A method and system for dynamic task composition based on user digital profiles

Abstract

This invention relates to the field of big data analytics, and discloses a method and system for dynamic task combination based on user digital profiles. The method includes: acquiring historical user behavior data, extracting preference features, performing noise injection processing to obtain a noise-enhanced profile, and performing desensitization and encryption processing to obtain an encrypted noise profile; performing clustering operations to form preliminary user groups; if the ambiguity of the group boundaries exceeds a preset threshold, injecting compensation noise for iterative verification until a group with clear boundaries is obtained; determining the parameter set and reconfiguring the clustering; if the number of iterations is less than a preset threshold, fusing the compensation noise to obtain the final user group; performing a second noise injection processing to obtain a group with minimized bias and evaluating it; if the integrity index meets a preset index threshold, performing bias analysis and noise injection processing again to obtain the final enhanced group. This invention can improve the privacy protection effect and adaptability in big data analytics.

Description

Technical Field

[0001] This invention relates to the field of big data analytics, and in particular to a method and system for dynamic task combination based on user digital profiles. Background Technology

[0002] With the rapid development of the internet and mobile devices, user behavior data is experiencing explosive growth, making user profiling a key technology for personalized services and accurate recommendations. Task combination methods based on user digital profiles can enhance user experience and system efficiency by analyzing users' historical behavior and preference characteristics through reinforcement learning and dynamically adjusting service content and recommendation strategies.

[0003] Traditional techniques primarily involve collecting users' historical behavior data, extracting user preference features, and using desensitization tools to anonymize sensitive information. Subsequently, a static K-means clustering algorithm is used to cluster the feature vectors, forming initial user groups and their boundaries. During the clustering process, the group boundaries are adjusted based on the degree of boundary ambiguity, and the accuracy of the groups is evaluated through bias analysis. Finally, personalized recommendations or task assignments are made based on the group results.

[0004] However, in real-world business scenarios, traditional technologies typically rely on simple noise injection and desensitization for privacy protection. These methods cannot dynamically adjust noise intensity to adapt to changes in user behavior, leading to blurred user characteristics and unclear cluster boundaries. Furthermore, due to the lack of in-depth encryption mechanisms for sensitive features, leaked data can potentially be recovered and restored to its original form. This results in existing technologies exhibiting insufficient privacy protection, inadequate clustering accuracy, and poor adaptability in real-world business scenarios. Summary of the Invention

[0005] This invention provides a dynamic task combination method and system based on user digital profiles to solve the problems of insufficient privacy protection, insufficient clustering accuracy, and poor adaptability of existing technologies in actual business scenarios.

[0006] In a first aspect, to address the aforementioned technical problems, this invention provides a dynamic task combination method based on user digital profiles, comprising:

[0007] Obtain historical user behavior data, extract preference features, and form a preliminary user profile;

[0008] Based on the initial user profile, noise injection processing is performed to obtain a noise-enhanced profile, which is then desensitized and encrypted to obtain an encrypted noise profile.

[0009] Clustering is performed on the encrypted noise profile to form preliminary user groups;

[0010] If the degree of boundary ambiguity of the initial user group exceeds the preset boundary ambiguity threshold, compensation noise is injected for iterative verification until a group with clear boundaries is obtained. Then, delay-related parameters are extracted to determine the low-latency parameter set.

[0011] The clustering is reconfigured based on the low-latency parameter set. If the number of iterations for verification is less than a preset verification threshold, the noise is fused and compensated to obtain the final user group.

[0012] The end-user group is subjected to secondary noise injection processing to obtain a group with minimized deviation, which is then evaluated to obtain an integrity index.

[0013] If the integrity index meets the preset index threshold, additional deviation analysis and additional noise injection processing are performed to obtain the final enhanced group.

[0014] Secondly, the present invention provides a dynamic task combination system based on user digital profiles, comprising:

[0015] User data extraction module: acquires historical user behavior data, extracts preference features, and forms a preliminary user profile;

[0016] Noise injection protection module: Based on the preliminary user profile, perform noise injection processing to obtain a noise-enhanced profile, and perform desensitization and encryption processing to obtain an encrypted noise profile;

[0017] Clustering to form groups module: Performs clustering operations on the encrypted noise profile to form preliminary user groups;

[0018] Boundary blur correction module: If the boundary blur degree of the initial user group exceeds the preset boundary blur threshold, compensation noise is injected for iterative verification until a clear boundary group is obtained. Then, delay-related parameters are extracted to determine the low-latency parameter set.

[0019] Low-latency clustering module: Reconfigures clustering based on the low-latency parameter set; if the number of iterations for verification is less than a preset verification threshold, then merges and compensates for noise to obtain the final user group.

[0020] Secondary noise assessment module: Performs secondary noise injection processing on the end-user group to obtain the group with minimal deviation and performs assessment to obtain the integrity index;

[0021] Additional noise enhancement module: If the integrity index meets the preset index threshold, additional deviation analysis and additional noise injection processing are performed to obtain the final enhanced group.

[0022] Thirdly, the present invention also provides an electronic device including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the dynamic task composition method described in any one of the foregoing.

[0023] Fourthly, the present invention also provides a computer-readable storage medium comprising a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform the dynamic task composition method described in any one of the above.

[0024] Compared with the prior art, the present invention has the following beneficial effects:

[0025] (1) This invention injects noise into user historical behavior data through a differential privacy module to form a noise-enhanced profile, and combines this with desensitization and encryption to ensure multi-layered privacy protection for user data. After differential privacy noise injection and privacy protection mechanisms, the risk of privacy leakage of user profile data is significantly reduced; subsequently, these privacy-protected profiles are used for clustering to ensure data security while maintaining the basic accuracy of user group segmentation. This technology effectively solves the contradiction between user behavior data privacy protection and high-precision group segmentation, and realizes accurate digital profile generation under the premise of protecting user privacy.

[0026] (2) This invention addresses user groups formed by clustering by employing a boundary ambiguity judgment and iterative compensation noise injection mechanism, combined with low-latency parameter adjustment, to optimize the clarity and stability of group boundaries. Multiple iterative corrections are performed through noise compensation to adjust the feature distribution of boundary points, making group boundaries clearer. Simultaneously, a low-latency parameter set is extracted, and the clustering framework is reconfigured to accelerate the dynamic response of group partitioning. This dynamic optimization process significantly improves the accuracy and real-time performance of user group partitioning, reduces the uncertainty of group boundaries, ensures stable group structure and rapid response, and is beneficial for application in time-sensitive recommendation and task allocation systems.

[0027] (3) After optimizing the group boundaries, this invention performs secondary differential privacy noise injection to obtain a group with minimized deviation. Further deviation analysis and enhanced noise injection are then performed to output a final stable user group with strong privacy protection. By implementing secondary noise injection on the optimized user group, the impact of deviation is further reduced, and the group's integrity index is calculated to assess its stability. Groups meeting the stability threshold undergo additional noise injection and deviation verification to ensure that the group is not only stable but also meets high privacy protection standards. This technology ensures that the user group maintains good integrity and stability while possessing high privacy protection. Attached Figure Description

[0028] Figure 1 This is a schematic diagram of a dynamic task combination method based on user digital profiles provided in an embodiment of the present invention;

[0029] Figure 2 This is a schematic diagram of a dynamic task combination system module based on user digital profiles provided in an embodiment of the present invention. Detailed Implementation

[0030] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] Reference Figure 1 The first embodiment of the present invention provides a dynamic task combination method based on user digital profiles, comprising the following steps:

[0032] S101, Obtain user historical behavior data, extract preference features, and form a preliminary user profile;

[0033] S102, based on the preliminary user profile, perform noise injection processing to obtain a noise-enhanced profile, and perform desensitization and encryption processing to obtain an encrypted noise profile;

[0034] S103, perform clustering operation on the encrypted noise profile to form a preliminary user group;

[0035] S104, if the degree of boundary ambiguity of the initial user group exceeds the preset boundary ambiguity threshold, then inject compensation noise for iterative verification until a clear boundary group is obtained, then extract the delay-related parameters and determine the low-latency parameter set.

[0036] S105, reconfigure clustering according to the low latency parameter set; if the number of iterations for verification is less than a preset verification threshold, then fuse and compensate for noise to obtain the final user group.

[0037] S106, Perform secondary noise injection processing on the end-user group to obtain the deviation-minimized group and evaluate it to obtain the integrity index;

[0038] S107, If the integrity index meets the preset index threshold, then perform additional deviation analysis and additional noise injection processing to obtain the final enhanced group.

[0039] In step S101, user historical behavior data is acquired, preference features are extracted, and a preliminary user profile is formed, including:

[0040] S1011, Obtain user behavior logs and divide them according to the time dimension to obtain user historical behavior data;

[0041] S1012, Extract the time distribution features of user behavior from the user's historical behavior data to form a preference feature set;

[0042] S1013, Based on the preference feature set, the time distribution features and interaction frequency are weighted and fused to generate an initial user description structure and determine a preliminary user profile.

[0043] In step S1011, user behavior logs are obtained and divided by time dimension to obtain user historical behavior data.

[0044] In one implementation, user behavior logs are retrieved from a distributed database, in the form of {"user_id":"U001","action":"click","timestamp":"2023-05-01 14:30:45"}. Then, the time dimension is divided by hourly granularity (e.g., 14:30:45 is classified into the "14:00-15:00" time period), and the number of behaviors in each time period is counted (e.g., the number of clicks in the "14:00-15:00" time period is 120, accounting for 65% of the total clicks for the day). Structured historical behavior data is generated, in the form of {"peak_click":{"time":"14:00-15:00","count":120,"ratio":0.65}}).

[0045] In step S1012, the time distribution features of user behavior are extracted from the user's historical behavior data to form a preference feature set.

[0046] In one implementation, from the user's historical behavior data, in the form of {"peak_click":{"time":"14:00-15:00","count":120,"ratio":0.65}}, the frequency of behavior is statistically analyzed by hour (e.g., the number of clicks in the "14:00-15:00" time period is 120, accounting for 65% of the total clicks of the day). Through weighted calculation (the product of the time period weight and the normalized "ratio" plus the product of the interaction frequency weight and the normalized "count"), a time distribution feature value is generated, ultimately forming a structured preference feature set, in the form of {"peak_search":{"time":"14:00-15:00","score":0.772}}.

[0047] It should be noted that in the weighted calculation, the optimal weight combination is determined through grid search cross-validation. The specific steps are as follows: First, set the weight search range (e.g., time period weight [0.3, 0.7], step size 0.1). Second, use the click-through rate (CTR) or conversion rate of the recommendation system as the target indicator. Then, test different combinations on the training set to make the system's click-through rate as high as possible (e.g., the CTR of "0.6 + 0.4" reaches 23.5%, which is better than the 21.8% of "0.5 + 0.5"). Finally, combine the business scenario and historical data to determine the weight allocation in the weighted calculation (e.g., in the e-commerce scenario, the final decision is "time period 0.65 + interaction 0.35").

[0048] In step S1013, the time distribution features and interaction frequency are weighted and fused according to the preference feature set to generate an initial user description structure and determine a preliminary user profile.

[0049] In one implementation, the process of determining the initial user profile is as follows: First, obtain the time distribution features and interaction frequency data from the user preference feature set from a distributed database. For example, the time distribution weight of a user between 9:00 AM and 11:00 AM is 0.75, and the interaction frequency is 0.5. Then, set weighting coefficients, assuming the time distribution weight is 0.65 and the interaction frequency weight is 0.35, and calculate the initial user description structure through weighted fusion, i.e., the fusion value is 0.65×0.75+0.35×0.5=0.6375. Finally, use this initial description structure as the user's initial profile to accurately depict user behavior preferences and support subsequent personalized services and decisions.

[0050] It should be noted that the determination of weights here is consistent with the previous weight determination process in terms of overall approach. Both involve combining business scenarios and historical data, and adjusting and optimizing the weight ratios through expert guidance, experimental verification, and other methods. However, the specific weight values ​​and adjustment details will vary depending on different application backgrounds, data characteristics, and business needs.

[0051] In step S102, based on the preliminary user profile, noise injection processing is performed to obtain a noise-enhanced profile, followed by desensitization and encryption processing to obtain an encrypted noise profile, including:

[0052] S1021, Inject privacy-preserving noise into the preliminary user profile to obtain a noise-enhanced profile;

[0053] S1022, Anonymize the sensitive information in the noise-enhanced image to obtain a desensitized feature set;

[0054] S1023, adjust the encryption strength of the desensitized feature set and generate an encrypted noise profile.

[0055] In step S1021, privacy-preserving noise is injected into the preliminary user profile to obtain a noise-enhanced profile.

[0056] In one implementation, firstly, for the key feature values ​​of the initial user profile, such as a user's activity level of 0.65 at a certain time period, privacy protection is achieved by introducing Laplace noise. The privacy protection parameter ε=1.0 is set, and noise Δ is injected into this feature value. Δ is sampled from the Laplace distribution Lap(0,1 / ε). Assuming that the sampled noise is 0.12, the feature value after noise enhancement is 0.65+0.12=0.77. This process is repeated to inject independent noise into all sensitive features in the profile, generating a noise-enhanced profile.

[0057] In step S1022, the sensitive information in the noise-enhanced image is anonymized to obtain a desensitized feature set.

[0058] In one implementation, for sensitive information contained in the noise-enhanced profile, such as a user's activity feature value of 0.77 over a certain time period, this feature is first anonymized. Using the k-anonymity algorithm with k=3, the user's feature value is obfuscated with the feature values ​​of two other similar users (0.75 and 0.80 respectively). The mean of the three values ​​(0.77+0.75+0.80) / 3=0.7733 is calculated, and the original feature value is replaced, thus forming a de-anonymized feature. All sensitive features in the profile are processed in the same way, ultimately resulting in a de-anonymized feature set containing multiple anonymized feature values. It should be noted that, to further prevent inference attacks based on data distribution, after differential privacy protection, k-anonymity technology is additionally used to smooth the data distribution.

[0059] In step S1023, the encryption strength of the desensitized feature set is adjusted, and an encrypted noise profile is generated.

[0060] In one implementation, for a specific feature value of 0.7733 in the desensitized feature set, the encryption strength is first adjusted according to security requirements. Assuming the AES symmetric encryption algorithm is used, the encryption strength parameter is set to a 128-bit key. Then, during the encryption process, injected encryption noise is added, with a noise amplitude of 0.05, to perturb the desensitized feature value. This means the feature value is adjusted to a random value within the range of 0.7733 ± 0.05, assuming the perturbed value is 0.78. Finally, the adjusted feature value is encrypted using the encryption algorithm to generate an encrypted noise profile. This process is performed sequentially on all sensitive features in the desensitized feature set, ensuring that the profile possesses a certain degree of perturbation while maintaining data privacy, thereby improving security and resistance to attacks.

[0061] In step S103, the encrypted noise profile is clustered to form a preliminary user group, including:

[0062] S1031, initialize clustering parameters, perform clustering processing on the encrypted noise profile to obtain a preliminary user group.

[0063] In step S1031, clustering parameters are initialized, and the encrypted noise profile is clustered to obtain a preliminary user group.

[0064] In one implementation, the clustering parameters are first initialized, the clustering algorithm is set to K-means, the number of clusters K=3, and the initial cluster centers are the feature vectors of three typical users in the encrypted noise profile [0.75, 0.80, 0.70]. Then, the Euclidean distance from each encrypted noise profile sample to the cluster center is calculated, and the sample is assigned to the nearest cluster. The cluster center position is iteratively updated until the change in the cluster center is less than 0.01 or the maximum number of iterations of 100 is reached. Finally, based on the clustering results, the users are divided into three preliminary user groups. For example, the first user group contains 45 users with an average feature value of 0.76, the second user group contains 35 users with an average feature value of 0.82, and the third user group contains 20 users with an average feature value of 0.69.

[0065] In step S104, if the boundary ambiguity of the initial user group exceeds a preset boundary ambiguity threshold, compensation noise is injected for iterative verification until a group with clear boundaries is obtained. Then, delay-related parameters are extracted to determine a low-latency parameter set, including:

[0066] S1041, if the number of boundary points of the initial user group exceeds a preset boundary point number threshold, the boundary of the initial user group is adjusted to obtain an optimized group boundary.

[0067] S1042, Based on the distribution of boundary points in the preliminary user group, calculate the distance distribution between the boundary points and neighboring points to obtain the boundary ambiguity value;

[0068] S1043, if the boundary ambiguity value is higher than the preset boundary ambiguity threshold, then compensation noise is injected into the boundary point for iterative verification until the boundary ambiguity value meets the threshold, and the adjusted group boundary is obtained.

[0069] S1044, Based on the adjusted group boundaries, analyze the connectivity of boundary points, generate clear boundary groups, and determine the set of clear groups;

[0070] S1045, Based on the clear group set, calculate the delay distribution of data points in each group, obtain delay-related parameters, and obtain a low-latency parameter set.

[0071] In step S1041, if the number of boundary points of the initial user group exceeds a preset boundary point number threshold, the boundary of the initial user group is adjusted to obtain an optimized group boundary.

[0072] In one implementation, assuming the initial user group boundary points are 50 and the preset boundary point threshold is 40, if 50 > 40, a boundary adjustment mechanism is triggered. By introducing a weighted distance adjustment method, the boundary points are recalculated and filtered, compressing the number of boundary points from 50 to less than 40, for example, to 38. Specifically, the distance threshold is adjusted from 0.5 to 0.45. Through iterative updates to the group boundary curve, the shape of the group boundary is smoothed and boundary point redundancy is reduced, resulting in optimized user group boundaries. This improves the stability and accuracy of group segmentation, providing a more reasonable user segmentation basis for subsequent personalized services.

[0073] It should be noted that the specific steps for iteratively updating the group boundary curve are as follows: First, an initial boundary curve is constructed based on the current set of boundary points. Then, a weighted screening is performed to remove redundant or abnormal boundary points. Next, a moving average algorithm is applied to smooth the boundary curve and reduce sharp abrupt changes. Then, the boundary point positions are readjusted according to the updated boundary, and it is determined whether the number of boundary points is lower than a preset threshold and whether the curve change has converged. If the conditions are not met, the above steps are repeated until the number and shape of the boundary points reach the optimal balance or the maximum number of iterations is reached.

[0074] It should be noted that the weighted screening of boundary points mentioned above is usually achieved by calculating the comprehensive weight of each boundary point. The weight is obtained by combining distance indicators (such as the distance from a point to adjacent points or the center of a group) and curvature indicators (such as the curvature of the boundary curve at that point) with a certain weight coefficient. Specifically, the distance and curvature are first normalized, and then weights are assigned according to a preset proportional coefficient to calculate the comprehensive weight value of each point. Points with lower weights (indicating redundancy or anomalies) are preferentially eliminated, while key boundary points with higher weights are retained, thereby achieving the goal of reducing redundant boundary points and maintaining the boundary shape characteristics.

[0075] It should be noted that the specific formula for defining the weights of boundary points in the weighted filtering of boundary points is as follows:

[0076]

[0077] in, A normalized distance metric (distance from the boundary point to the center of the group); The normalized curvature index (the absolute value of the curvature of the boundary curve at that point); These are the weighting parameters for the distance metric. The weighting parameter of the curvature index and , ∈[0,1], α+β=1. Since curvature estimation is sensitive to noise / sampling density, α will be greater than 0.5 according to the SNR scaling rule; in most test scenarios between synthetic boundaries and real samples, α≈0.6~0.67, so α=0.6 is a conservative and universal default value, balancing robustness and shape fidelity; β=0.4 is calculated from α and is used to balance the influence of distance and curvature.

[0078] In step S1042, based on the distribution of boundary points in the preliminary user group, the distance distribution between the boundary points and neighboring points is calculated to obtain the boundary ambiguity value.

[0079] In one implementation, assuming the initial user group contains 100 boundary points, the coordinate information of each boundary point is first extracted. For each boundary point, its Euclidean distance to its five adjacent boundary points is calculated, resulting in a distance set such as [0.12, 0.15, 0.10, 0.14, 0.13] (Euclidean distance values ​​in the feature space). Then, the mean and standard deviation of this distance set are calculated; the mean is 0.128, and the standard deviation is 0.018. Next, the mean distances of all boundary points are statistically analyzed. If the overall mean is large and the standard deviation exceeds a set threshold, such as 0.02, it indicates that the group boundaries are relatively blurry. Finally, the normalized distance standard deviation is used as the boundary blur value, for example, a blur value of 0.9 (ranging from 0 to 1). A higher value indicates a more blurred boundary, thus providing a basis for subsequent boundary optimization and smoothing.

[0080] It should be noted that the calculation of the selected point and its neighbors... The formula for the average Euclidean distance of the boundary points is as follows:

[0081]

[0082] in, This represents the number of adjacent boundary points included in the calculation; i represents the nth point of the selected point; j represents the jth point adjacent to the selected point. This represents the Euclidean distance between the selected point and its adjacent j-th point; Represents the selected point and its neighbors. The average Euclidean distance between the boundary points.

[0083] The formula for calculating the population standard deviation of the average distance of all boundary points is as follows:

[0084]

[0085] in, Standard deviation used to calculate boundary ambiguity , is a key intermediate variable for measuring the dispersion of group boundaries; This is the total number of boundary points; It is the overall standard deviation of the average distance between all boundary points.

[0086] The formula for obtaining the boundary ambiguity through normalization is as follows: (the maximum standard deviation is denoted as...) ):

[0087]

[0088] in, It is the maximum standard deviation of the average distance of all boundary points obtained from historical data statistics; the ambiguity F ranges from [0,1], with a larger value indicating a more ambiguous boundary. Recommended threshold for ambiguity: If F≥ If the result is positive, proceed to the subsequent noise compensation iteration process; otherwise, proceed directly to the next processing step.

[0089] In step S1043, if the boundary ambiguity value is higher than the preset boundary ambiguity threshold, compensation noise is injected into the boundary point for iterative verification until the boundary ambiguity value meets the threshold, and the adjusted group boundary is obtained.

[0090] It should be noted that during the iterative verification process, the ambiguity value of the initial group boundary is first checked to see if it exceeds a preset threshold. If so, a certain amount of compensation noise is injected into the coordinates of each boundary point, for example, a random disturbance following a Gaussian distribution with a mean of 0 and a progressively adjusted standard deviation. After each noise injection, the distance distribution between the boundary point and its neighboring points and the new ambiguity value are recalculated to determine if the threshold requirement is met. If it still exceeds the threshold, the noise standard deviation or injection strategy is adjusted according to the current ambiguity, and the next iteration continues. This process forms a closed-loop feedback, repeatedly adjusting the boundary point positions until the boundary ambiguity value drops below the preset threshold, at which point the group boundary is considered effectively optimized, and the final boundary result is output.

[0091] In one implementation, assume the initially calculated group boundary ambiguity value is 0.95, exceeding the preset boundary ambiguity threshold of 0.8. At this point, the system injects compensation noise into each boundary point. Specifically, Gaussian noise with a mean of 0 and a standard deviation of 0.01 is randomly added to the point's coordinates to perturb the boundary point's position and alleviate boundary overlap and ambiguity. After injecting noise, the distance distribution between the boundary point and its neighboring points is recalculated, and the ambiguity value is updated. After the first iteration, the ambiguity value drops to 0.85, still higher than the threshold. A second iteration is then performed, adjusting the noise standard deviation to 0.008 and recalculating. The ambiguity value decreases to 0.78, meeting the threshold requirement. The iteration then terminates, and the adjusted set of group boundary points is output, achieving boundary optimization and smoothing, reducing redundancy, and improving the accuracy of user group segmentation.

[0092] It should be noted that the specific operations of the iteration are as follows:

[0093] First, set the initial noise standard deviation. The value is 0.01 (consistent with the data measurement unit); then, in the t-th iteration, based on the current ambiguity... With ambiguity threshold The relationship adjustment is performed using the following formula:

[0094]

[0095] in, It is the noise standard deviation at the t-th iteration; It is the noise standard deviation of the previous iteration in the t-th iteration; It is the ambiguity at the t-th iteration; It is the ambiguity threshold at the t-th iteration.

[0096] Next, compensation noise is injected into the boundary points, using the following formula:

[0097]

[0098] in, It is the coordinate vector of boundary point i at the end of the t-th iteration; It is the coordinate vector of boundary point i at the end of the (t-1)th iteration; It is injected compensation noise.

[0099] Recalculate until ≤ Stop the iteration directly or the maximum number of iterations is reached. Recommended value: .

[0100] If convergence is not achieved after reaching the maximum number of iterations, the weighted selection formula for the boundary points can be used:

[0101]

[0102] The weighting parameter α of the distance index is lowered, and the weighting parameter β of the curvature index is raised to increase the proportion of curvature in the weighting and accelerate boundary convergence.

[0103] Alternatively, the current boundary that performs best during the iteration process can be used as an approximate output along with diagnostic information and confidence level, while marking the group as "uncertain" to trigger manual review or further processing.

[0104] In step S1044, based on the adjusted group boundaries, the connectivity of boundary points is analyzed to generate clear boundary groups and determine the clear group set.

[0105] In one implementation, based on the adjusted group boundaries, assuming a total of 120 boundary points, the Euclidean distance between a point and its neighbors (e.g., three adjacent points) is calculated, and a connectivity threshold of 0.05 is set. If the distance between two points is less than this threshold, they are considered connected. A boundary point connectivity graph is further constructed, and a connected component algorithm is used to identify five clearly defined boundary groups. The largest group contains 50 points, and the smallest group contains 12 points. Statistical analysis is performed on the boundary shape of each clearly defined group, eliminating isolated nodes and discontinuous boundary points. Finally, a set containing four clearly defined groups is determined to represent the stable boundaries of user groups, achieving more accurate group partitioning and subsequent boundary optimization.

[0106] In step S1045, based on the clear group set, the delay distribution of data points within each group is calculated, delay-related parameters are obtained, and a low-latency parameter set is obtained.

[0107] In one implementation, based on the clear group set, assuming it contains four groups, each with 30, 45, 25, and 20 data points respectively. First, network latency data for each data point within the corresponding group is obtained from a distributed device database. For example, the latency data range for group 1 is 10ms to 40ms, and for group 2 it is 15ms to 50ms. The latency data is then denoised, removing outliers exceeding three times the standard deviation, resulting in a cleaned latency dataset. The mean latency for group 1 is 25ms, and the standard deviation is 5ms. Then, latency distribution characteristic parameters for each group are calculated, such as mean, variance, and kurtosis, and latency-related parameters are determined in conjunction with environmental parameter fluctuation data. For example, the mean latency for group 3 is 20ms, and the variance is 3... Finally, groups with average latency below 30ms and high stability were selected to form a low-latency parameter set. For example, the low-latency parameter set includes groups 1 and 3 with average latency of 25ms and 20ms, respectively.

[0108] In step S105, clustering is reconfigured based on the low-latency parameter set. If the number of iterations for verification is less than a preset verification threshold, the noise is fused and compensated to obtain the final user group, including:

[0109] S1051, Based on the low-latency parameter set, adjust the initialization parameters of the clustering process, re-execute the clustering, and generate reconfigured groups;

[0110] S1052, if the number of iteration verifications is less than a preset threshold for the number of iteration verifications, then extract compensation noise features from the boundary points in the reconfigured group and perform compensation noise fusion to generate a noise fusion group.

[0111] S1053, perform a stability test on the noise fusion group. If the variance of the data points in the group is lower than the preset variance threshold, it is determined to be a stable group and is used as the final user group.

[0112] In step S1051, the initialization parameters of the clustering process are adjusted according to the low-latency parameter set, and the clustering is re-executed to generate reconfigured groups.

[0113] In one implementation, based on the low-latency parameter set, two groups with a mean network latency below 30ms are selected. This information is used to adjust the initialization parameters of the clustering algorithm. For example, the number of clusters in the K-means algorithm is reduced from 5 to 4, and the initial cluster centers are selected to favor the coordinates of the groups with lower network latency. Specifically, the centers of groups with mean network latency of 20ms and 25ms are used as initial seed points, the maximum number of iterations is set to 100, and the distance threshold is set to 0.05. The clustering process is then re-executed, reassigning 120 data points to new groups that better match the low-latency characteristics. After iteration, four reconfigured groups are generated, with the mean network latency of two groups optimized to 18ms and 22ms respectively, significantly improving system response speed and task scheduling efficiency.

[0114] It should be noted that the above iterative process first adjusts the initialization parameters of the clustering algorithm based on the low-latency parameter set, such as the number of clusters, the initial cluster center position, and the distance threshold. Then, the clustering algorithm is executed to re-divide the groups. In each iteration, the latency performance of the current group division is evaluated by calculating the mean network latency and distribution of data points within each group, and the group division results are scored. If the mean latency of some groups does not meet expectations, such as still being higher than the set mean latency threshold, the clustering parameters are adjusted based on feedback (e.g., moving the cluster center position or adjusting the distance limit), and the clustering is run again. This process is repeated 50 times or until the latency index converges, ultimately generating four reconfigured groups. The mean network latency of two of these groups is optimized to 18ms and 22ms, respectively, thereby significantly improving the system response speed and task scheduling efficiency.

[0115] In step S1052, if the number of iteration verifications is less than a preset threshold for the number of iteration verifications, then the boundary points in the reconfigured group are extracted with compensation noise features and compensation noise fusion is performed to generate a noise fusion group.

[0116] In one implementation, if the current iteration verification count is less than a preset iteration verification count threshold (e.g., the threshold is set to 50 times, and the current iteration count is 30 times), then compensated noise features of the boundary points are extracted from the reconfigured four groups. The specific steps include: first, injecting Gaussian compensated noise with a mean of 0 and a standard deviation of 0.02 into the location data of each group boundary point, and calculating the amplitude, frequency, and correlation characteristics of the noise; second, weighting and fusing the compensated noise features of each boundary point, with weights allocated based on noise intensity and boundary point influence (e.g., weights of 0.6 and 0.4 respectively); finally, generating a noise fusion group boundary containing the fused noise features to improve the stability and ambiguity of the group boundaries. This method ensures that even with insufficient iterations, compensated noise fusion effectively reduces boundary ambiguity, improving clustering accuracy and overall system robustness.

[0117] In step S1053, a stability test is performed on the noise fusion group. If the variance of the data points in the group is lower than the preset variance threshold, it is determined to be a stable group and is used as the final user group.

[0118] In one implementation, the specific steps for stability testing of noise fusion groups are as follows: First, calculate the variance of the network delay values ​​for all data points within each noise fusion group. Assuming the delay data for a certain group is [18ms, 19ms, 17ms, 20ms, 18ms], the calculated variance is approximately 1.04. Then, this variance is compared with a preset variance threshold, which is set to 2.0. If the variance of data points within a group is below the threshold, the group is considered a stable group. Finally, this stable group is used as the end-user group for priority scheduling and resource allocation of subsequent tasks. This step ensures that the selected groups have good internal consistency and network performance stability, improving the overall reliability and efficiency of the system.

[0119] In step S106, a secondary noise injection process is performed on the end-user group to obtain a deviation-minimized group, which is then evaluated to obtain an integrity index, including:

[0120] S1061, Perform secondary noise injection processing on the end user group to further adjust user characteristics and obtain a group with minimized deviation;

[0121] S1062, Based on the deviation-minimizing group, calculate the consistency and stability of the distribution of user behavior features within the group, and then obtain a quantified integrity index.

[0122] In step S1061, a secondary noise injection process is performed on the final user group to further adjust the user characteristics and obtain a group with minimized deviation.

[0123] In one implementation, for the end-user group, a secondary noise injection process is first performed. Specifically, Gaussian noise with a mean of 0 and a standard deviation of 0.01 is injected into the user characteristics (such as CPU utilization and network latency) of each data point within the group to simulate the impact of real-world environmental fluctuations on system performance. Subsequently, an iterative optimization algorithm (such as a genetic algorithm) is used to adjust the weight ratios of the user characteristics. For example, the weight of CPU utilization is adjusted from 0.6 to 0.65, and the weight of network latency is adjusted accordingly to 0.35. The number of iterations is set to 100 to find the feature combination that minimizes the network latency and energy consumption deviation within the group. By calculating the mean variance of network latency and energy consumption fluctuation within the group after each adjustment, the configuration with the smallest deviation is selected to form the deviation-minimizing group. For example, after optimization, the mean network latency within the group decreases from the initial 22ms to 19ms, and the network latency variance decreases from 2.5... Reduced to 1.1 Energy consumption fluctuations are reduced by about 15%, significantly improving the stability of the group and the overall energy efficiency of the system.

[0124] In step S1062, based on the deviation minimization group, the consistency and stability of the distribution of user behavior features within the group are calculated, thereby obtaining a quantified integrity index.

[0125] In one implementation, based on a bias-minimizing group, user behavior characteristic data (such as CPU utilization, network latency, energy consumption monitoring, etc.) is first collected within the group. Assuming 1000 records are collected, the statistical distribution parameters of each characteristic are calculated, including the mean, standard deviation, and covariance matrix. Then, consistency indices (such as the Pearson correlation coefficient between features, assuming an average correlation coefficient of 0.85) and stability indices (such as the variance stability of features, with an average variance of 0.005) are used to quantify the distribution of behavioral characteristics. Next, an integrity index is calculated by weighted summation of these indices. The formula is: the integrity index is obtained by calculating the product of consistency weight and consistency, and the product of stability weight and stability, and then summing the two products. Combining the above values, the integrity index is calculated to be 0.77. This process ensures that the distribution of user behavior characteristics within the group is both highly consistent and stable, providing a reliable quantitative basis for further system optimization and scheduling.

[0126] In step S107, if the integrity index meets a preset index threshold, additional deviation analysis and additional noise injection processing are performed to obtain the final enhanced group, including:

[0127] S1071, if the integrity index is higher than the preset first index threshold, then calculate the dispersion of the user behavior feature distribution within the deviation minimization group to obtain the deviation analysis result;

[0128] S1072, when the deviation analysis result meets the preset second index threshold, a directional enhancement noise vector is generated based on the deviation analysis result;

[0129] S1073, the directional enhancement noise vector is injected into the bias minimization group, and privacy enhancement and bias optimization are performed to generate a privacy enhancement group;

[0130] S1074, Perform stability verification on the privacy enhancement group. If the stability verification result meets the preset stability index threshold, the final enhancement group is obtained.

[0131] In step S1071, if the integrity index is higher than the preset first index threshold, the dispersion of the user behavior feature distribution within the deviation minimization group is calculated to obtain the deviation analysis result.

[0132] In one implementation, when the integrity index of the deviation-minimizing group exceeds a preset first threshold, the system first extracts user behavior characteristic data within that group, such as CPU utilization, network latency, and energy consumption, and performs distribution statistical analysis. Specific steps include calculating the dispersion indices of each characteristic—such as variance and coefficient of variation. For example, assuming the variance of CPU utilization is 0.004, the coefficient of variation of network latency is 12%, and the variance of energy consumption is 3... Subsequently, the dispersion of each feature is integrated, and the overall dispersion is calculated by weighted averaging. The weights can be set as CPU utilization 0.5, network latency 0.3, and energy consumption 0.2, resulting in a comprehensive dispersion value of 0.0065. Finally, based on this comprehensive dispersion, deviation analysis results are generated to determine the stability and uniformity of user behavior within the group, providing a basis for further system optimization.

[0133] It should be noted that the preset first indicator threshold is determined based on a comprehensive consideration of requirements such as business trigger frequency and misjudgment cost, through cross-validation and statistical analysis on representative synthetic and real samples.

[0134] It should be noted that the specific process of weight allocation involves assigning values ​​based on the importance of different user behavior characteristics to the overall dispersion. First, the relative contribution of each characteristic (such as CPU utilization, network latency, and energy consumption) to the overall stability assessment is determined. Here, CPU utilization is assigned a weight of 0.5, indicating that this indicator has the greatest impact on the overall dispersion; network latency is assigned a weight of 0.3, indicating a secondary impact; and energy consumption is assigned a weight of 0.2, indicating a relatively small impact. Then, the dispersion value of each characteristic is multiplied by its corresponding weight, and finally, the weighted results are summed to obtain the overall comprehensive dispersion value of 0.0065. This weight allocation reflects the system's understanding of the importance of the stability of different characteristics, ensuring that the calculated overall dispersion accurately reflects the stability and uniformity of user behavior within the group, thus providing a reasonable basis for generating effective deviation analysis results in subsequent deviation analysis.

[0135] In step S1072, when the deviation analysis result meets the preset second index threshold, a directional enhancement noise vector is generated based on the deviation analysis result.

[0136] In one implementation, when the deviation analysis result meets a preset second threshold (e.g., an interference threshold exceeding 5dB), the system first imports the deviation analysis result data. For example, if the actual interference is 45dB, exceeding the threshold of 40dB, the deviation is 5dB. Subsequently, based on this deviation value and the corresponding environmental parameter fluctuations (e.g., temperature fluctuation of 5°C, humidity fluctuation of 10%), an iterative optimization algorithm (e.g., gradient descent) is used to generate a directional enhancement noise vector. This vector enhances the disturbance in the direction of the most significant deviation. Assuming the noise vector is [0.03, -0.02, 0.05], corresponding to the adjustment ranges of CPU utilization, network latency, and energy consumption, respectively. This vector is applied to the system through dynamic data flow to adjust task allocation and resource scheduling strategies, thereby suppressing deviations and optimizing network load, thus improving the overall system stability and performance robustness.

[0137] In step S1073, the directional enhancement noise vector is injected into the bias minimization group, and privacy enhancement and bias optimization are performed to generate a privacy enhancement group.

[0138] In one implementation, a targeted noise vector (e.g., [0.03, -0.02, 0.05], corresponding to adjustments in CPU utilization, network latency, and energy consumption, respectively) is injected into a bias-minimizing group. First, the user behavior feature data within the group is perturbed to enhance privacy protection. Specifically, after injecting the noise vector into each data point, a differential privacy mechanism is applied to limit the noise amplitude, ensuring that the risk of privacy leakage is below a preset threshold (e.g., privacy budget). =0.5). Subsequently, based on the injected data, a bias optimization algorithm is used to iteratively adjust the feature weights and data distribution within the group, ensuring that the overall bias is within 0.005 while maintaining privacy and security. Ultimately, a privacy-enhanced group with privacy enhancement and minimized bias is formed, maintaining an average network latency of 19ms and reducing energy consumption fluctuations to 85% of the original, achieving a balance between group performance and privacy protection, and providing support for intelligent scheduling and energy management.

[0139] In step S1074, the privacy enhancement group is subjected to stability verification. If the stability verification result meets the preset stability index threshold, the final enhancement group is obtained.

[0140] In one implementation, when verifying the stability of the privacy-enhanced group, outliers are first removed (e.g., CPU utilization fluctuations exceeding 3 standard deviations are eliminated). Then, stability indicators for key features (CPU utilization, network latency, and energy consumption) within the group are calculated. Assuming the calculated stability score is 0.82 and the preset stability threshold is 0.80, if the stability score meets the preset threshold, the group's stability is deemed satisfactory, and the final enhanced group is output. This final enhanced group maintains a stable average network latency within 20ms and energy consumption fluctuations within 5%, ensuring stable system performance while meeting privacy enhancement requirements, providing reliable support for energy management and task scheduling in smart factories.

[0141] In summary, this invention discloses a dynamic task combination method based on user digital profiles, comprising:

[0142] Obtain historical user behavior data, extract preference features, and form a preliminary user profile;

[0143] Based on the initial user profile, noise injection processing is performed to obtain a noise-enhanced profile, which is then desensitized and encrypted to obtain an encrypted noise profile.

[0144] Clustering is performed on the encrypted noise profile to form preliminary user groups;

[0145] If the degree of boundary ambiguity of the initial user group exceeds the preset boundary ambiguity threshold, compensation noise is injected for iterative verification until a group with clear boundaries is obtained. Then, delay-related parameters are extracted to determine the low-latency parameter set.

[0146] The clustering is reconfigured based on the low-latency parameter set. If the number of iterations for verification is less than a preset verification threshold, the noise is fused and compensated to obtain the final user group.

[0147] The end-user group is subjected to secondary noise injection processing to obtain a group with minimized deviation, which is then evaluated to obtain an integrity index.

[0148] If the integrity index meets the preset index threshold, additional deviation analysis and additional noise injection processing are performed to obtain the final enhanced group.

[0149] Reference Figure 2 The second embodiment of the present invention provides a dynamic task combination system based on user digital profiles, including:

[0150] User data extraction module: acquires historical user behavior data, extracts preference features, and forms a preliminary user profile;

[0151] Noise injection protection module: Based on the preliminary user profile, perform noise injection processing to obtain a noise-enhanced profile, and perform desensitization and encryption processing to obtain an encrypted noise profile;

[0152] Clustering to form groups module: Performs clustering operations on the encrypted noise profile to form preliminary user groups;

[0153] Boundary blur correction module: If the boundary blur degree of the initial user group exceeds the preset boundary blur threshold, compensation noise is injected for iterative verification until a clear boundary group is obtained. Then, delay-related parameters are extracted to determine the low-latency parameter set.

[0154] Low-latency clustering module: Reconfigures clustering based on the low-latency parameter set; if the number of iterations for verification is less than a preset verification threshold, then merges and compensates for noise to obtain the final user group.

[0155] Secondary noise assessment module: Performs secondary noise injection processing on the end-user group to obtain the group with minimal deviation and performs assessment to obtain the integrity index;

[0156] Additional noise enhancement module: If the integrity index meets the preset index threshold, additional deviation analysis and additional noise injection processing are performed to obtain the final enhanced group.

[0157] It should be noted that the dynamic task combination system based on user digital profile provided in this embodiment of the invention is used to execute all the process steps of the dynamic task combination method based on user digital profile in the above embodiment. The working principle and beneficial effects of the two are one-to-one, so they will not be described again.

[0158] This invention also provides an electronic device. The electronic device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a dynamic task composition program. When the processor executes the computer program, it implements the steps described in the various embodiments of the pipe weld defect analysis methods above, for example... Figure 1The step S101 is shown. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the above-described device embodiments, for example... Figure 2 The user data extraction module shown.

[0159] The electronic device may be a desktop computer, laptop, handheld computer, or smart tablet, etc. The electronic device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above components are merely examples of electronic devices and do not constitute a limitation on the electronic device. It may include more or fewer components than described above, or combine certain components, or different components. For example, the electronic device may also include input / output devices, network access devices, buses, etc.

[0160] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the electronic device, connecting all parts of the electronic device via various interfaces and lines.

[0161] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0162] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0163] The above specific embodiments have further described in detail the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and do not limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A dynamic task combination method based on user digital portrait, characterized in that, The method comprises the following steps: acquiring user historical behavior data, extracting preference features, and forming a preliminary user portrait; performing noise injection processing according to the preliminary user portrait to obtain a noise-enhanced portrait, and performing desensitization processing and encryption processing to obtain an encrypted noise portrait; performing clustering operation on the encrypted noise portrait to form a preliminary user group; if the boundary fuzziness degree of the preliminary user group exceeds a preset boundary fuzziness threshold, injecting compensation noise for iterative verification until a boundary clear group is obtained, extracting delay-related parameters, and determining a low-delay parameter set; reconfiguring clustering according to the low-delay parameter set, fusing compensation noise if the number of iterations is less than a preset verification number threshold, and obtaining a final user group; performing secondary noise injection processing on the final user group to obtain a deviation-minimized group and performing evaluation to obtain an integrity index; if the integrity index meets a preset index threshold, performing additional deviation analysis and additional noise injection processing to obtain a final enhanced group; wherein, based on the clear group set, the delay-related parameters are extracted, the low-delay parameter set is determined, which comprises: obtaining network delay data of each data point in the corresponding group from a distributed device database; denoising the delay data to obtain a cleaned delay data set; calculating the delay distribution feature parameters of each group, including mean, variance and kurtosis, and combining environmental parameter fluctuation data to determine the delay-related parameters; finally, the groups with low delay mean and high stability are screened out to form the low-delay parameter set; wherein, the secondary noise injection processing on the final user group to obtain a deviation-minimized group and the evaluation to obtain an integrity index comprise: performing secondary noise injection processing on the final user group to further adjust the user features to obtain a deviation-minimized group; based on the deviation-minimized group, calculating the consistency and stability of the user behavior feature distribution in the group to obtain a quantitative integrity index.

2. The dynamic task combination method based on user digital portrait according to claim 1, characterized in that, The method comprises the following steps: acquiring user behavior logs and performing time dimension division to obtain user historical behavior data; extracting time distribution features of user behavior from the user historical behavior data to form a preference feature set; according to the preference feature set, weighting and fusing the time distribution features and interaction frequency to generate an initial user description structure and determine a preliminary user portrait.

3. The dynamic task combination method based on user digital portrait according to claim 1, characterized in that, The method comprises the following steps: injecting privacy protection noise into the preliminary user portrait to obtain a noise-enhanced portrait; performing anonymization processing on sensitive information in the noise-enhanced portrait to obtain a desensitization feature set; adjusting the encryption strength of the desensitization feature set and generating an encrypted noise portrait.

4. The dynamic task combination method based on user digital portrait according to claim 1, characterized in that, The method comprises the following steps: initializing clustering parameters, performing clustering processing on the encrypted noise portrait to obtain a preliminary user group.

5. The dynamic task combination method based on user digital portrait according to claim 1, characterized in that, If the boundary ambiguity of the preliminary user group exceeds a preset boundary ambiguity threshold, compensation noise is injected for iterative verification until a boundary clear group is obtained, and delay correlation parameters are extracted to determine a low delay parameter set, including: If the number of boundary points of the preliminary user group exceeds a preset boundary point number threshold, the boundary of the preliminary user group is adjusted to obtain an optimized group boundary; According to the distribution of the boundary points in the preliminary user group, the distance distribution of the boundary points and the adjacent points is calculated to obtain a boundary ambiguity value; If the boundary ambiguity value is higher than a preset boundary ambiguity threshold, compensation noise is injected to the boundary points for iterative verification until the boundary ambiguity value meets the threshold, and an adjusted group boundary is obtained; Based on the adjusted group boundary, the connectivity of the boundary points is analyzed to generate a boundary clear group and determine a clear group set; Based on the clear group set, the delay distribution of the data points in each group is calculated to obtain delay correlation parameters and obtain a low delay parameter set.

6. The dynamic task combination method based on user digital portrait according to claim 1, characterized in that, According to the low delay parameter set, the cluster is reconfigured, and if the number of iterative verifications is less than a preset verification number threshold, the compensation noise is fused to obtain a final user group, including: According to the low delay parameter set, the initialization parameters of the clustering processing are adjusted, the clustering is re-executed, and a reconfigured group is generated; If the number of iterative verifications is less than a preset iterative verification number threshold, compensation noise features are extracted from the boundary points in the reconfigured group, and compensation noise fusion is performed to generate a noise fusion group; The noise fusion group is subjected to stability test, and if the variance of the data points in the group is lower than a preset variance threshold, it is determined as a stable group as the final user group.

7. The dynamic task combination method based on user digital portrait according to claim 1, characterized in that, If the integrity index meets a preset index threshold, additional bias analysis and additional noise injection processing are performed to obtain a final enhanced group, including: If the integrity index is higher than a preset first index threshold, the dispersion of the user behavior feature distribution in the bias minimization group is calculated to obtain a bias analysis result; When the bias analysis result meets a preset second index threshold, a directional enhancement noise vector is generated based on the bias analysis result; The directional enhancement noise vector is injected into the bias minimization group, and privacy enhancement and bias optimization are performed to generate a privacy enhanced group; The privacy enhanced group is subjected to stability verification, and if the stability verification result meets a preset stability index threshold, a final enhanced group is obtained.

8. A dynamic task combination system based on user digital portrait, for implementing the dynamic task combination method based on user digital portrait as claimed in any one of claims 1-7, characterized in that, Including: User data extraction module: Obtain user historical behavior data, extract preference features, and form a preliminary user portrait; Noise injection protection module: According to the preliminary user portrait, noise injection processing is performed to obtain a noise enhanced portrait, and desensitization processing and encryption processing are performed to obtain an encrypted noise portrait; Cluster formation group module: The encrypted noise portrait is subjected to clustering operation to form a preliminary user group; The boundary blur correction module: if the boundary blur degree of the preliminary user group exceeds the preset boundary blur threshold, compensation noise is injected for iterative verification until a clear boundary group is obtained, delay-related parameters are extracted, and a low-delay parameter set is determined; The low-delay clustering module: the clustering is reconfigured according to the low-delay parameter set, if the number of iterations is less than the preset verification number threshold, the compensation noise is fused, and a final user group is obtained; The secondary noise evaluation module: the final user group is subjected to secondary noise injection processing, a deviation minimization group is obtained, and evaluation is performed to obtain an integrity index; The additional noise enhancement module: if the integrity index meets the preset index threshold, additional deviation analysis and additional noise injection processing are performed to obtain a final enhancement group.

Images