keyword item in the document, wi is the weight, m ​​is the medical concept, M is the medical dictionary set, sim(ti,M) is the semantic similarity between the keyword item and the medical concept, and the calculation formula is: sim(ti,M)=max{cos(v_ti,v_m)|m∈M} Among them, v_ti and v_m are word vector representations of key words and medical concepts respectively. Through the word vector representation of v_ti and v_m, a semantic bridge between ordinary text and professional medical concepts is established. This method breaks through the traditional keyword matching limitations and can identify potential health risk-related content.

[0029] The weight wi is calculated by the improved TF-IDF method: wi = tf(ti)·log(N / df(ti)) Among them, tf(ti) is the frequency of the keyword term, df(ti) is the number of documents containing the keyword term, and N is the total number of documents. This improved weight calculation method shows better results than the traditional TF-IDF in medical text analysis, and the accuracy on the test set is improved by 15%.

[0030] Specifically, the risk analysis layer is used for: when the time distribution law of user access to the page is that the more frequently the user accesses, the larger the time feature value output by the risk assessment model; when the document health relevance of the user accessing the page is higher, the content feature value output by the risk assessment model is larger; when the user switches the access page more frequently and the more focused the attention is, the larger the behavior feature value output by the risk assessment model is. For example, when the user frequently accesses the page, browses the relevant page in the middle of the night or at a short interval, that is, the time distribution law of the user's access to the page is that the user visits frequently, and the content of the document in the page is highly related to "cold", indicating that the user is more concerned about the health of the body and frequently visits the web page due to worry. And the more frequently the page is switched and the more focused the attention is, the more urgent the attention is and the unwillingness to miss any page information. This situation can predict that the user may have symptoms related to "cold" and the risk level of catching a cold is high.

[0031] Specifically, the risk score is calculated as follows: Risk_base = w1×F1+w2×F2+w3×F3 In the formula, Risk_base is the risk score, w1 is the time feature weight, F1 is the time feature value, w2 is the content feature weight, F2 is the content feature value, w3 is the behavior feature weight, and F3 is the behavior feature value. The initial configuration of the weight is determined based on expert knowledge and large-scale data analysis, with the content feature weight being 0.4, the time feature weight being 0.3, and the behavior feature weight being 0.3. Furthermore, considering the impact of timing factors, a timing adjustment term is introduced into the risk score calculation formula, and the final risk score is: Risk_final=Risk_base·(1+β·T_factor) Where β is the timing adjustment coefficient and T_factor is the timing adjustment item.

[0032] Preferably, β=0.3. In a long-term tracking experiment of 1,000 users, this value amplifies abnormal behavior signals while maintaining a low false alarm rate (less than 5%).

[0033] Disease risk warning information includes disease risk levels, which can be divided into high, medium and low levels, corresponding to a certain range of risk scores. And the warning output layer is specifically used for: when the risk score is larger, the output disease risk level is higher, indicating that the user is more likely to suffer from the disease.

[0034] Based on the same inventive concept, another embodiment of the present invention provides a disease risk early warning method based on digital behavior analysis, using the disease risk early warning system based on digital behavior analysis of the above embodiment, the method includes: Step 1: The data collection layer collects browsing behavior data, mouse behavior data, search behavior data and time pattern data when users browse pages; Step 2: The feature extraction layer extracts the time dimension features, content dimension features and behavior dimension features of the user when browsing the page based on the browsing behavior data, mouse behavior data, search behavior data and time pattern data; Step 3: The risk analysis layer normalizes the time dimension features, content dimension features, and behavior dimension features respectively to obtain the time feature value, content feature value, and behavior feature value, and then inputs the time feature value, content feature value, and behavior feature value into the preset risk assessment model to obtain the risk score; Step 4: The warning output layer determines the user's disease risk warning information based on the risk score.

[0035] The disease risk early warning system and method based on digital behavior analysis provided by the present invention have been verified through large-scale experiments and have demonstrated excellent early warning performance: the overall prediction accuracy is approximately 55%, obvious abnormalities are approximately 65%, and minor abnormalities are 40%.

[0036] In summary, the disease risk warning system and method based on user digital behavior analysis proposed in the present invention have significant technical advantages and practical application effects. In terms of timeliness, by analyzing the digital behavior characteristics of users, a warning can be issued an average of 10 days before the actual symptoms appear. Experimental data show that the predicted risk value curve shows an upward trend before the actual risk value rises significantly. This warning time is 3 to 5 days earlier than the traditional method based on physiological indicator monitoring. This early warning capability provides users and medical institutions with sufficient time windows for preventive intervention, effectively reducing the risk of disease development.

[0037] In terms of early warning accuracy, through the continuous optimization of the deep learning model, significant performance improvements have been achieved. As shown in the experimental results, the system's early warning accuracy has gradually increased from the initial 75% to 87%, and the verification accuracy has increased from 72% to 85%. This high accuracy greatly reduces the false alarm rate, avoids unnecessary waste of medical resources, and ensures the reliability of early warnings.

[0038] Compared with traditional health monitoring methods, the present invention adopts a non-invasive monitoring method, which does not require users to wear any equipment or change their daily living habits. Health risk warning is achieved by analyzing the user's daily digital behavior, and the acquired data has the characteristics of good continuity and high authenticity. This method not only improves the user's willingness to use, but also overcomes the problems of large sampling intervals and discontinuous data in traditional methods.

[0039] The present invention has personalized early warning capabilities. By establishing a behavioral baseline for individual users, the early warning threshold can be dynamically adjusted according to the individual characteristics of the user. Experimental data show that this personalized early warning mechanism reduces the false alarm rate by 30% and the missed alarm rate by 25% compared with the traditional fixed threshold method. In particular, it maintains a stable early warning effect among user groups of different age groups, verifying its good adaptability.

[0040] The present invention also has wide applicability. In the test of user groups aged 20 to 50, it showed a stable early warning effect. At the same time, the design architecture supports the expansion of early warning models for a variety of common diseases and can be easily integrated with existing medical systems. This scalability and compatibility lays the foundation for the promotion and application of the system.

[0041] In order to verify the practicality and effectiveness of the present invention, the present invention conducted a 3-month experimental study from April to July 2024. After obtaining informed consent, a 25-year-old male volunteer A was selected as a typical case for in-depth analysis, and a group verification was carried out on an experimental group of 50 people.

[0042] In the case study, volunteer A developed cold symptoms during the experiment. By analyzing his daily digital behavior, early warning was successfully achieved. Figure 2 As shown in the figure, 9 days before the onset of symptoms, the user's risk score began to rise significantly. Specifically, it was manifested in increased nighttime activities and frequent searches and browsing of cold-related content. When the risk score exceeded the warning threshold of 0.7, the system issued a health risk warning in a timely manner. Subsequent tracking confirmed that the warning was highly consistent with the actual time of symptom onset, reflecting the system's warning ability.

[0043] In the group validation study, the present invention selected 50 volunteers aged between 20 and 50 for follow-up study. Figure 3 As shown in the figure, the system has a stable early warning effect in different age groups, and the overall accuracy rate remains at around 85%. During the three-month experiment, a total of 35 health events were recorded, and the system successfully warned 30 of them, with only 5 missed reports and 3 false alarms. It is particularly noteworthy that the system's average advance warning time reached 10.5 days, providing a sufficient time window for timely intervention.

[0044] Experimental data show that the present invention has good early warning performance and practical value. While ensuring a high accuracy rate (85.3%), it also achieved a high recall rate (82.7%). Statistical analysis shows that users of different age groups have good feedback on the system and believe that the early warning information provided by the system is of practical help in their health management. The average response time of the system is controlled within 1 second, and the stability reaches 99.9%, which fully meets the actual application needs.

[0045] This embodiment fully verifies the feasibility and effectiveness of the present invention in early warning of disease risks through specific experimental cases and data analysis. The experimental results show that the present invention can timely identify potential health risks by analyzing the digital behavior characteristics of users, providing a scientific basis for preventive health intervention.

[0046] Those skilled in the art will readily appreciate other embodiments of the present invention after considering the specification and practicing the embodiments disclosed herein. The present invention is intended to cover any variations, uses or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary techniques in the art that are not disclosed by the present invention. It should be understood that the present invention is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims

1. A disease risk early warning system based on digital behavior analysis, characterized in that: It includes a data collection layer, a feature extraction layer, a risk analysis layer and an early warning output layer. The data collection layer is used to collect browsing behavior data, mouse behavior data, search behavior data and time pattern data when users browse pages; The feature extraction layer is used to extract the time dimension features, content dimension features and behavior dimension features when the user browses the page according to the browsing behavior data, mouse behavior data, search behavior data and time pattern data; the risk analysis layer is used to normalize the time dimension features, content dimension features and behavior dimension features respectively to obtain the time feature value, content feature value and behavior feature value, and then input the time feature value, content feature value and behavior feature value into the preset risk assessment model to obtain the risk score; The warning output layer is used to determine the disease risk warning information of the user according to the risk score.

2. The disease risk early warning system based on digital behavior analysis according to claim 1 is characterized in that: The browsing behavior data includes URL address, page title, access timestamp, source page, dwell time, page content update and HTTP request response; the mouse behavior data includes mouse movement track, mouse movement speed, mouse clicks and mouse scrolling; the search behavior data includes search keywords, search time, search frequency and search result clicks; the time pattern data includes the time distribution of visited pages.

3. The disease risk early warning system based on digital behavior analysis according to claim 2 is characterized in that: The data collection layer is specifically used to collect and record the URL address, page title, access timestamp, source page, and dwell time through a browser plug-in, and obtain page content updates and HTTP request responses through DOM change monitoring.

4. The disease risk early warning system based on digital behavior analysis according to claim 2 is characterized in that: The time dimension features include the time distribution pattern of page visits; the content dimension features include the document health relevance of the page; and the behavior dimension features include the switching behavior and attention distribution of page visits.

5. The disease risk early warning system based on digital behavior analysis according to claim 4 is characterized in that: The feature extraction layer is specifically used to calculate the document health relevance of the page based on the deep semantic analysis method.

6. The disease risk early warning system based on digital behavior analysis according to claim 5 is characterized in that: The calculation formula for the document health relevance of the page is: H(d)=∑(wi·sim(ti,M)) Among them, sim(ti,M)=max{cos(v_ti,v_m)|m∈M}, wi=tf(ti)·log(N / df(ti)); Where H(d) is the health relevance of the document, wi is the weight, ti is the keyword item in the document, M is the medical dictionary set, m is the medical concept, v_ti is the word vector representation of the keyword item, v_m is the word vector representation of the medical concept, sim(ti,M) is the semantic similarity between the keyword item and the medical concept, tf(ti) is the frequency of the keyword item, df(ti) is the number of documents containing the keyword item, and N is the total number of documents.

7. The disease risk early warning system based on digital behavior analysis according to claim 6 is characterized in that: The risk analysis layer is specifically used for: when the time distribution law of user access to the page is that the more frequently the user accesses, the larger the time feature value output by the risk assessment model; when the document health relevance of the user accessing the page is higher, the content feature value output by the risk assessment model is larger; The more frequently the user switches access pages and the more focused his attention is, the greater the behavior feature value output by the risk assessment model will be.

8. The disease risk early warning system based on digital behavior analysis according to claim 7 is characterized in that: The risk score is calculated as follows: Risk_base = w1×F1+w2×F2+w3×F3 In the formula, Risk_base is the risk score, w1 is the time feature weight, F1 is the time feature value, w2 is the content feature weight, F2 is the content feature value, w3 is the behavior feature weight, and F3 is the behavior feature value.

9. The disease risk early warning system based on digital behavior analysis according to claim 8 is characterized in that: The disease risk warning information includes the disease risk level; the warning output layer is specifically used for: when the risk score is larger, the output disease risk level is higher, indicating that the possibility of the user suffering from the disease is greater.

10. A disease risk early warning method based on digital behavior analysis, characterized in that: Using the disease risk early warning system based on digital behavior analysis as described in any one of claims 1 to 9, the method comprises: Step 1: The data collection layer collects browsing behavior data, mouse behavior data, search behavior data and time pattern data when users browse pages; Step 2: The feature extraction layer extracts the time dimension features, content dimension features and behavior dimension features of the user when browsing the page according to the browsing behavior data, mouse behavior data, search behavior data and time pattern data; Step 3: The risk analysis layer normalizes the time dimension features, content dimension features, and behavior dimension features respectively to obtain a time feature value, a content feature value, and a behavior feature value, and then inputs the time feature value, content feature value, and behavior feature value into a preset risk assessment model to obtain a risk score; Step 4: The warning output layer determines the user's disease risk warning information based on the risk score.

Images