Data desensitization processing method, device, terminal and equipment

Abstract

The application relates to the technical field of data desensitization, in particular to a data desensitization processing method and device, a terminal and equipment. The method comprises the following steps: obtaining, at a local intelligent terminal, biological data on a wearable side generated by a wearable device and touch side interaction data generated by the intelligent terminal; obtaining multi-source heterogeneous feature data; performing irreversible feature projection to obtain a continuous emotional state vector; when the continuous emotional state vector is used for local processing of the intelligent terminal, the continuous emotional state vector is provided to a local downstream processing module; when the continuous emotional state vector is used for network synchronization, the continuous emotional state vector is subjected to discrete quantization processing to generate a discrete state feature code, and the discrete state feature code is subjected to network synchronization. The application can solve the technical problems that in the prior art, original sensitive data is easily exposed in the correlation processing of multi-source physiological data and interaction data, and effective state features are easily lost in traditional desensitization processing.

Description

Technical Field

[0001] This application relates to the field of data anonymization technology, and in particular to a data anonymization processing method, apparatus, terminal and equipment. Background Technology

[0002] With the development of smart wearable devices, smart terminals, and digital health technologies, physiological data such as heart rate, heart rate variability, exercise status, and sleep status are obtained through wearable devices such as smartwatches and wristbands. Combined with interactive data such as touch, swipe, and operation pauses generated by terminals such as smartphones, the analysis of users' physical and mental state is gradually being applied to scenarios such as digital health, human-computer interaction, and proactive psychological intervention.

[0003] To improve the ability to identify and assess users' physical and mental states, it is usually necessary to comprehensively utilize data from multiple data sources. For example, wearable devices can continuously generate physiological data reflecting users' autonomic nervous activity and physical state, while smart terminals can generate interactive data reflecting users' operational rhythm and behavioral changes. Correlation analysis of data from different sources is beneficial for characterizing users' actual state from multiple dimensions.

[0004] However, data itself possesses strong privacy attributes. For example, raw heart rate values ​​and exercise data generated by wearable devices can reflect a user's specific physiological state; absolute touch coordinates, input information, and application operation information generated by smart terminals may reflect the user's input content, operation trajectory, and application usage. When this data is stored for a long period or transmitted to external servers for processing, it easily increases the risk of leakage of user physiological information and terminal operation information.

[0005] Existing technologies typically employ data encryption, data truncation, or perturbation of the original data to mitigate the risk of data leakage. Encryption can reduce the risk of data being directly accessed during transmission; however, subsequent data processing usually requires restoring the encrypted data to its original form, allowing external processing nodes to still access the original physiological and interactive data. Therefore, it is difficult to block the flow of sensitive original data at the source. On the other hand, desensitization processes involving truncation, adding noise, or reducing data precision can easily destroy subtle changes in the original data that characterize the user's state. For example, minute-level fluctuations in wearable physiological data and changes in operational rhythm in terminal-side interactive data may both be related to the user's current state. When this information is excessively weakened during desensitization, it reduces the effectiveness of the data used for subsequent state analysis.

[0006] Furthermore, wearable biometric data and terminal interaction data are heterogeneous time-series data from different sources. Wearable data is typically generated according to a relatively stable cycle, while touch interaction data is usually generated with user operations, exhibiting characteristics of high-frequency bursts and uneven sampling intervals.

[0007] Therefore, there is an urgent need for a data desensitization processing method, device, terminal and equipment to solve the technical problem that the original sensitive data of multi-source physiological data and interactive data is easily exposed during the correlation processing of existing technologies, and that traditional desensitization processing is prone to loss of effective state characteristics. Summary of the Invention

[0008] Therefore, it is necessary to provide a data desensitization processing method and system to address the above-mentioned technical problems. This method and system can solve the technical problems in the prior art where original sensitive data is easily exposed during the correlation processing of multi-source physiological data and interactive data, and where traditional desensitization processing easily causes the loss of effective state characteristics.

[0009] The technical solution of this invention is as follows: A data anonymization processing method, the method comprising: The wearable biometric data generated by the wearable device and the touch-side interaction data generated by the smart terminal are acquired locally on the smart terminal. Based on a unified time source, the wearable biological data and the touch interaction data are timestamped and corrected. Then, time alignment, resampling and outlier processing are performed according to a sliding time window to obtain multi-source heterogeneous feature data. The multi-source heterogeneous feature data is standardized, and the standardized multi-source heterogeneous feature data is input into the feature translation layer deployed on the smart terminal. Irreversible feature projection is performed through the feature translation layer to obtain a continuous emotional state vector. When the continuous emotion state vector is used for local processing on a smart terminal, the continuous emotion state vector is provided to the local downstream processing module; when the continuous emotion state vector is used for network synchronization, the continuous emotion state vector is subjected to discrete quantization processing to generate a discrete state feature code, and the discrete state feature code is synchronized to the network.

[0010] Optionally, the wearable-side biometric data generated by the wearable device and the touch-side interaction data generated by the smart terminal are acquired locally on the smart terminal, including: Establish wear-side data acquisition tasks for the wearable device and touch-side data acquisition tasks for the smart terminal; The wearable-side data acquisition task acquires time-series indicators of heart rate variability, transient heart rate, triaxial acceleration, step count data, and sleep stage data generated by the wearable device, forming wearable-side biological data. The touch-side data acquisition task listens to the touch interaction events generated by the smart terminal, and extracts touch timestamps, touch coordinates, touch area, touch pressure, sliding speed, operation pause time and trajectory path curvature from each touch interaction event to form touch-side interaction data; The wearable biometric data and the touch interaction data are written into the local memory of the smart terminal as dual-track heterogeneous data to be processed for timing alignment.

[0011] Optionally, the wearable biometric data and the touch-screen interaction data are timestamped based on a unified time source, and time-series aligned according to a sliding time window, including: The wearable-side clock corresponding to the wearable device and the terminal-side clock corresponding to the smart terminal are obtained, and the basic delay offset and clock drift correction rate are determined based on the timestamp interaction between the wearable device and the smart terminal. Based on the basic delay offset and the clock drift correction rate, the original timestamp corresponding to the wearable biological data is corrected to obtain the corrected timestamp corresponding to the terminal clock. Based on the preset window length and preset sliding step size, multiple sliding time windows are established on the time axis corresponding to the clock on the terminal side. Based on the correction timestamp and the touch timestamp corresponding to the touch-side interaction data, the wearable biometric data and the touch-side interaction data are divided into corresponding sliding time windows to obtain time-aligned dual-track data corresponding to each sliding time window.

[0012] Optionally, resampling the time-aligned dual-track data includes: Multiple target sampling times are established for the target sliding time window, arranged according to a preset time interval; Based on the timestamps of adjacent real touch sampling points within the target sliding time window, determine the touch sampling state corresponding to each target sampling moment; When the time interval between two adjacent real touch sampling points does not exceed the preset maximum interpolation time interval, the touch coordinates, touch pressure and touch area corresponding to the real touch sampling points are piecewise linearly interpolated to obtain the resampled touch data corresponding to each target sampling time. When no new data is generated for the wearable physiological indicators within the target sliding time window, a forward zero-order hold method is adopted to fill the data at the corresponding target sampling time with the most recent valid wearable physiological indicators to obtain resampled dual-track data.

[0013] Optionally, the resampled dual-track data undergoes anomaly processing, including: The touch swipe speed and transient heart rate in the resampled dual-track data are obtained respectively; The touch sliding speed is compared with a preset touch speed threshold, and touch data that exceeds the preset touch speed threshold is identified as abnormal touch data. The transient heart rate is compared with a preset effective heart rate range, and wearable-side biodata that is not within the preset effective heart rate range is identified as abnormal physiological data. The abnormal touch data is removed, and the abnormal physiological data is subjected to sliding median filtering to obtain the multi-source heterogeneous feature data.

[0014] Optionally, the multi-source heterogeneous feature data is standardized, including: The wearable biological data and touch interaction data corresponding to the same target sliding time window are vectorized and assembled according to a preset feature order to obtain a joint input feature vector. Determine the feature mean and feature standard deviation of each feature component in the joint input feature vector within the current processing window; Based on the feature mean and feature standard deviation corresponding to each feature component, Z-score standardization is performed on each feature component to obtain a standardized joint feature vector.

[0015] Optionally, irreversible feature projection is performed through the feature translation layer to obtain a continuous emotional state vector, including: The standardized joint feature vector is input into the feature translation layer deployed locally on the smart terminal. Obtain the pre-fixed projection weight matrix and bias vector in the feature translation layer, wherein the projection weight matrix is ​​obtained by joint comparative training targeting wearable biometric features and touch-screen interaction features; The standardized joint feature vector is projected based on the projection weight matrix and the bias vector, and the projection result is subjected to nonlinear activation processing to obtain low-dimensional mapping features. The low-dimensional mapping features are defined as continuous emotional state vectors, wherein the continuous emotional state vectors include emotional valence and emotional arousal.

[0016] Optionally, when the continuous emotion state vector is used for local processing on the smart terminal, the continuous emotion state vector is provided to the local downstream processing module, including: The target position of the continuous emotional state vector in the preset two-dimensional emotional state space is determined based on the emotional valence and the emotional arousal. The target emotional state corresponding to the target location is determined according to the preset region division rules, and hysteresis comparison is performed on the target emotional states corresponding to multiple consecutive sampling periods. When the target emotional state continuously meets the preset state switching conditions, corresponding local state output data is generated. The continuous emotional state vector and the local state output data are provided to the local downstream processing module through local inter-process communication on the smart terminal.

[0017] Optionally, when the continuous emotion state vector is used for network synchronization, the continuous emotion state vector undergoes discrete quantization processing, including: Obtain the emotional valence and emotional arousal from the continuous emotional state vector; The continuous value range corresponding to the emotional valence and the continuous value range corresponding to the emotional arousal are respectively divided into multiple discrete intervals; Based on the discrete intervals to which the emotional valence and emotional arousal belong, the corresponding first grid index and second grid index are determined to obtain the discrete grid position corresponding to the continuous emotional state vector.

[0018] Optionally, generating the discrete state feature code and synchronizing the discrete state feature code over the network includes: A combined index is generated based on the first grid index and the second grid index to characterize the discrete grid position; The combined index is encoded into the discrete state feature code; The discrete state feature code is transmitted as network synchronization data, wherein the wearable biometric data, the touch interaction data, and the continuous emotional state vector are not transmitted during the network synchronization process.

[0019] Optionally, a data anonymization processing system is also provided, the system comprising: An interactive data acquisition module is used to acquire wearable-side biometric data generated by the wearable device and touch-side interactive data generated by the smart terminal locally. The feature data generation module is used to perform timestamp correction on the wearable biological data and the touch interaction data based on a unified time source, and to perform time sequence alignment, resampling and abnormal data processing according to a sliding time window to obtain multi-source heterogeneous feature data. The state vector generation module is used to standardize the multi-source heterogeneous feature data and input the standardized multi-source heterogeneous feature data into the feature translation layer deployed on the smart terminal. Irreversible feature projection is performed through the feature translation layer to obtain a continuous emotional state vector. The discrete decryption processing module is used to provide the continuous emotion state vector to the local downstream processing module when the continuous emotion state vector is used for local processing on the smart terminal; when the continuous emotion state vector is used for network synchronization, the module performs discrete quantization processing on the continuous emotion state vector to generate a discrete state feature code, and then performs network synchronization on the discrete state feature code.

[0020] Optionally, the interactive data acquisition module is further configured to: establish a wearable-side data acquisition task for the wearable device and a touch-side data acquisition task for the smart terminal; acquire heart rate variability time-series indicators, transient heart rate, triaxial acceleration, step count data, and sleep stage data generated by the wearable device through the wearable-side data acquisition task to form wearable-side biological data; monitor touch interaction events generated by the smart terminal through the touch-side data acquisition task, and extract touch timestamps, touch coordinates, touch area, touch pressure, sliding speed, operation pause time, and trajectory path curvature from each touch interaction event to form touch-side interactive data; and write the wearable-side biological data and the touch-side interactive data into the local memory of the smart terminal as dual-track heterogeneous data to be time-aligned.

[0021] Optionally, the interactive data acquisition module is further configured to: acquire the wearable-side clock corresponding to the wearable device and the terminal-side clock corresponding to the smart terminal, and determine the basic latency offset and clock drift correction rate based on the timestamp interaction between the wearable device and the smart terminal; correct the original timestamp corresponding to the wearable-side biometric data according to the basic latency offset and the clock drift correction rate to obtain a corrected timestamp corresponding to the terminal-side clock; establish multiple sliding time windows on the time axis corresponding to the terminal-side clock according to a preset window length and a preset sliding step size; and divide the wearable-side biometric data and the touch-side interactive data into corresponding sliding time windows according to the corrected timestamp and the touch timestamp corresponding to the touch-side interactive data to obtain time-aligned dual-track data corresponding to each sliding time window.

[0022] Optionally, the feature data generation module is further configured to: establish multiple target sampling times arranged according to a preset time interval for the target sliding time window; determine the touch sampling state corresponding to each target sampling time based on the timestamps of adjacent real touch sampling points within the target sliding time window; when the time interval between two adjacent real touch sampling points does not exceed a preset maximum interpolation time interval, perform piecewise linear interpolation on the touch coordinates, touch pressure, and touch area corresponding to the real touch sampling points to obtain resampled touch data corresponding to each target sampling time; when no new data is generated by the wearable physiological indicators within the target sliding time window, adopt a forward zero-order hold method to fill the data of the corresponding target sampling time based on the most recent effective wearable physiological indicators to obtain resampled dual-track data.

[0023] Optionally, the feature data generation module is further configured to: acquire the touch sliding speed and transient heart rate in the resampled dual-track data respectively; compare the touch sliding speed with a preset touch speed threshold, and determine touch data exceeding the preset touch speed threshold as abnormal touch data; compare the transient heart rate with a preset effective heart rate range, and determine wearable-side biological data not within the preset effective heart rate range as abnormal physiological data; remove the abnormal touch data, and perform sliding median filtering on the abnormal physiological data to obtain the multi-source heterogeneous feature data.

[0024] Optionally, the state vector generation module is further configured to: vectorize and assemble wearable biological data and touch interaction data corresponding to the same target sliding time window according to a preset feature order to obtain a joint input feature vector; determine the feature mean and feature standard deviation of each feature component in the joint input feature vector in the current processing window; and perform Z-score standardization on each feature component according to the feature mean and feature standard deviation of each feature component to obtain a standardized joint feature vector.

[0025] Optionally, the state vector generation module is further configured to: input the standardized joint feature vector into a feature translation layer deployed locally on the smart terminal; obtain a pre-fixed projection weight matrix and bias vector in the feature translation layer, wherein the projection weight matrix is ​​obtained by joint comparative training targeting wearable biometric features and touch-screen interaction features; perform feature projection on the standardized joint feature vector according to the projection weight matrix and the bias vector, and perform nonlinear activation processing on the projection result to obtain a low-dimensional mapping feature; determine the low-dimensional mapping feature as a continuous emotional state vector, wherein the continuous emotional state vector includes emotional valence and emotional arousal.

[0026] Optionally, the discrete decryption processing module is further configured to: determine the target position of the continuous emotional state vector in a preset two-dimensional emotional state space based on the emotional valence and the emotional arousal; determine the target emotional state corresponding to the target position based on a preset region division rule, and perform hysteresis comparison on the target emotional states corresponding to multiple consecutive sampling periods; generate corresponding local state output data when the target emotional state continuously meets preset state switching conditions; and provide the continuous emotional state vector and the local state output data to the local downstream processing module through local inter-process communication on the smart terminal.

[0027] Optionally, the discrete decryption processing module is further configured to: obtain the emotional valence and emotional arousal in the continuous emotional state vector; divide the continuous value range corresponding to the emotional valence and the continuous value range corresponding to the emotional arousal into multiple discrete intervals respectively; determine the corresponding first grid index and second grid index according to the discrete intervals to which the emotional valence and the emotional arousal belong, and obtain the discrete grid position corresponding to the continuous emotional state vector.

[0028] Optionally, the discrete decryption processing module is further configured to: generate a combined index for characterizing the discrete grid position based on the first grid index and the second grid index; encode the combined index into the discrete state feature code; and transmit the discrete state feature code as network synchronization data, wherein the wearable biometric data, the touch interaction data, and the continuous emotional state vector are not transmitted during network synchronization.

[0029] Optionally, a computer terminal is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps described in the above data desensitization processing method.

[0030] Optionally, a computer-readable storage medium is also provided, on which a computer program is stored, which, when executed by a processor, implements the steps described in the above-described data desensitization processing method.

[0031] The technical effects achieved by this invention are as follows: The aforementioned data anonymization methods, devices, terminals, and equipment acquire wearable biological data generated by the wearable device and touch interaction data generated by the smart terminal locally. This allows physiological information and interactive behavior information from different sources to enter a unified data processing process before the original data is transmitted externally, thus providing a data foundation for multi-source heterogeneous data processing and anonymization at the data source. By performing timestamp correction on wearable biological data and touch interaction data based on a unified time source, and performing time alignment, resampling, and abnormal data processing according to a sliding time window, a unified time correspondence can be established between wearable biological data and touch interaction data with different sampling periods and different data generation methods. This reduces the impact of clock offset, sampling interval differences, and abnormal sampling data between the wearable device and the smart terminal on subsequent feature fusion, improving the time consistency and data validity among multi-source heterogeneous data. By standardizing the multi-source heterogeneous feature data and inputting the standardized multi-source heterogeneous feature data into the feature translation layer deployed locally on the smart terminal for irreversible feature projection, the original physiological values ​​and original touch interaction physical quantities are converted into continuous emotional state vectors to represent the user's state. While reducing the possibility of reverse recovery of raw heart rate data, touch coordinates, and other sensitive physical quantities, the system retains the feature relationships related to the user's physical and mental state in multi-source data, thereby reducing the damage to effective state features caused by traditional truncation or noise perturbation-based desensitization processing. When the continuous emotional state vector is only used for local processing on the smart terminal, it is directly provided to the local downstream processing module, enabling the local application to obtain information reflecting the user's continuous state changes without transmitting the original wearable biological data and touch interaction data to the network side. This helps to balance the data validity of local state processing with the privacy and security of the original data. When the continuous emotional state vector needs to be synchronized with the network, the continuous state information is converted into discrete state feature codes by further discretization of the continuous emotional state vector. Only the discrete state feature codes are synchronized with the network, so that the data obtained by the network side is only used to represent the corresponding macroscopic state interval, without including the original physiological values, precise touch trajectories, and continuous state coordinates. This further reduces the possibility of external nodes reversing the user's original sensitive data based on the network-synchronized data. Furthermore, this application forms a complete data processing chain from multi-source heterogeneous data acquisition, unified time-series processing, and irreversible feature translation from the edge to local and network hierarchical output. While reducing the risk of privacy leakage during multi-source data association processing, it retains effective features for user status analysis and provides different granular data expression forms according to different data usage needs for local use and network synchronization, thereby taking into account data anonymization security, status feature validity, and downstream data usage needs. Attached Figure Description

[0032] Figure 1This is a flowchart illustrating a data anonymization process in one embodiment; Figure 2 This is a structural block diagram of a data desensitization processing system in one embodiment. Detailed Implementation

[0033] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0034] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0035] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0036] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0037] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0038] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0039] In one embodiment, such as Figure 1 As shown, a data anonymization processing method is provided, the method comprising: Step S100: Step S100: Acquire wearable biometric data generated by the wearable device and touch-side interaction data generated by the smart terminal locally; Step S200: Based on a unified time source, timestamp correction is performed on the wearable biological data and the touch interaction data, and time alignment, resampling and abnormal data processing are performed according to a sliding time window to obtain multi-source heterogeneous feature data; Step S300: Standardize the multi-source heterogeneous feature data, and input the standardized multi-source heterogeneous feature data into the feature translation layer deployed on the smart terminal. Through the feature translation layer, irreversible feature projection is performed to obtain a continuous emotional state vector. Step S400: When the continuous emotion state vector is used for local processing on the smart terminal, the continuous emotion state vector is provided to the local downstream processing module; when the continuous emotion state vector is used for network synchronization, the continuous emotion state vector is subjected to discrete quantization processing to generate a discrete state feature code, and the discrete state feature code is synchronized to the network.

[0040] This application acquires wearable biological data and touch interaction data generated by wearable devices locally on the smart terminal, enabling physiological information and interactive behavior information from different sources to enter a unified data processing process before the raw data is transmitted externally. This provides a data foundation for multi-source heterogeneous data processing and desensitization at the data source. By performing timestamp correction on wearable biological data and touch interaction data based on a unified time source, and performing time alignment, resampling, and abnormal data processing according to a sliding time window, a unified time correspondence can be established between wearable biological data and touch interaction data with different sampling periods and different data generation methods. This reduces the impact of clock offset, sampling interval differences, and abnormal sampling data between wearable devices and smart terminals on subsequent feature fusion, improving the time consistency and data validity among multi-source heterogeneous data. By standardizing the multi-source heterogeneous feature data and inputting the standardized multi-source heterogeneous feature data into the feature translation layer deployed locally on the smart terminal for irreversible feature projection, the original physiological values ​​and original touch interaction physical quantities are converted into continuous emotional state vectors to represent the user's state. While reducing the possibility of reverse recovery of raw heart rate data, touch coordinates, and other sensitive physical quantities, the system retains the feature relationships related to the user's physical and mental state in multi-source data, thereby reducing the damage to effective state features caused by traditional truncation or noise perturbation-based desensitization processing. When the continuous emotional state vector is only used for local processing on the smart terminal, it is directly provided to the local downstream processing module, enabling the local application to obtain information reflecting the user's continuous state changes without transmitting the original wearable biological data and touch interaction data to the network side. This helps to balance the data validity of local state processing with the privacy and security of the original data. When the continuous emotional state vector needs to be synchronized with the network, the continuous state information is converted into discrete state feature codes by further discretization of the continuous emotional state vector. Only the discrete state feature codes are synchronized with the network, so that the data obtained by the network side is only used to represent the corresponding macroscopic state interval, without including the original physiological values, precise touch trajectories, and continuous state coordinates. This further reduces the possibility of external nodes reversing the user's original sensitive data based on the network-synchronized data. Furthermore, this application forms a complete data processing chain from multi-source heterogeneous data acquisition, unified time-series processing, and irreversible feature translation from the edge to local and network hierarchical output. While reducing the risk of privacy leakage during multi-source data association processing, it retains effective features for user status analysis and provides different granular data expression forms according to different data usage needs for local use and network synchronization, thereby taking into account data anonymization security, status feature validity, and downstream data usage needs.

[0041] In one embodiment, step S100, acquiring wearable-side biometric data generated by the wearable device and touch-side interaction data generated by the smart terminal locally, includes: Step S110: Establish a wearable-side data acquisition task for the wearable device and a touch-side data acquisition task for the smart terminal; Step S120: Acquire the heart rate variability time series index, transient heart rate, triaxial acceleration, step count data and sleep stage data generated by the wearable device through the wearable side data acquisition task to form wearable side biological data; Step S130: Listen to the touch interaction events generated by the smart terminal through the touch side data acquisition task, and extract the touch timestamp, touch coordinates, touch area, touch pressure, sliding speed, operation pause time and trajectory path curvature from each touch interaction event to form touch side interaction data; Step S140: Write the wearable biometric data and the touch interaction data into the local memory of the smart terminal as dual-track heterogeneous data to be time-aligned.

[0042] In this embodiment, in step S100, the smart terminal locally acquires wearable-side biometric data generated by the wearable device and touch-side interaction data generated by the smart terminal itself. This provides dual-track heterogeneous data for subsequent time-series alignment and feature translation without uploading the original biometric data and original touch data to an external server. The wearable device is a smartwatch, smart bracelet, or other wearable electronic device capable of collecting the user's physiological and movement states. The smart terminal is a smartphone, tablet, or other electronic device with touch interaction capabilities.

[0043] Specifically, in step S110, the smart terminal establishes a wearable-side data acquisition task for the wearable device and a touch-side data acquisition task for the smart terminal itself. The wearable-side data acquisition task is used to acquire data generated by the wearable device according to the data transmission methods supported by the wearable device, while the touch-side data acquisition task is used to monitor touch interaction events occurring locally on the smart terminal.

[0044] In one embodiment, the smart terminal and the wearable device establish a wireless communication connection beforehand. After the wearable device completes identity verification and data access authorization, the smart terminal registers a wearable-side data acquisition task according to the data interface provided by the wearable device. The wireless communication connection can use Bluetooth, Bluetooth Low Energy, or other data communication methods suitable for communication between the smart terminal and the wearable device. The wearable-side data acquisition task can acquire wearable-side data through periodic reading or by the wearable device actively pushing the data. For data with different sampling frequencies, the wearable-side data acquisition task saves the original acquisition time corresponding to each data point for subsequent timestamp correction.

[0045] The touch-side data acquisition task is run locally on the smart terminal, and listens for touch interaction events generated during user operations through the touch event listening interface, input event interface, or application interaction event interface provided by the operating system. The touch-side data acquisition task only obtains the underlying interaction parameters that characterize the physical characteristics of the user's touch behavior, and does not obtain the text content entered by the user, the semantic information corresponding to the input content, or the business content in the application page.

[0046] In step S120, the smart terminal acquires heart rate variability time-series indicators, transient heart rate, triaxial acceleration, step count data and sleep stage data generated by the wearable device through the wearable side data acquisition task, forming wearable side biological data.

[0047] Among them, the heart rate variability time series index is used to characterize the change of adjacent heartbeat intervals over time, and is a heart rate variability index calculated by the wearable device based on continuous heartbeat information; transient heart rate is used to characterize the heart rate state at the corresponding sampling time or within the corresponding short time period; triaxial acceleration includes acceleration data collected by the wearable device along three mutually orthogonal directions, used to characterize the user's body movement state; step count data is used to characterize walking activities within a preset time range; sleep stage data is used to characterize at least one of the wakefulness state, light sleep state, deep sleep state, or REM sleep state identified by the wearable device.

[0048] Different types of wearable biometric data can have different data update cycles. For example, triaxial acceleration can be generated continuously at a higher sampling frequency, transient heart rate and heart rate variability time-series indicators can be generated according to the corresponding physiological detection cycle, and step count data and sleep stage data can be updated at a lower frequency. When acquiring the above data, the smart terminal retains the original timestamps of each data point on the wearable side. It does not require different types of wearable biometric data to have the same sampling frequency in step S120. Subsequently, time correction and resampling processing are performed uniformly in step S200.

[0049] In step S130, the smart terminal listens to the touch interaction events generated by the smart terminal through the touch side data acquisition task, and extracts the touch timestamp, touch coordinates, touch area, touch pressure, sliding speed, operation pause time and trajectory path curvature from each touch interaction event to form touch side interaction data.

[0050] Touch interaction events include press events, move events, and release events generated after a user's finger comes into contact with the touch interface of a smart terminal. Touch timestamps indicate the time of occurrence of the corresponding touch interaction event under the terminal's clock; touch coordinates indicate the position of the corresponding touch point on the touch interface; touch area represents the size of the contact area between the user's finger and the touch interface; and touch pressure represents the intensity of the touch action, which can be the pressure parameters directly provided by the smart terminal's touch interface.

[0051] The swipe speed can be determined based on the positional change between consecutive touch points and the corresponding time interval. The operation pause time can be determined based on the time interval between consecutive touch points during the same complete touch operation. For example, when the positional change between two consecutive touch points is less than a preset positional change threshold, the corresponding duration is used as the operation pause time. The trajectory path curvature is used to characterize the degree of curvature of the touch trajectory and can be determined based on the change in the direction of movement between multiple consecutive touch points. Therefore, the touch-side interaction data primarily reflects the speed, pressure, pauses, and trajectory changes during the user's touch operation, without involving the semantics of the actual user input.

[0052] In a specific example, when a user performs a swipe operation on the touch interface of a smart terminal, the touch-side data acquisition task records the touch timestamps, touch coordinates, touch area, and touch pressure of consecutive touch points from the press event onwards, and continues recording until the corresponding release event ends. Based on the positional and temporal changes corresponding to the consecutive touch points, the smart terminal further determines the swipe speed, operation pause time, and trajectory path curvature, and arranges the data generated by one or more touch operations according to their corresponding touch timestamps to form touch-side interaction data.

[0053] In step S140, after obtaining wearable biometric data and touch interaction data, the smart terminal writes the two types of data into the local memory of the smart terminal as dual-track heterogeneous data to be processed for timing alignment.

[0054] In this embodiment, wearable biometric data and touch interaction data each retain their own data source identifier, data type identifier, data value, and corresponding timestamp, enabling differentiation of different data sources during subsequent processing. The smart terminal establishes corresponding memory cache areas for wearable biometric data and touch interaction data, and writes them into the corresponding cache areas in the order of data generation time; alternatively, the two types of data can be written into the same local cache area according to a unified data recording format and differentiated by the data source identifier.

[0055] The dual-track heterogeneous data here refers to a data set originating from two different data acquisition links—wearable devices and smart terminals—that differ in data type, physical quantity, sampling period, and data update method. Specifically, wearable-side biometric data primarily characterizes the user's physiological and physical activity states, while touch-side interaction data primarily characterizes the user's operational behavior during human-computer interaction with the smart terminal. Step S140 only temporarily stores the aforementioned raw dual-track heterogeneous data locally on the smart terminal, without directly sending the wearable-side biometric data and the touch-side interaction data to external devices as network synchronization data.

[0056] Furthermore, the dual-track heterogeneous data in local memory can be stored in a rolling manner according to a preset cache duration. When the preset cache duration is exceeded or the corresponding data has completed subsequent time alignment and feature processing, the corresponding original data is overwritten or cleared, thereby reducing the long-term retention of original biological data and original touch data in the smart terminal. In the subsequent step S200, the dual-track heterogeneous data in local memory is used as input to uniformly correct the wearable-side timestamp and the terminal-side timestamp, and further performs sliding time window division, resampling, and abnormal data processing.

[0057] In one embodiment, step S200 involves performing timestamp correction on the wearable biometric data and the touch-based interaction data based on a unified time source, and aligning the sequences according to a sliding time window, including: Step S210: Obtain the wearable-side clock corresponding to the wearable device and the terminal-side clock corresponding to the smart terminal, and determine the basic delay offset and clock drift correction rate based on the timestamp interaction between the wearable device and the smart terminal; Step S220: Based on the basic delay offset and the clock drift correction rate, correct the original timestamp corresponding to the wearable bio data to obtain a corrected timestamp corresponding to the terminal clock. Step S230: Based on the preset window length and preset sliding step size, establish multiple sliding time windows on the time axis corresponding to the clock on the terminal side; Step S240: Based on the correction timestamp and the touch timestamp corresponding to the touch-side interaction data, divide the wearable biometric data and the touch-side interaction data into corresponding sliding time windows to obtain time-aligned dual-track data corresponding to each sliding time window.

[0058] In this embodiment, in step S200, the smart terminal performs timestamp correction on the wearable biometric data and the touch interaction data based on a unified time source, and performs time sequence alignment on the two types of data according to a sliding time window, so as to eliminate the time deviation caused by the wearable device and the smart terminal using different local clocks and different data sampling periods, so that the wearable biometric data and the touch interaction data generated within the same time range are divided into the same sliding time window.

[0059] Specifically, in step S210, after the smart terminal and the wearable device establish a data communication connection, the smart terminal periodically sends a time synchronization request to the wearable device. The wearable device responds to the time synchronization request by returning a wearable-side timestamp recorded by its own clock. The smart terminal records the sending time of the time synchronization request and the receiving time of the corresponding response data. Based on the sending time, receiving time, and wearable-side timestamp corresponding to the same timestamp interaction, the smart terminal determines the time offset of the wearable-side clock relative to the terminal-side clock, and determines the basic delay offset based on the time offset obtained from multiple timestamp interactions.

[0060] The base delay offset is used to characterize the basic time difference between the wearable-side timestamp and the terminal-side timestamp during the current time synchronization phase. When the smart terminal performs multiple timestamp interactions, data interaction results that are significantly longer than the normal communication delay are discarded, and the base delay offset is determined based on the remaining timestamp interaction results to reduce the impact of instantaneous communication congestion or wireless transmission fluctuations on the time correction results.

[0061] As the wearable device continues to operate, a gradually changing time deviation will occur between the wearable-side clock and the terminal-side clock due to errors in their respective clock source frequencies. Therefore, the smart terminal further determines the trend of the time deviation over time based on multiple time offset results obtained from timestamp interactions at different times, and determines the clock drift correction rate based on this trend. The clock drift correction rate is used to characterize the cumulative degree of deviation of the wearable-side clock relative to the terminal-side clock over time.

[0062] In step S220, for any wearable bio data, the smart terminal first performs an overall time offset correction on the original timestamp corresponding to the wearable bio data based on the basic time delay offset, and then corrects the clock drift that accumulates over time based on the time interval between the original timestamp and the reference synchronization time and the clock drift correction rate, thereby converting the original timestamp recorded by the wearable clock to the time axis corresponding to the terminal clock.

[0063] When the smart terminal re-establishes a communication connection with the wearable device, or detects that the current time offset exceeds the preset time offset threshold, the smart terminal re-executes the timestamp interaction and updates the basic latency offset and clock drift correction rate to ensure that the wearable data and the terminal data maintain a consistent time base during continuous operation.

[0064] In step S230, the preset window length is used to limit the time range covered by each data joint processing, and the preset sliding step size is used to limit the time interval between the start times of two adjacent sliding time windows. When the preset sliding step size is less than the preset window length, the adjacent sliding time windows partially overlap, thereby preserving the data correlation between consecutive time periods.

[0065] For example, a specific moment on the terminal's clock is used as the start time of the first sliding time window. The end time of the first sliding time window is determined according to a preset window length. Then, the start time of the next sliding time window is determined according to a preset sliding step size, and multiple sliding time windows are established sequentially. The preset window length and preset sliding step size are preset based on the update frequency of wearable biometric data, the generation frequency of touch interaction data, and the local processing load of the smart terminal.

[0066] In step S240, for any sliding time window, wearable biometric data whose correction timestamp is within the time range of the sliding time window is determined as wearable data corresponding to the sliding time window, and touch-side interactive data whose touch timestamp is within the time range of the sliding time window is determined as touch-side data corresponding to the sliding time window, thereby establishing a time correspondence between the two types of data under a unified terminal time axis.

[0067] In this embodiment, the timing-aligned dual-track data obtained in step S240 still maintains the data types and sampling times of the wearable side data and the touch side data, and does not require the two types of data to have the same data quantity or the same sampling interval. Subsequently, the data with different sampling frequencies are further mapped to a unified target sampling time through resampling processing.

[0068] In one embodiment, step S200, resampling the time-aligned dual-track data, includes: Step S250: Establish multiple target sampling times arranged according to preset time intervals for the target sliding time window; Step S260: Determine the touch sampling state corresponding to each target sampling moment based on the timestamps of adjacent real touch sampling points within the target sliding time window; Step S270: When the time interval between two adjacent real touch sampling points does not exceed the preset maximum interpolation time interval, perform piecewise linear interpolation on the touch coordinates, touch pressure and touch area corresponding to the real touch sampling points to obtain the resampled touch data corresponding to each target sampling time. Step S280: When no new data is generated for the wearable physiological indicators within the target sliding time window, the forward zero-order hold method is adopted to fill the data at the corresponding target sampling time according to the most recent effective wearable physiological indicators to obtain resampled dual-track data.

[0069] In this embodiment, in steps S200 and S250, for any target sliding time window to be processed, the smart terminal establishes multiple target sampling moments according to a preset time interval. The multiple target sampling moments are distributed in chronological order within the target sliding time window, forming a unified resampling time sequence corresponding to the target sliding time window.

[0070] In this embodiment, the preset time interval is determined in advance based on the sampling characteristics of touch-side interaction data and wearable-side biometric data. For continuously generated high-frequency data, the data misalignment caused by differences in sampling frequencies between different devices is reduced by unifying the target sampling time; for low-frequency updated data, corresponding processing is performed based on the valid data state closest to the target sampling time.

[0071] In step S260, the actual touch sampling point is the touch sampling point actually generated by the smart terminal's touch system. For any target sampling moment, the smart terminal determines whether there are actual touch sampling points belonging to the same continuous touch operation before and after the target sampling moment, and determines whether the target sampling moment is in a continuous effective touch segment, a non-touch segment, or a segment with missing touch data.

[0072] When the target sampling moment falls between a press event and its corresponding release event, and there are consecutive real touch sampling points before and after it, the target sampling moment is determined as a valid touch sampling state. When there is no corresponding touch operation at the target sampling moment, it is determined as a non-touch state. When the time interval between adjacent real touch sampling points increases abnormally, the corresponding time range is determined as a missing touch data segment to avoid performing continuous interpolation across long data gaps.

[0073] In step S270, piecewise linear interpolation determines the touch coordinates, touch pressure, and touch area corresponding to the target sampling time based on the relative position of the target sampling time within the time interval of two adjacent real touch sampling points. Thus, without altering the overall trend of the original touch trajectory, irregularly generated touch sampling points are converted into touch data arranged according to a uniform time interval.

[0074] When the time interval between two adjacent actual touch sampling points exceeds the preset maximum interpolation time interval, the smart terminal does not perform continuous interpolation between the two actual touch sampling points, but marks the corresponding target sampling moment as having no valid touch data or missing touch data. The preset maximum interpolation time interval is preset based on the normal touch sampling cycle of the smart terminal and the allowable degree of data loss, thereby avoiding the erroneous connection of two actual discontinuous touch processes into a continuous touch trajectory.

[0075] In step S280, forward zero-order hold refers to maintaining the most recently obtained valid data value before new valid data is generated. For example, between two adjacent transient heart rate update times, the previous valid transient heart rate is used as the transient heart rate at the corresponding target sampling time; for low-frequency update data in heart rate variability time series indicators, step count data, and sleep stage data, the state corresponding to the most recently obtained valid data is also maintained backward until new valid data is obtained.

[0076] For continuously updated data such as triaxial acceleration, the data is mapped to the target sampling time series according to the time correspondence between the actual sampling time and the target sampling time. This unifies the wearable and touch-based data within the target sliding time window to multiple target sampling times, resulting in resampled dual-track data.

[0077] In one embodiment, step S200, performing abnormal data processing on the resampled dual-track data, includes: Step S291: Obtain the touch swipe speed and transient heart rate from the resampled dual-track data respectively; Step S292: Compare the touch sliding speed with a preset touch speed threshold, and determine the touch data that exceeds the preset touch speed threshold as abnormal touch data; Step S293: Compare the transient heart rate with the preset effective heart rate range, and identify the wearable biological data that is not within the preset effective heart rate range as abnormal physiological data; Step S294: Remove the abnormal touch data and perform sliding median filtering on the abnormal physiological data to obtain the multi-source heterogeneous feature data.

[0078] In this embodiment, the touch swiping speed is determined based on the change in touch coordinates corresponding to the target sampling time and the time interval between adjacent target sampling times; the transient heart rate is the heart rate data output by the wearable device during the corresponding sampling period, or the heart rate data corresponding to the target sampling time after resampling in step S280.

[0079] In step S292, the smart terminal compares each touch sliding speed with a preset touch speed threshold. When any touch sliding speed exceeds the preset touch speed threshold, the corresponding touch data is identified as abnormal touch data.

[0080] In this embodiment, the preset touch speed threshold is predetermined based on the effective size of the smart terminal's touch interface, the normal touch sampling period, and the speed range corresponding to normal user touch operations. Touch data exceeding the preset touch speed threshold typically corresponds to touch coordinate jumps, abnormal sampling timestamps, or incorrect association of consecutive touch points; therefore, the corresponding data is marked as abnormal touch data.

[0081] In step S293, the smart terminal compares each transient heart rate with a preset effective heart rate range. When any transient heart rate is lower than the lower limit of the preset effective heart rate range or higher than the upper limit of the preset effective heart rate range, the corresponding wearable biological data is identified as abnormal physiological data.

[0082] In this embodiment, the preset effective heart rate range is pre-set based on the effective detection range of the wearable device and the normal physiological data range of the human body under the target application scenario. For significantly abnormal heart rate values ​​caused by loose wearable devices, momentary failure of optical sensors, or abnormal wireless data, the preset effective heart rate range is used for identification.

[0083] In step S294, the smart terminal removes abnormal touch data and performs sliding median filtering on abnormal physiological data to obtain multi-source heterogeneous feature data.

[0084] Specifically, for abnormal touch data, the smart terminal deletes the corresponding abnormal touch sampling point or marks the corresponding target sampling time as having no valid touch data, so that the abnormal data does not participate in subsequent joint feature assembly. For abnormal physiological data, a local data window of a preset length is established with the time where the abnormal data is located as the center, and multiple valid physiological data in the local data window are obtained. The abnormal physiological data is replaced according to the median of the multiple valid physiological data, thereby suppressing the influence of a single outlier on the trend of continuous physiological data changes.

[0085] After processing the abnormal data, the remaining wearable biological data and touch interaction data together constitute the multi-source heterogeneous feature data corresponding to the target sliding time window.

[0086] In one embodiment, step S300, standardizing the multi-source heterogeneous feature data, includes: Step S310: The wearable biological data and touch interaction data corresponding to the same target sliding time window are vectorized and assembled according to a preset feature order to obtain a joint input feature vector; Step S320: Determine the feature mean and feature standard deviation of each feature component in the joint input feature vector in the current processing window; Step S330: Based on the feature mean and feature standard deviation corresponding to each feature component, perform Z-score standardization on each feature component to obtain a standardized joint feature vector.

[0087] In this embodiment, after obtaining multi-source heterogeneous feature data, the smart terminal first performs standardization processing on the multi-source heterogeneous feature data to reduce the numerical differences between heart rate, acceleration, touch coordinates, touch pressure and trajectory parameters caused by different physical quantities and value scales.

[0088] In step S310, the preset feature order is determined in advance when the feature translation layer is deployed on the smart terminal, and it is consistent with the input dimension corresponding to the feature translation layer. For example, the data are arranged in the order of heart rate variability time series index, transient heart rate, triaxial acceleration, step count data, sleep stage data, touch coordinates, touch area, touch pressure, swipe speed, operation pause time and trajectory path curvature.

[0089] For data containing multiple target sampling times within the target sliding time window, the smart terminal arranges multiple valid sampled values ​​corresponding to the same feature in chronological order, and then assembles the data of different features according to the preset feature order, thereby forming a joint input feature vector to characterize the physiological state of the wearer and the behavioral state of the touch device within the target sliding time window.

[0090] In step S320, for any feature component, the smart terminal determines the feature mean based on multiple valid data points corresponding to that feature within the target sliding time window, and determines the corresponding feature standard deviation based on the dispersion of each valid data point relative to the feature mean. Thus, the statistical scales corresponding to different data types within the current target sliding time window are obtained.

[0091] If the number of valid data for a certain feature within the target sliding time window is insufficient or the standard deviation of the corresponding feature is less than the preset minimum standard deviation, the preset minimum standard deviation will be used as the standard deviation corresponding to the feature in subsequent standardization processing, thereby avoiding abnormal increases in the standardization result due to the feature value remaining unchanged for a long time.

[0092] In step S330, for each feature component, the difference between the corresponding original feature value and the feature mean is used to characterize its offset relative to the current window average state. Then, the offset is scaled according to the corresponding feature standard deviation, so that different features after standardization have similar numerical scales, thereby reducing the impact of differences in the numerical range of different physical quantities on subsequent feature projection.

[0093] In one embodiment, step S300 involves irreversible feature projection through the feature translation layer to obtain a continuous emotional state vector, including: Step S340: Input the standardized joint feature vector into the feature translation layer local to the smart terminal; Step S350: Obtain the pre-fixed projection weight matrix and bias vector in the feature translation layer, wherein the projection weight matrix is ​​obtained by joint comparative training targeting wearable biometric features and touch interaction features; Step S360: Project the standardized joint feature vector according to the projection weight matrix and the bias vector, and perform nonlinear activation processing on the projection result to obtain low-dimensional mapping features; Step S370: Determine the low-dimensional mapping feature as a continuous emotional state vector, wherein the continuous emotional state vector includes emotional valence and emotional arousal.

[0094] In this embodiment, after obtaining the standardized joint feature vector, the smart terminal performs irreversible feature projection on the standardized joint feature vector through a feature translation layer deployed locally, converting the high-dimensional joint features containing wearable physiological information and touch-based behavioral information into a low-dimensional continuous emotional state vector.

[0095] In step S340, the feature translation layer runs locally on the smart terminal with pre-trained and solidified parameters. The standardized joint feature vector completes input, projection and output processing locally, without needing to send the wearable bio data and touch interaction data before or after standardization to an external server.

[0096] In step S350, before deploying the feature translation layer to the smart terminal, training samples including multiple sets of wearable biometric features and touch-screen interaction features are pre-established. Each set of training samples corresponds to pre-determined emotional state information. For samples with similar emotional states, the differences between their projected features are reduced during joint contrastive training; for samples with large differences in emotional states, the discriminative power between their projected features is increased, thereby adjusting the projection weight matrix and bias vector in the feature translation layer.

[0097] After completing the joint contrastive training, the determined projection weight matrix and bias vector are embedded into the feature translation layer of the smart terminal. During actual operation of the smart terminal, it is not necessary to retrain the projection weight matrix using the current user's original wearable biometric data and touch interaction data.

[0098] In step S360, the projection weight matrix cross-combines and compresses the dimensions of multiple high-dimensional feature components in the standardized joint feature vector, enabling both wearable biometric features and touch-screen interaction features to be mapped to a low-dimensional feature space. Nonlinear activation processing further alters the linear correspondence between each input feature and the low-dimensional mapped features.

[0099] In this embodiment, the irreversible feature projection refers to the process where, after the standardized joint feature vector undergoes high-dimensional to low-dimensional feature compression and nonlinear activation, multiple different combinations of original features are mapped to state positions in the low-dimensional feature space. The smart terminal subsequently only uses the low-dimensional mapped features and does not reconstruct the original heart rate, acceleration, touch coordinates, touch pressure, or touch trajectory data based on these features. Therefore, the low-dimensional mapped features do not retain the one-to-one correspondence used to directly reconstruct the original wearable-side biological data and touch-side interaction data.

[0100] In step S370, emotional valence is used to characterize the degree of continuous change in the user's current emotional state between negative and positive directions, and emotional arousal is used to characterize the degree of continuous change in the user's current emotional state between low and high activation states. Emotional valence and emotional arousal together form a two-dimensional continuous emotional state representation, so that the original wearable biometrics and touch interaction features constituting the continuous emotional state vector are no longer directly used in subsequent processing by the smart terminal.

[0101] In one embodiment, in step S400, when the continuous emotion state vector is used for local processing on a smart terminal, the continuous emotion state vector is provided to the local downstream processing module, including: Step S410: Determine the target position of the continuous emotional state vector in the preset two-dimensional emotional state space based on the emotional valence and the emotional arousal. Step S420: Determine the target emotional state corresponding to the target location according to the preset region division rules, and perform hysteresis comparison on the target emotional states corresponding to multiple consecutive sampling periods; Step S430: When the target emotional state continuously meets the preset state switching conditions, generate corresponding local state output data; Step S440: Provide the continuous emotional state vector and the local state output data to the local downstream processing module through local inter-process communication on the smart terminal.

[0102] In this embodiment, in step S410, the smart terminal determines the target position of the continuous emotional state vector in a preset two-dimensional emotional state space based on the emotional valence and emotional arousal in the continuous emotional state vector.

[0103] In this embodiment, a preset two-dimensional emotional state space is established, with emotional valence corresponding to one coordinate direction and emotional arousal corresponding to another coordinate direction. After obtaining a continuous emotional state vector each time, the corresponding coordinate position in the two-dimensional emotional state space is determined based on the current emotional valence and the current emotional arousal, and this coordinate position is designated as the target position.

[0104] In step S420, the preset two-dimensional emotional state space is divided into multiple emotional state regions based on the range of emotional valence and emotional arousal. For example, the two-dimensional emotional state space is divided into a positive high arousal region, a positive low arousal region, a negative high arousal region, a negative low arousal region, and a neutral region. When the target position falls into any emotional state region, the state corresponding to that emotional state region is determined as the target emotional state for the current sampling period.

[0105] To avoid frequent switching of the target emotional state caused by small fluctuations in the continuous emotional state vector near the boundary of adjacent emotional state regions, the smart terminal does not immediately change the current output state based on the target emotional state of a single sampling period. Instead, it records the target emotional state corresponding to multiple consecutive sampling periods and compares it with the emotional state currently held by the smart terminal using hysteresis.

[0106] In step S430, the preset state switching conditions include the target emotional state remaining consistent for multiple consecutive sampling periods, or the target position continuously being outside the region corresponding to the current state and entering the region of another target emotional state for multiple consecutive sampling periods. Only after the preset state switching conditions are met will the smart terminal switch the currently maintained emotional state to the new target emotional state.

[0107] The local state output data includes a state identifier for the current target emotional state and corresponding state change information. The local downstream processing module performs local functional processing corresponding to the current target emotional state based on the local state output data, without needing to read the original wearable biometric data and the original touch interaction data.

[0108] In step S440, local inter-process communication occurs within the smart terminal. The processing process containing the feature translation layer writes the continuous emotion state vector and local state output data into the local inter-process communication channel, and the downstream processing module reads the corresponding data from the local inter-process communication channel. During this local inter-process communication, it is not necessary to send the continuous emotion state vector to the external network of the smart terminal.

[0109] In one embodiment, in step S400, when the continuous emotional state vector is used for network synchronization, the continuous emotional state vector undergoes discrete quantization processing, including: Step S450: Obtain the emotional valence and emotional arousal in the continuous emotional state vector; Step S460: Divide the continuous value range corresponding to the emotional valence and the continuous value range corresponding to the emotional arousal into multiple discrete intervals respectively; Step S470: Based on the discrete intervals to which the emotional valence and emotional arousal belong, determine the corresponding first grid index and second grid index to obtain the discrete grid position corresponding to the continuous emotional state vector.

[0110] In this embodiment, when the continuous emotional state vector is used for network synchronization, the smart terminal does not directly use the continuous emotional state vector as network synchronization data. Instead, it first performs discrete quantization on the emotional valence and emotional arousal, converting the continuous values ​​into discrete grid positions.

[0111] In step S450, both emotional valence and emotional arousal have corresponding preset continuous value ranges. These continuous value ranges are consistent with the output range of the feature translation layer and are predetermined during the feature translation layer deployment phase.

[0112] In step S460, the continuous range of emotional valence is divided into multiple discrete valence intervals according to a preset interval width, and the continuous range of emotional arousal is divided into multiple discrete arousal intervals according to a preset interval width. Each discrete interval is assigned a corresponding interval number in ascending order of its corresponding continuous value.

[0113] The number of discrete intervals is predetermined based on the emotional state resolution and data anonymization required for network synchronization. Increasing the number of discrete intervals improves the granularity of the discrete grid in distinguishing continuous emotional states; decreasing the number of discrete intervals increases the range of continuous states covered by the same discrete interval, thereby reducing the accuracy of the network synchronization data in representing the original continuous states.

[0114] In step S470, the smart terminal determines the first grid index based on the discrete interval of the current emotional valence and the second grid index based on the discrete interval of the current emotional arousal. The first grid index and the second grid index are used to determine the discrete grid position corresponding to the continuous emotional state vector.

[0115] In this embodiment, each combination of the first grid index and each second grid index corresponds to a discrete grid in a preset two-dimensional emotional state space. Thus, the emotional valence and emotional arousal, which originally had continuous numerical values, are converted into a finite number of discrete grid positions, preventing the transmission of precise values ​​of continuous emotional state vectors during subsequent network synchronization.

[0116] In one embodiment, step S400, generating the discrete state feature code and synchronizing the discrete state feature code over the network, includes: Step S480: Generate a combined index to characterize the discrete grid position based on the first grid index and the second grid index; Step S490: Encode the combined index into the discrete state feature code; Step S491: Transmit the discrete state feature code as network synchronization data, wherein the wearable biometric data, the touch interaction data, and the continuous emotional state vector are not transmitted during the network synchronization process.

[0117] In this embodiment, after obtaining the first grid index and the second grid index, the smart terminal generates a discrete state feature code based on the first grid index and the second grid index, and transmits the discrete state feature code as network synchronization data.

[0118] In step S480, the smart terminal maps the first grid index and the second grid index to a unique combined index according to a preset grid numbering rule. For example, the two-dimensional grid is numbered sequentially row by row or column by column starting from a preset starting position, so that each set of first grid indexes and second grid indexes has a corresponding combined index. Alternatively, the first grid indexes and the second grid indexes are concatenated according to a preset field order to form a corresponding combined index.

[0119] In step S490, the combined index is converted into a fixed-length discrete state feature code according to the preset encoding rules, so that the network receiver can identify the corresponding discrete state category or discrete grid position only based on the discrete state feature code, without having to obtain the wearable biometric data and touch interaction data on which the discrete state feature code is based.

[0120] For example, a fixed-length binary code can be pre-configured for each composite index, and the binary code corresponding to the target composite index can be determined as a discrete state feature code; or a mapping relationship between composite indexes and preset state codes can be established, and the preset state code corresponding to the current composite index can be determined as a discrete state feature code.

[0121] In step S491, the smart terminal establishes a network synchronization data packet and writes the discrete state feature code into the state data field corresponding to the network synchronization data packet. Then, the smart terminal sends the network synchronization data packet to the target network node through its network communication module. The target network node determines the corresponding discrete state based on the received discrete state feature code, but cannot directly obtain the user's original transient heart rate, heart rate variability time series index, three-axis acceleration, touch coordinates, touch pressure, swipe speed, or continuous emotional valence and emotional arousal from the discrete state feature code.

[0122] During the local processing stage of the smart terminal, a continuous emotional state vector with high state resolution is retained to meet the processing needs of the local downstream processing module. During the network synchronization stage, the state precision of the synchronized data is further reduced by discrete quantization and state coding, and only discrete state feature codes are sent, thereby reducing the transmission of raw wearable biological data, raw touch interaction data and continuous emotional state data to the outside of the smart terminal from the data processing link.

[0123] In one embodiment, a data desensitization processing system is also provided, such as... Figure 2 As shown, the system includes: The interactive data acquisition module is used to: acquire wearable-side biometric data generated by the wearable device and touch-side interactive data generated by the smart terminal locally on the smart terminal. The feature data generation module is used to perform timestamp correction on the wearable biological data and the touch interaction data based on a unified time source, and to perform time sequence alignment, resampling and abnormal data processing according to a sliding time window to obtain multi-source heterogeneous feature data. The state vector generation module is used to standardize the multi-source heterogeneous feature data and input the standardized multi-source heterogeneous feature data into the feature translation layer deployed on the smart terminal. Irreversible feature projection is performed through the feature translation layer to obtain a continuous emotional state vector. The discrete decryption processing module is used to provide the continuous emotion state vector to the local downstream processing module when the continuous emotion state vector is used for local processing on the smart terminal; when the continuous emotion state vector is used for network synchronization, the module performs discrete quantization processing on the continuous emotion state vector to generate a discrete state feature code, and then performs network synchronization on the discrete state feature code.

[0124] In one embodiment, the interactive data acquisition module is further configured to: establish a wearable-side data acquisition task for the wearable device and a touch-side data acquisition task for the smart terminal; acquire heart rate variability time-series indicators, transient heart rate, triaxial acceleration, step count data, and sleep stage data generated by the wearable device through the wearable-side data acquisition task to form wearable-side biological data; monitor touch interaction events generated by the smart terminal through the touch-side data acquisition task, and extract touch timestamps, touch coordinates, touch area, touch pressure, sliding speed, operation pause time, and trajectory path curvature from each touch interaction event to form touch-side interactive data; and write the wearable-side biological data and the touch-side interactive data into the local memory of the smart terminal as dual-track heterogeneous data to be time-aligned.

[0125] In one embodiment, the interactive data acquisition module is further configured to: acquire the wearable-side clock corresponding to the wearable device and the terminal-side clock corresponding to the smart terminal, and determine the basic latency offset and clock drift correction rate based on the timestamp interaction between the wearable device and the smart terminal; correct the original timestamp corresponding to the wearable-side biometric data according to the basic latency offset and the clock drift correction rate to obtain a corrected timestamp corresponding to the terminal-side clock; establish multiple sliding time windows on the time axis corresponding to the terminal-side clock according to a preset window length and a preset sliding step size; and divide the wearable-side biometric data and the touch-side interactive data into corresponding sliding time windows according to the corrected timestamp and the touch timestamp corresponding to the touch-side interactive data to obtain time-aligned dual-track data corresponding to each sliding time window.

[0126] In one embodiment, the feature data generation module is further configured to: establish multiple target sampling times arranged according to a preset time interval for a target sliding time window; determine the touch sampling state corresponding to each target sampling time based on the timestamps of adjacent real touch sampling points within the target sliding time window; when the time interval between two adjacent real touch sampling points does not exceed a preset maximum interpolation time interval, perform piecewise linear interpolation on the touch coordinates, touch pressure, and touch area corresponding to the real touch sampling points to obtain resampled touch data corresponding to each target sampling time; when no new data is generated by the wearable physiological indicators within the target sliding time window, adopt a forward zero-order hold method to fill the data of the corresponding target sampling time based on the most recent effective wearable physiological indicators to obtain resampled dual-track data.

[0127] In one embodiment, the feature data generation module is further configured to: acquire the touch sliding speed and transient heart rate in the resampled dual-track data respectively; compare the touch sliding speed with a preset touch speed threshold, and determine touch data exceeding the preset touch speed threshold as abnormal touch data; compare the transient heart rate with a preset effective heart rate range, and determine wearable-side biological data not within the preset effective heart rate range as abnormal physiological data; remove the abnormal touch data, and perform sliding median filtering on the abnormal physiological data to obtain the multi-source heterogeneous feature data.

[0128] In one embodiment, the state vector generation module is further configured to: vectorize and assemble wearable biometric data and touch interaction data corresponding to the same target sliding time window according to a preset feature order to obtain a joint input feature vector; determine the feature mean and feature standard deviation of each feature component in the joint input feature vector in the current processing window; and perform Z-score standardization on each feature component according to the feature mean and feature standard deviation of each feature component to obtain a standardized joint feature vector.

[0129] In one embodiment, the state vector generation module is further configured to: input the standardized joint feature vector into a feature translation layer deployed locally on the smart terminal; obtain a pre-fixed projection weight matrix and bias vector in the feature translation layer, wherein the projection weight matrix is ​​obtained by joint comparative training targeting wearable biometric features and touch-screen interaction features; perform feature projection on the standardized joint feature vector according to the projection weight matrix and the bias vector, and perform nonlinear activation processing on the projection result to obtain a low-dimensional mapping feature; determine the low-dimensional mapping feature as a continuous emotional state vector, wherein the continuous emotional state vector includes emotional valence and emotional arousal.

[0130] In one embodiment, the discrete decryption processing module is further configured to: determine the target position of the continuous emotional state vector in a preset two-dimensional emotional state space based on the emotional valence and the emotional arousal; determine the target emotional state corresponding to the target position based on a preset region division rule, and perform hysteresis comparison on the target emotional states corresponding to multiple consecutive sampling periods; generate corresponding local state output data when the target emotional state continuously meets preset state switching conditions; and provide the continuous emotional state vector and the local state output data to the local downstream processing module through local inter-process communication on the smart terminal.

[0131] In one embodiment, the discrete decryption processing module is further configured to: obtain the emotional valence and emotional arousal in the continuous emotional state vector; divide the continuous value range corresponding to the emotional valence and the continuous value range corresponding to the emotional arousal into multiple discrete intervals respectively; determine the corresponding first grid index and second grid index according to the discrete intervals to which the emotional valence and the emotional arousal belong respectively, and obtain the discrete grid position corresponding to the continuous emotional state vector.

[0132] In one embodiment, the discrete decryption processing module is further configured to: generate a combined index for characterizing the discrete grid position based on the first grid index and the second grid index; encode the combined index into the discrete state feature code; and transmit the discrete state feature code as network synchronization data, wherein the wearable biometric data, the touch interaction data, and the continuous emotional state vector are not transmitted during network synchronization.

[0133] It should be noted that the information interaction and execution process between the above modules are based on the same concept as the method embodiments of this application. For details on their specific functions and technical effects, please refer to the method embodiments section, which will not be repeated here.

[0134] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0135] This application also provides a network device, which includes: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the processor executes the computer program to implement the steps in any of the above method embodiments.

[0136] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0137] This application provides a computer program product that, when run on a mobile terminal, enables the mobile terminal to execute the steps described in the various method embodiments above.

[0138] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A data anonymization processing method, characterized in that, The method includes: The wearable biometric data generated by the wearable device and the touch-side interaction data generated by the smart terminal are acquired locally on the smart terminal. Based on a unified time source, the wearable biological data and the touch interaction data are timestamped and corrected. Then, time alignment, resampling and abnormal data processing are performed according to a sliding time window to obtain multi-source heterogeneous feature data. The multi-source heterogeneous feature data is standardized, and the standardized multi-source heterogeneous feature data is input into the feature translation layer deployed on the smart terminal. Irreversible feature projection is performed through the feature translation layer to obtain a continuous emotional state vector. When the continuous emotion state vector is used for local processing on a smart terminal, the continuous emotion state vector is provided to the local downstream processing module; when the continuous emotion state vector is used for network synchronization, the continuous emotion state vector is subjected to discrete quantization processing to generate a discrete state feature code, and the discrete state feature code is synchronized to the network.

2. The data anonymization processing method according to claim 1, characterized in that, The smart terminal locally acquires wearable-side biometric data generated by the wearable device and touch-side interaction data generated by the smart terminal, including: Establish wear-side data acquisition tasks for the wearable device and touch-side data acquisition tasks for the smart terminal; The wearable-side data acquisition task acquires time-series indicators of heart rate variability, transient heart rate, triaxial acceleration, step count data, and sleep stage data generated by the wearable device, forming wearable-side biological data. The touch-side data acquisition task listens to the touch interaction events generated by the smart terminal, and extracts touch timestamps, touch coordinates, touch area, touch pressure, sliding speed, operation pause time and trajectory path curvature from each touch interaction event to form touch-side interaction data; The wearable biometric data and the touch interaction data are written into the local memory of the smart terminal as dual-track heterogeneous data to be processed for timing alignment.

3. The data desensitization processing method according to claim 2, characterized in that, Based on a unified time source, timestamp correction is performed on the wearable biometric data and the touch-based interaction data, and time sequence alignment is performed according to a sliding time window, including: The wearable-side clock corresponding to the wearable device and the terminal-side clock corresponding to the smart terminal are obtained, and the basic delay offset and clock drift correction rate are determined based on the timestamp interaction between the wearable device and the smart terminal. Based on the basic delay offset and the clock drift correction rate, the original timestamp corresponding to the wearable biological data is corrected to obtain the corrected timestamp corresponding to the terminal clock. Based on the preset window length and preset sliding step size, multiple sliding time windows are established on the time axis corresponding to the clock on the terminal side. Based on the correction timestamp and the touch timestamp corresponding to the touch-side interaction data, the wearable biometric data and the touch-side interaction data are divided into corresponding sliding time windows to obtain time-aligned dual-track data corresponding to each sliding time window.

4. The data anonymization processing method according to claim 3, characterized in that, Resampling the time-aligned dual-track data includes: Multiple target sampling times are established for the target sliding time window, arranged according to a preset time interval; Based on the timestamps of adjacent real touch sampling points within the target sliding time window, determine the touch sampling state corresponding to each target sampling moment; When the time interval between two adjacent real touch sampling points does not exceed the preset maximum interpolation time interval, the touch coordinates, touch pressure and touch area corresponding to the real touch sampling points are piecewise linearly interpolated to obtain the resampled touch data corresponding to each target sampling time. When no new data is generated for the wearable physiological indicators within the target sliding time window, a forward zero-order hold method is adopted to fill the data at the corresponding target sampling time with the most recent valid wearable physiological indicators to obtain resampled dual-track data.

5. The data desensitization processing method according to claim 4, characterized in that, The resampled dual-track data undergoes anomaly processing, including: The touch swipe speed and transient heart rate in the resampled dual-track data are obtained respectively; The touch sliding speed is compared with a preset touch speed threshold, and touch data that exceeds the preset touch speed threshold is identified as abnormal touch data. The transient heart rate is compared with a preset effective heart rate range, and wearable-side biodata that is not within the preset effective heart rate range is identified as abnormal physiological data. The abnormal touch data is removed, and the abnormal physiological data is subjected to sliding median filtering to obtain the multi-source heterogeneous feature data.

6. The data anonymization processing method according to claim 1, characterized in that, The standardization process for the multi-source heterogeneous feature data includes: The wearable biological data and touch interaction data corresponding to the same target sliding time window are vectorized and assembled according to a preset feature order to obtain a joint input feature vector. Determine the feature mean and feature standard deviation of each feature component in the joint input feature vector within the current processing window; Based on the feature mean and feature standard deviation corresponding to each feature component, Z-score standardization is performed on each feature component to obtain a standardized joint feature vector.

7. The data desensitization processing method according to claim 6, characterized in that, Irreversible feature projection is performed through the feature translation layer to obtain a continuous emotional state vector, including: The standardized joint feature vector is input into the feature translation layer deployed locally on the smart terminal. Obtain the pre-fixed projection weight matrix and bias vector in the feature translation layer, wherein the projection weight matrix is ​​obtained by joint comparative training targeting wearable biometric features and touch-screen interaction features; The standardized joint feature vector is projected based on the projection weight matrix and the bias vector, and the projection result is subjected to nonlinear activation processing to obtain low-dimensional mapping features. The low-dimensional mapping features are defined as continuous emotional state vectors, wherein the continuous emotional state vectors include emotional valence and emotional arousal.

8. A data anonymization processing system, characterized in that, The system includes: An interactive data acquisition module is used to acquire wearable-side biometric data generated by the wearable device and touch-side interactive data generated by the smart terminal locally. The feature data generation module is used to perform timestamp correction on the wearable biological data and the touch interaction data based on a unified time source, and to perform time sequence alignment, resampling and abnormal data processing according to a sliding time window to obtain multi-source heterogeneous feature data. The state vector generation module is used to standardize the multi-source heterogeneous feature data and input the standardized multi-source heterogeneous feature data into the feature translation layer deployed on the smart terminal. Irreversible feature projection is performed through the feature translation layer to obtain a continuous emotional state vector. The discrete decryption processing module is used to provide the continuous emotion state vector to the local downstream processing module when the continuous emotion state vector is used for local processing on the smart terminal; when the continuous emotion state vector is used for network synchronization, the module performs discrete quantization processing on the continuous emotion state vector to generate a discrete state feature code, and then performs network synchronization on the discrete state feature code.

9. A computer terminal, comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images