User value evaluation method based on cross-application data analysis

Abstract

The invention provides a user value evaluation method based on cross-application data analysis, and the method comprises the steps: constructing a user behavior track matrix according to the switching sequence data of a user among a starter application, a keyboard application and a camera application, and obtaining an original behavior data set containing an application identifier, a time interval and a use frequency; calculating a user cross-application migration value by adopting a real-time weight distribution scheme, and integrating contribution degree scores, calculated on the basis of the use frequency, the staying duration and the switching efficiency, of the applications in a weighted summation mode; and establishing a mapping relationship between the user behavior pattern and the weight configuration, and calling the corresponding configuration when a similar behavior pattern is detected to form a dynamic value evaluation system adaptive to the user behavior change.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to a user value assessment method based on cross-application data analysis. Background Technology

[0002] In the mobile application ecosystem, user behavior migration value assessment has become a core indicator for measuring the synergistic effect of product matrices (i.e., the phenomenon where multiple application products cooperate and promote each other, generating an overall benefit greater than the sum of the individual benefits of each application), directly impacting the accuracy of product strategy formulation and resource allocation optimization. Given the importance of this indicator, in tool application clusters, user cross-application usage patterns contain enormous commercial value and user insight potential. Accurately assessing this migration value is crucial for enhancing the overall competitiveness of the product ecosystem, as accurate value assessment can guide product optimization direction, resource allocation focus, and user retention strategies, thereby gaining a differentiated advantage in fierce market competition. However, current mainstream user behavior analysis methods generally adopt static weight allocation mechanisms, failing to effectively capture the dynamic changes in user usage patterns. At the same time, existing assessment systems often treat each application as an independent entity, ignoring the complex relationships and mutual influences between applications, leading to significant deviations between value assessment results and actual user behavior. In reality, as users migrate from launcher apps to core tools like keyboard and camera apps, their behavioral patterns exhibit high dynamism and interconnectedness. Changes in user frequency trigger a chain reaction, altering not only the intensity of individual app usage but also reshaping the entire app usage sequence, forming new behavioral path patterns. This path restructuring further impacts the switching time intervals between apps; users may adjust their overall operational rhythm due to changes in their usage habits of a particular app, leading to systemic changes in the original time interval patterns. How to construct a dynamic weight allocation mechanism that can respond in real-time to changes in user behavior, accurately capture the value flow patterns across app usage paths, and adjust evaluation parameters when user behavior patterns change to effectively conduct overall value assessment has become a critical issue that urgently needs to be addressed in the field of user behavior migration value assessment. Summary of the Invention

[0003] This invention provides a user value assessment method based on cross-application data analysis, mainly including:

[0004] Based on the user's switching sequence data between the launcher app, keyboard app, and camera app, a user behavior trajectory matrix is ​​constructed to obtain the raw behavior dataset containing app identifiers, time intervals, and usage frequencies.

[0005] The system calculates the time interval between adjacent application switching in the original behavior dataset. If the standard deviation of the time interval exceeds a preset threshold, it determines that the user behavior pattern has changed significantly and obtains the behavior pattern change identifier and the corresponding time node information. If the standard deviation of the time interval does not exceed the preset threshold, it maintains the current behavior pattern classification and continues to monitor subsequent usage data.

[0006] Based on behavioral pattern change identifiers, and using the usage frequency, switching probability, and dwell time of each application as inputs, an inter-application relationship model is established, and the dependency strength and influence coefficient of each application are output. Combined with time node information, the influence coefficient of each application node is calculated according to the magnitude of usage frequency change and the degree of path reconstruction, and an association weight matrix reflecting the mutual influence strength between applications is obtained.

[0007] The Markov chain algorithm is used to predict the probability distribution of users' subsequent application usage. The transmission impact on the probability of other applications is analyzed based on the rate of change of the probability of any application usage. By analyzing the connection strength and transmission path of each application node in the association weight matrix, the transmission path of high-frequency application to low-frequency application is determined.

[0008] Based on the degree of influence of each application node in the transmission path on the probability of using the target application, and combined with the scope of influence, the weight coefficients of each application in the value evaluation system are redistributed to obtain the timeliness characteristics of user behavior changes, the weight decay factor is dynamically adjusted, and a real-time weight allocation scheme that adapts to the current behavior pattern is obtained.

[0009] The value of users' cross-application migration is calculated using a real-time weighting scheme, and the contribution scores of each application are integrated by weighted summation based on usage frequency, dwell time, and switching efficiency.

[0010] Establish a mapping relationship between user behavior patterns and weight configurations. When similar behavior patterns are detected, the corresponding configuration is invoked to form a dynamic value assessment system that adapts to changes in user behavior.

[0011] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0012] This invention discloses a user value assessment method based on cross-application data analysis. It acquires user switching sequence data across multiple applications, constructs a behavior trajectory matrix, analyzes time interval distribution characteristics to identify behavioral pattern changes, establishes an application association model using a Bayesian network, combines Markov chains to predict usage probabilities, and dynamically adjusts application weight coefficients. Based on the timeliness of behavioral changes, this invention updates the weight allocation scheme in real time, recalculates the user's cross-application migration value, and stores the configuration in a model library to establish mapping relationships. By continuously optimizing the weight strategy and association parameters, this invention achieves a dynamic value assessment system that adapts to changes in user behavior, accurately capturing trends in user habits and providing data support for personalized service recommendations and user experience optimization. Attached Figure Description

[0013] Figure 1 This is a flowchart of a user value assessment method based on cross-application data analysis according to the present invention.

[0014] Figure 2 This is a schematic diagram of a user value assessment method based on cross-application data analysis according to the present invention. Detailed Implementation

[0015] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0016] like Figure 1 -2, This embodiment of a user value assessment method based on cross-application data analysis may specifically include:

[0017] S101. Based on the user's switching sequence data between the launcher application, keyboard application, and camera application, construct a user behavior trajectory matrix to obtain the original behavior dataset containing application identifiers, time intervals, and usage frequencies.

[0018] This process acquires user switching records between the launcher app, keyboard app, and camera app. It extracts the timestamp and app ID for each app launch, calculates the time interval between two consecutive switching operations, and determines the user's dwell time in the current app based on the time interval. Finally, it constructs a switching sequence data containing the app ID, launch timestamp, and dwell time. A launcher app is a core application on a mobile device (such as a smartphone or tablet) whose main function is to help users launch or open other applications.

[0019] Launchers are typically located on the device's home screen, providing users with an interface to access all installed applications. They are the first applications a user encounters when starting to use the phone, and can be the starting point for a series of tasks. Users can use launcher apps to view the application list, search for specific applications, or launch applications directly by clicking their icons.

[0020] For the switching sequence data, a fixed-duration time window is used to segment the data. Within each time window, the usage frequency of each application is counted. A switching path vector is constructed based on the order in which applications are switched, where each element represents the number of transitions from one application to another. A sliding time window mechanism is used to traverse the switching sequence data. Each slide retrieves the switching path vector and usage frequency statistics within the current window. The sum of the absolute differences in the number of transitions between pairs of the same application in adjacent windows is calculated as a continuity index. If the continuity index is less than a preset threshold, the switching path vectors and usage frequency data of adjacent windows are merged. A user behavior trajectory matrix is ​​constructed based on the merged data. The rows of the matrix represent the time window sequence number, and the columns represent the application identifier. The matrix element value equals the usage frequency of the corresponding application within that window multiplied by the average dwell time, then divided by the window duration, resulting in the original behavior dataset containing application identifiers, time intervals, and usage frequencies.

[0021] Specifically, data collection on user application switching behavior involves monitoring interactions between multiple applications.

[0022] In one possible implementation, when a user switches from the launcher app to the keyboard app, the system records the precise timestamp of the switch, such as 2:32:18 PM on March 15, 2024. It also records the duration the user spends on the keyboard app, for example, 45 seconds before switching to the camera app. This time interval is calculated by subtracting the previous switch time from the subsequent switch time, forming a complete switch sequence. Time window segmentation is a key technology for understanding user behavior patterns.

[0023] Specifically, each time window is set to 5 minutes, within which the usage of each application is statistically analyzed. If a user launches the keyboard application 3 times, the camera application 2 times, and the launcher application 1 time within 5 minutes, the usage frequency is recorded as 3, 2, and 1 respectively. A switching path vector records the transition relationships between applications; for example, the number of transitions from the launcher to the keyboard is 2, and the number of transitions from the keyboard to the camera is 3. These transition counts constitute the elements of the vector, reflecting the user's usage habits. The sliding time window mechanism can capture dynamic changes in user behavior.

[0024] For example, the first window covers 0-5 minutes, and the second window covers 1-6 minutes, with a 4-minute overlap between the windows. The continuity index is determined by calculating the change in the number of transitions between the same application pairs in adjacent windows. If the number of transitions from launcher to keyboard is 3 in the first window and 4 in the second window, the absolute difference is 1. The continuity index is obtained by summing the absolute differences of all application pairs. When this index is less than a preset threshold, such as 5, it indicates that user behavior is relatively stable, and the data from these two windows can be merged. The construction of the user behavior trajectory matrix transforms the time-series data into a structured representation.

[0025] In one embodiment, each row of the matrix represents a time window, and each column represents an application. The matrix element values ​​are calculated considering usage intensity, obtained by multiplying usage frequency by average dwell time and then dividing by window duration. If a keyboard application is used 3 times within a window, with an average dwell time of 40 seconds each time, and a window duration of 300 seconds, then the element value is 3 × 40 ÷ 300 = 0.4. This weighted calculation method considers both usage frequency and usage depth, more accurately reflecting the user's dependence on each application. The resulting raw behavior dataset contains multi-dimensional information such as application identifiers, time intervals, and usage frequency, providing a solid data foundation for subsequent user behavior analysis.

[0026] S102. Statistically analyze the time interval between adjacent application switching in the original behavior dataset. If the standard deviation of the time interval exceeds a preset threshold, it is determined that the user behavior pattern has changed significantly. Obtain the behavior pattern change identifier and the corresponding time node information. If the standard deviation of the time interval does not exceed the preset threshold, maintain the current behavior pattern classification and continue to monitor subsequent usage data.

[0027] Interval data of adjacent application switching are extracted from the original behavior dataset. The sum of all time intervals is divided by the number of intervals to obtain the mean. The variance is obtained by summing the squares of the differences between each time interval and the mean and dividing by the number of intervals. The standard deviation is obtained by taking the square root of the variance, forming a statistical feature set of time intervals. Based on the mean and standard deviation in the statistical feature set, the distribution intervals of time intervals are determined. The frequency of time intervals in each interval is counted, and the proportion of the frequency to the total is calculated as the distribution probability. If there are multiple probability peaks and the peak intervals show a regularity, a periodic change pattern is identified, and the peak positions and intervals are recorded as periodic parameters. The standard deviation is compared with a preset threshold, and the changes in the periodic parameter are also considered. If the standard deviation exceeds the preset threshold or the periodic parameter changes abruptly, a significant change in user behavior pattern is judged, and a behavior pattern change identifier containing the standard deviation value, periodic parameter, and timestamp is generated. If the standard deviation does not exceed the preset threshold and the periodic parameter is stable, the current behavior pattern classification is maintained, and the time interval distribution features in the behavior trajectory matrix are obtained.

[0028] Specifically, extracting statistical features of time intervals is fundamental to understanding patterns in user behavior.

[0029] In one possible implementation, when the system records time intervals of 2 seconds for a user to switch from the launcher to the keyboard, 5 seconds for switching from the keyboard to the camera, and 3 seconds for returning from the camera to the launcher, these time intervals constitute the raw material for analysis. The mean is calculated by summing all intervals and dividing by the total number of intervals; for example, if the sum of 100 switching intervals is 400 seconds, the mean is 4 seconds. Variance reflects the dispersion of time intervals; it is calculated by squaring the differences between each interval and the mean, summing the results, and then averaging them. A larger variance indicates more irregular user behavior. The standard deviation, as the square root of the variance, provides a measure of dispersion with the same dimensions as the original data. This set of statistical characteristics lays the numerical foundation for subsequent distribution analysis.

[0030] Specifically, using a mean of 4 seconds and a standard deviation of 1.5 seconds, it can be determined that most time intervals fall within the range of 2.5 to 5.5 seconds. The system divides the entire time range into multiple intervals, such as 0-2 seconds, 2-4 seconds, and 4-6 seconds, and counts the number of intervals within each interval. If there are 20 intervals in the 0-2 second interval, 60 intervals in the 2-4 second interval, and 15 intervals in the 4-6 second interval, the corresponding probability distributions are 0.2, 0.6, and 0.15, respectively. This probability distribution directly reflects the user's time preference when switching applications. The identification of periodic changes reveals deep patterns in user behavior.

[0031] For example, analysis revealed a significant increase in the frequency of app switching between 8-9 AM and 8-9 PM, forming two probability peaks. These peaks are 12 hours apart, exhibiting a clear periodicity. The system recorded the first peak at 8.5 hours and the second at 20.5 hours, with the 12-hour interval serving as a periodic parameter. This periodicity reflects users' different usage habits during commutes and rest periods. The mechanism for detecting changes in behavioral patterns ensures timely awareness of changes in user habits.

[0032] In one embodiment, the system sets a standard deviation threshold of 2 seconds. When the standard deviation for a certain time period reaches 3.5 seconds, exceeding the threshold, it indicates abnormal fluctuations in user behavior. Simultaneously, the previously stable 12-hour cycle suddenly changes to 8 hours, representing a significant change in the cycle parameter. The behavior pattern change identifier generated by the system includes the standard deviation of 3.5 seconds, the new cycle parameter of 8 hours, and the detection timestamp of March 20, 2024, at 15:30. This information collectively constitutes the time interval distribution characteristics in the behavior trajectory matrix, providing a quantitative basis for subsequent user behavior prediction and personalized services.

[0033] S103. Based on behavioral pattern change identifiers, using the usage frequency, switching probability, and dwell time of each application as inputs, establish an inter-application relationship model, output the dependency strength and influence coefficient between each application, combine time node information, calculate the influence coefficient of each application node according to the magnitude of usage frequency change and the degree of path reconstruction, and obtain the association weight matrix reflecting the mutual influence strength between applications.

[0034] The dynamic weight adjustment process is initiated based on behavioral pattern change identifiers. Data on usage frequency, number of app switches, and dwell time per use for each application are extracted to construct a Bayesian network. Nodes represent applications, and directed edges represent the relationship between switching from one application to another. The switching probability of each edge is calculated as the initial weight. Using the initial weights and usage frequency data of the Bayesian network, the conditional probability of switching to other applications while using one application is calculated. Multiplying the conditional probability by the average dwell time of the target application yields the dependency strength value between applications. Based on the dependency strength value and time node information, the change in dependency strength within adjacent time periods is calculated as the magnitude of change. The number of newly added or disappeared edges in the application switching path is counted as the path reconstruction degree. Multiplying the magnitude of change by the path reconstruction degree yields the dynamic adjustment factor. The influence coefficient of each application node is calculated by multiplying the dependency strength value by the dynamic adjustment factor. An association weight matrix is ​​constructed based on the influence coefficients, where each element equals the product of the influence coefficients of the corresponding two applications, resulting in an association weight matrix reflecting the strength of mutual influence between applications.

[0035] Specifically, Bayesian networks play a central role in modeling application relationships.

[0036] In one possible implementation, when a change in user behavior pattern is detected, the system begins to construct a network structure reflecting the relationships between applications. Each application acts as a node in the network; for example, the launcher, keyboard, and camera each constitute three independent nodes. Directed edges between nodes represent the user's behavioral path when switching from one application to another. If a user switches from the launcher to the keyboard 15 times and from the launcher to the camera 5 times within an hour, the probability of switching from the launcher to the keyboard is 0.75, and the probability of switching to the camera is 0.25. These probability values ​​become the initial weights of the corresponding edges. The calculation of conditional probabilities and dependency strength reveals the deep connections between applications.

[0037] Specifically, the system analyzes the probability distribution of users switching to other applications while using the keyboard application. If there is an 80% probability of switching to the camera after using the keyboard, and the average time a user spends in the camera application is 120 seconds, then the dependency strength value from keyboard to camera is calculated as 0.8 multiplied by 120, which equals 96. This value reflects the strength of the triggering effect of the keyboard application on the camera application; the higher the value, the stronger the usage association between the two applications. The introduction of a dynamic adjustment factor allows the weights to adapt to changes in user behavior.

[0038] For example, the system compares the dependency strength values ​​at 8 AM and 2 PM, finding that the dependency strength from keyboard to camera changed from 96 to 48, a change of 48. Simultaneously, it was observed that the original launcher-to-keyboard switching path disappeared, while a new reverse path from keyboard to launcher was added, with the path reconstruction degree recorded as 2. The dynamic adjustment factor equals 48 multiplied by 2, resulting in 96. This factor reflects the intensity of user behavior patterns. The construction of the association weight matrix completes the transformation from behavioral data to a structured representation.

[0039] In one embodiment, the influence coefficient of the keyboard application is obtained by multiplying its dependency strength value (96) by a dynamic adjustment factor (96) and then dividing by a normalization coefficient. Assuming the keyboard's influence coefficient is 0.8 and the camera's influence coefficient is 0.6, then the element value from keyboard to camera in the weight matrix is ​​0.8 multiplied by 0.6, which equals 0.48. This matrix comprehensively depicts the mutual influence relationships between applications, with each element quantifying the influence strength between the corresponding two applications. In this way, the association patterns of users using different applications can be accurately grasped, providing a quantitative basis for subsequent behavior prediction and personalized recommendations.

[0040] S104. Use the Markov chain algorithm to predict the probability distribution of users' subsequent application usage. Analyze the transmission impact on the probability of other applications based on the rate of change of the probability of any application usage. By analyzing the connection strength and transmission path of each application node in the correlation weight matrix, determine the boundary of the influence range of high-frequency applications on low-frequency applications.

[0041] A Markov chain state transition matrix is ​​constructed based on the association weight matrix, where matrix elements represent the probability of transitioning from the current application state to the next application state. Multi-step transition probabilities are obtained by iteratively calculating the power of the state transition matrix, yielding the application usage probability distribution for each subsequent time step. Using this application usage probability distribution, the probability difference between adjacent time steps for any application is calculated as the rate of change. This rate of change is multiplied by the elements of the corresponding row in the association weight matrix to obtain the transmission impact value of the application probability change on other applications. Application pairs with influence exceeding a preset threshold are identified based on the transmission impact value. The connections between these application pairs are defined as strong connections. Starting from the application with the highest usage frequency, the path is traced along the strong connections to the application with the lowest usage frequency. All intermediate applications and connections are recorded, forming the transmission path for each application node, thus determining the influence path of high-frequency applications on low-frequency applications. The end-to-end transmission strength is calculated by multiplying the weight values ​​of each segment on the influence path. A preset attenuation coefficient is applied to each intermediate application node. When the cumulative transmission strength falls below a preset threshold, the current position is recorded as the influence boundary. The number of hops from the starting application to the boundary is counted to determine the effective range boundary of the influence transmission.

[0042] Specifically, the core role of Markov chains in application usage prediction lies in the calculation of state transition probabilities.

[0043] In one possible implementation, the process of transforming the association weight matrix into a state transition matrix involves normalization. Assuming the weight from the launcher to the keyboard is 0.6, to the camera is 0.3, and to itself is 0.1, the sum of the elements in the corresponding row is 1.0, and these values ​​are directly used as transition probabilities. By calculating the square of the matrix, the two-step transition probability can be obtained; for example, the probability that a user will use the keyboard after two switches, starting from the launcher, is 0.45. Multi-step prediction provides insights into long-term user behavior trends. The dynamic changes in application usage probabilities reveal the evolution of user behavior.

[0044] Specifically, the probability of keyboard usage was detected as 0.4 at 10:00 AM, rising to 0.6 by 10:30 AM, a change rate of 0.4 per hour. This rate of change is multiplied by the element in the row containing the keyboard in the association weight matrix. For example, if the weight from the keyboard to the camera is 0.5, the resulting impact value is 0.4 multiplied by 0.5, which equals 0.2. This means that an increase in the probability of keyboard usage will lead to an increase of 0.2 in the probability of camera usage per hour, demonstrating the linkage effect between applications. The identification of strong connections helps the system focus on the most important application associations.

[0045] For example, when the influence propagation value exceeds a threshold of 0.15, the corresponding application pair is marked as a strong connection. In a real-world scenario, the launcher, being the most frequently used application, is used 30 times per hour, while a certain professional editing application is used only twice per hour. Tracking revealed two strong connections from the launcher through the keyboard to the editing application, forming a clear influence path. This path reflects the user's usage pattern of gradually moving from general functions to professional functions. The boundary of influence propagation determines the effective distance of interaction between applications.

[0046] In one embodiment, when calculating the conduction strength along the influence path, the weight from the initiator to the keyboard is 0.6, the weight from the keyboard to the input method is 0.4, and the attenuation coefficient is 0.8 for each intermediate node. The conduction strength of the first hop is 0.6, and the weight of the second hop is 0.6 multiplied by 0.4 multiplied by 0.8, which equals 0.192. When the conduction strength drops below the threshold of 0.1, the system records the number of hops at this time as 3, determining the effective range boundary of the influence conduction as 3 hops. This boundary recognition mechanism avoids misjudging weak indirect influences as important associations, improving the accuracy of behavior analysis.

[0047] S105. Based on the degree of influence of each application node in the transmission path on the probability of using the target application, and combined with the boundary of the influence range, the weight coefficients of each application in the value assessment system are redistributed to obtain the timeliness characteristics of user behavior changes, the weight decay factor is dynamically adjusted, and a real-time weight allocation scheme that adapts to the current behavior pattern is obtained.

[0048] Based on the weight coefficients of each application node in the transmission path, the cumulative product from the source application to the target application is calculated. For each intermediate node, the corresponding weight coefficient is multiplied to obtain the influence level value of each application node on the probability of using the target application. A normalization coefficient is calculated by dividing the influence level value by the number of hops within the influence range boundary and adding one. The original weight coefficient is multiplied by the normalization coefficient to obtain the adjusted weight coefficient for each application. Based on the temporal changes of the adjusted weight coefficients, the difference between weight coefficients at adjacent time points is calculated. The pattern of the difference decreasing over time is statistically analyzed, and the time required for it to decrease to half of its initial value is taken as the duration. The decrease ratio per unit time is calculated as the decay rate. The adjustment coefficient of the weight decay factor is determined by the ratio of the decay rate to a preset threshold. The adjusted weight coefficient is multiplied by the time factor calculated based on the duration, and then multiplied by the weight decay factor to obtain a real-time weight allocation scheme adapted to the current behavior pattern.

[0049] Specifically, the cumulative product calculation plays a crucial role in assessing the degree of impact.

[0050] In one possible implementation, when a user travels from the launcher through the keyboard to the camera application, the transmission path contains two connections. The weight coefficient from the launcher to the keyboard is 0.7, and the weight coefficient from the keyboard to the camera is 0.6. Multiplying these two weights gives 0.42, which represents the degree of influence of the launcher on the probability of camera use. This cumulative effect reflects the principle that the indirect influence gradually weakens as the transmission distance increases. Normalization ensures the rationality of the weight adjustment.

[0051] Specifically, for an impact value of 0.42, if the transmission path contains two hops, the hop count is increased by one to equal 3, and the normalization coefficient is calculated as 0.42 divided by 3, resulting in 0.14. The original weight coefficient of 0.5 is multiplied by the normalization coefficient of 0.14, yielding an adjusted weight coefficient of 0.07. This calculation method considers both the impact intensity and the transmission distance, avoiding the problem of excessively high weights for long-distance applications. Time-series change analysis reveals the dynamic characteristics of user behavior.

[0052] For example, the adjustment weighting coefficient of an application is 0.08 in the first hour, drops to 0.06 in the second hour, and reaches 0.045 in the third hour, showing a decreasing trend. Analysis reveals that the hourly decrease rate is approximately 25%, which is the decay rate. When the weighting coefficient drops from 0.08 to 0.04, exactly halving, it takes 2.5 hours; this time length is the duration. These two parameters together characterize the timeliness of behavioral changes. The generation of the real-time weighting allocation scheme reflects a comprehensive consideration of multiple factors.

[0053] In one embodiment, the ratio of the decay rate of 0.25 to the preset threshold of 0.2 is 1.25, and this ratio serves as the adjustment coefficient for the decay factor. The timeliness coefficient is calculated based on a duration of 2.5 hours, using an exponential decay function, resulting in a timeliness coefficient of 0.8 in the first hour. The adjusted weight coefficient of 0.07 is multiplied by the timeliness coefficient of 0.8, and then multiplied by the adjusted decay factor of 0.9 to obtain the real-time weight of 0.0504 at that moment.

[0054] S106. The real-time weighting scheme is used to calculate the user's cross-application migration value, and the contribution scores calculated by each application based on usage frequency, dwell time and switching efficiency are integrated by weighted summation.

[0055] The system obtains a baseline value calculated based on default weights during system initialization. A real-time weight allocation scheme replaces the default weights. Frequency contribution is obtained by dividing the usage frequency of each application by the total usage frequency; duration contribution is obtained by dividing the average dwell time by the total dwell time; and switching efficiency contribution is obtained by normalizing the inverse of the average switching time between applications. The user's cross-application migration value is obtained by multiplying the frequency contribution, duration contribution, and switching efficiency contribution by their respective real-time weights and then summing them. This value is recorded and compared with the baseline value, and the percentage difference is calculated. The cumulative average migration value within past time windows is used as the historical average evaluation value. The percentage deviation between the current migration value and the historical average evaluation value is calculated. If the percentage deviation exceeds a preset tolerance threshold, the weight ratios are adjusted based on the differences between the contribution of each indicator and the average contribution. Indicators with smaller deviations are multiplied by an increase factor, and indicators with larger deviations are multiplied by a decrease factor. The adjusted weight ratios are then used to recalculate and determine the updated user behavior migration value evaluation result.

[0056] Specifically, the contribution calculation reflects the standardized processing of different dimensional indicators.

[0057] In one possible implementation, the system counts 300 times a user uses the launcher, 200 times the keyboard, and 100 times the camera in a day, for a total of 600 uses. The frequency contribution of the launcher is calculated as 300 divided by 600, resulting in 0.5; the keyboard contribution is 0.33; and the camera contribution is 0.17. Regarding dwell time, if a user spends a total of 1800 seconds on the launcher, 3600 seconds on the keyboard, and 900 seconds on the camera, for a total dwell time of 6300 seconds, the duration contributions of each application are 0.29, 0.57, and 0.14, respectively. The calculation of switching efficiency involves normalizing the reciprocals of time. Switching from the launcher to other applications takes an average of 2 seconds, with a reciprocal of 0.5; switching from the keyboard takes 3 seconds, with a reciprocal of 0.33; and switching from the camera takes 4 seconds, with a reciprocal of 0.25. Normalizing these reciprocals yields the switching efficiency contribution. The weighted calculation of migration value reflects a multi-dimensional comprehensive evaluation approach.

[0058] Specifically, assuming the launcher's real-time weight is 0.4, the keyboard's is 0.35, and the camera's is 0.25, the launcher's migration value is calculated as follows: frequency contribution (0.5) multiplied by weight 0.4, plus duration contribution (0.29) multiplied by weight 0.4, plus switching efficiency contribution (0.45) multiplied by weight 0.4, resulting in 0.496. The migration values ​​of other applications are calculated using the same method and then summed to obtain the total user cross-application migration value of 1.2. Compared to the baseline value of 1.0 calculated based on default weights during system initialization, the difference is 20%, revealing the degree of change in user behavior patterns. The introduction of historical average evaluation values ​​provides a dynamically adjusted reference benchmark.

[0059] For example, the system maintains a sliding time window, recording the migration value calculated daily over the past 7 days as 1.1, 1.15, 1.12, 1.18, 1.14, 1.16, and 1.13, with a cumulative average of 1.14. The current calculated migration value of 1.2 deviates from the historical average of 1.14 by 5.3%. If the preset tolerance threshold is 10%, the current deviation is within an acceptable range and no adjustment is needed. The dynamic adjustment mechanism of the weight ratio ensures the stability of the evaluation.

[0060] In one embodiment, when the deviation exceeds a threshold, the system analyzes the deviation of the contribution of each indicator. The average contribution is 0.33, the frequency contribution is 0.5 (significant deviation), and the duration contribution is 0.29 (small deviation). During adjustment, the weight of the duration indicator is multiplied by an increase factor of 1.2, and the weight of the frequency indicator is multiplied by a decrease factor of 0.8. Through this differentiated adjustment, the recalculated migration value is closer to the historical average level, avoiding the excessive influence of abnormal fluctuations in a single indicator on the overall assessment, and achieving a robust assessment of user behavior migration value.

[0061] S107. Establish a mapping relationship between user behavior patterns and weight configurations. When similar behavior patterns are detected, the corresponding configuration is invoked to form a dynamic value assessment system that adapts to changes in user behavior.

[0062] Based on the updated user behavior migration value assessment results, the weight values ​​and corresponding contribution scores of each application in the real-time weight allocation scheme are extracted. These, combined with application identifiers and timestamps, form a configuration dataset, which is stored in the user behavior pattern historical model library, and a unique configuration identifier is generated. Behavioral feature vectors are constructed by extracting current usage frequency, dwell time, and switching paths. The cosine similarity between this vector and the corresponding behavioral feature vectors in the historical model library is calculated. When the similarity exceeds a preset threshold, a mapping relationship between the current behavior pattern and the historical weight configuration is established, and the configuration identifier is recorded as a mapping index. When a similar behavior pattern is detected, the corresponding weight configuration is retrieved from the historical model library based on the mapping index. This configuration is applied to predict the usage probability of each application. The sum of the absolute values ​​of the differences between the predicted probability and the actual usage frequency is calculated as the deviation value. If the deviation value is less than the expected standard, the configuration is retained; otherwise, the weight coefficients in the correlation parameters are increased or decreased accordingly based on whether the actual frequency is higher or lower than the predicted value. By recording the deviation value and adjustment direction after each configuration call, calculating the average adjustment range of the weight coefficients, correcting the corresponding weight coefficients and decay factors in the historical model library based on the adjustment range, updating the configuration data, and forming a dynamic value assessment system that is continuously optimized based on the verification results.

[0063] Specifically, the construction of the configuration dataset reflects the quantitative storage requirements of behavioral patterns.

[0064] In one possible implementation, the system extracts the initiator weight value (0.4), keyboard weight value (0.35), and camera weight value (0.25), corresponding to contribution scores of 0.5, 0.33, and 0.17, respectively. These values ​​are combined with application identifiers A001, A002, and A003, along with the timestamp 2024-03-20-15:30:00, to form a complete configuration record. The system generates a unique identifier CFG-20240320-001 for this record and stores it in the historical model library. This structured storage method allows for rapid retrieval and reuse of historical behavior patterns. Constructing behavioral feature vectors is a crucial step in achieving pattern matching.

[0065] Specifically, the launcher was used 300 times, the keyboard 200 times, and the camera 100 times during the current time period, forming a usage frequency vector [300, 200, 100]. The dwell time vector is [1800, 3600, 900], representing the cumulative number of seconds spent in each application. The switching path is formed by statistically analyzing the number of transitions between applications, such as 50 times from launcher to keyboard and 30 times to camera, constituting a switching feature. These three types of features are standardized and concatenated to form a comprehensive behavioral feature vector. Cosine similarity calculation reveals the degree of similarity in behavioral patterns.

[0066] For example, the current behavior feature vector is compared with the feature vector corresponding to the historical record CFG-20240319-003. A similarity of 0.92 is obtained by calculating the inner product of the two vectors and dividing by the product of their respective magnitudes. A threshold of 0.85 is set; if the current similarity exceeds the threshold, it indicates that the two behavior patterns are highly similar. The system establishes a mapping relationship, associating the current behavior pattern with the configuration identifier CFG-20240319-003, facilitating rapid retrieval of historically optimal configurations. The prediction and verification mechanism ensures the effectiveness of the configuration.

[0067] In one embodiment, the system calls historical configuration CFG-20240319-003 and predicts the launcher usage probability to be 0.5, keyboard usage to be 0.35, and camera usage to be 0.15 within the next hour based on its weight parameters. Actual monitoring revealed that the launcher usage frequency was 0.48, keyboard usage to 0.38, and camera usage to 0.14. The calculated deviation was |0.5-0.48|+|0.35-0.38|+|0.15-0.14|=0.06. The expected standard was set to 0.1; the current deviation was less than the standard, indicating the configuration was valid. If the actual keyboard frequency of 0.38 was higher than the predicted value of 0.35, the system adjusted the keyboard weight coefficient from 0.35 to 0.36. This dynamic optimization process demonstrates the system's adaptive capability. By recording multiple verification results, such as an average increase of 0.02 in the keyboard weight over five consecutive calls, the system identified a stable adjustment trend. Based on this trend, the keyboard weight of CFG-20240319-003 in the historical model library was updated to 0.37, and the decay factor was adjusted to reflect the persistence of behavioral changes. This continuous optimization mechanism based on actual feedback enables the value assessment system to keep up with the evolution of user behavior and provide more accurate application usage predictions and resource allocation suggestions.

[0068] The above description is merely a preferred embodiment of one or more embodiments of this specification and is not intended to limit the scope of one or more embodiments of this specification. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of one or more embodiments of this specification should be included within the protection scope of one or more embodiments of this specification.

Claims

1. A user value assessment method based on cross-application data analysis, characterized in that, The method includes: An original behavior dataset is constructed based on the user's switching sequences between various applications. By statistically analyzing the time intervals between switching between adjacent applications in the original behavior dataset, a time interval distribution feature is generated to identify the periodic variation pattern of the original behavior dataset. Based on the comparison of the time interval standard deviation with a preset threshold, behavior pattern change identifiers and time node information are generated. A dynamic weight adjustment mechanism is triggered by the behavior pattern change identifiers. A Bayesian network is used to construct an inter-application relationship model. Using the usage frequency, switching probability, and dwell time of each application as input, the dependency strength and influence coefficient between applications are generated. Combined with the time node information, an association weight matrix reflecting the mutual influence strength between applications is generated. A Markov chain algorithm is used to pre-... The system measures the probability distribution of subsequent application usage by users, determines the transmission path of each application node in the association weight matrix, and identifies the boundary of the influence range of high-frequency applications on low-frequency applications. Based on the degree of influence of each application node in the transmission path on the probability of using the target application, and combined with the boundary of the influence range, a real-time weight allocation scheme for each application is generated. Using the real-time weight allocation scheme, the contribution scores of each application based on usage frequency, dwell time, and switching efficiency are integrated through a weighted summation method to generate the user's cross-application migration value. The weight configuration and value assessment parameters of the real-time weight allocation scheme are stored in the user behavior pattern historical model library to establish a mapping relationship between behavior patterns and weight configuration, forming a dynamic value assessment system.

2. The user value assessment method based on cross-application data analysis according to claim 1, characterized in that, The process of constructing the original behavior dataset based on the user's switching sequence across applications includes: Extract the timestamp and application identifier of each application launch from the switching sequence data, calculate the time interval between two adjacent switching operations, and generate the dwell time of each application; segment the switching sequence data using a fixed time window, count the usage frequency of each application within each time window, and generate a switching path vector, where the elements of the switching path vector represent the number of transitions from one application to another; by sliding the time window, calculate the sum of the absolute values ​​of the differences in the number of transitions between pairs of the same application in adjacent windows to generate a continuity index; compare the continuity index with a preset threshold, and merge the switching path vectors and usage frequency data of adjacent windows; generate the user behavior trajectory matrix based on the merged data to obtain the original behavior dataset.

3. The user value assessment method based on cross-application data analysis according to claim 1, characterized in that, The step of generating time interval distribution features by statistically analyzing the time intervals between adjacent application switching in the original behavior dataset and identifying the periodic variation patterns of the original behavior dataset includes: Extract the time interval data of adjacent application switching from the original behavior dataset, calculate the mean and variance of all time intervals, and generate a statistical feature set; determine the distribution interval of the time interval based on the mean and variance in the statistical feature set, count the frequency of the time interval within each interval, and generate the distribution probability; identify the periodic change pattern based on the distribution probability, record the probability peak position and interval, and generate periodic parameters.

4. The user value assessment method based on cross-application data analysis according to claim 1, characterized in that, The dynamic weight adjustment mechanism triggered by the behavioral pattern change identifier uses a Bayesian network to construct an inter-application relationship model. Taking the usage frequency, switching probability, and dwell time of each application as input, it generates the dependency strength and influence coefficient between applications. Combined with the time node information, it generates a correlation weight matrix reflecting the strength of mutual influence between applications, including: Based on the behavioral pattern change identifiers, the usage frequency, number of switches, and dwell time data of each application are extracted, and a Bayesian network is constructed, where nodes represent applications and directed edges represent switching relationships. The switching probability of each edge is calculated. Based on the switching probability and usage frequency, a dependency strength value between applications is generated. Based on the dependency strength value and the time node information, the change in dependency strength and the degree of path reconstruction within adjacent time periods are calculated, and a dynamic adjustment factor is generated. Based on the dependency strength value and the dynamic adjustment factor, the influence coefficient of each application node is generated, and the association weight matrix is ​​constructed, where the element value is the product of the influence coefficients of two applications.

5. The user value assessment method based on cross-application data analysis according to claim 1, characterized in that, The step of using the Markov chain algorithm to predict the probability distribution of subsequent application usage by users and determining the transmission path of each application node in the association weight matrix includes: A state transition matrix is ​​generated based on the association weight matrix, multi-step transition probabilities are calculated, and the application usage probability distribution is generated. Based on the application usage probability distribution, the probability difference between adjacent time points is calculated, and the transmission influence value of each application is generated. The transmission path of each application node in the association weight matrix is ​​determined based on the transmission influence value.

6. The user value assessment method based on cross-application data analysis according to claim 1, characterized in that, The step of generating a real-time weight allocation scheme for each application based on the degree of influence of each application node in the transmission path on the probability of using the target application, combined with the boundary of the influence range, includes: Calculate the cumulative product of the weight coefficients from the source application to the target application in the transmission path to generate an influence level value; generate a normalization coefficient based on the influence level value and the number of hops within the influence range boundary; adjust the weight coefficients of each application based on the normalization coefficient to generate the real-time weight allocation scheme.

7. The user value assessment method based on cross-application data analysis according to claim 1, characterized in that, The aforementioned real-time weight allocation scheme integrates the contribution scores of each application based on usage frequency, dwell time, and switching efficiency through a weighted summation method to generate user cross-application migration value, including: Based on the real-time weight allocation scheme, the contribution scores of usage frequency, average dwell time, and switching efficiency of each application are calculated; by multiplying the contribution scores by their corresponding weights and summing them, a preliminary user cross-application migration value is generated; based on the deviation between the user cross-application migration value and the historical average evaluation value, the weight ratio of each indicator is adjusted to generate user behavior migration value.

8. The user value assessment method based on cross-application data analysis according to claim 1, characterized in that, The step of storing the weight configuration and value assessment parameters of the real-time weight allocation scheme into a user behavior pattern historical model library, establishing a mapping relationship between behavior patterns and weight configurations, and forming a dynamic value assessment system includes: Extract the weight values ​​and contribution scores of each application in the real-time weight allocation scheme, combine them with application identifiers and timestamps to generate a configuration dataset, and store it in the user behavior pattern historical model library; establish a mapping relationship between the behavior pattern and the weight configuration by calculating the similarity between the current behavior feature vector and the historical configuration; call the corresponding configuration according to the mapping relationship, predict the usage probability of each application, and calculate the prediction deviation; adjust the weight coefficient and decay factor according to the prediction deviation, update the configuration data in the user behavior pattern historical model library, and generate the dynamic value evaluation system.

Images