A social account operation data anomaly automatic diagnosis method and system
By combining data dimensionality reduction and feature extraction with ARIMA model prediction of residuals, the problem of insufficient accuracy of the isolated forest algorithm in the diagnosis of social account operation data is solved, and higher accuracy of anomaly diagnosis is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- BEIJING ZHUOXIN HUIZHI TECHNOLOGY CO LTD
- Filing Date
- 2025-10-30
- Publication Date
- 2026-05-19
AI Technical Summary
Existing technologies using the isolated forest algorithm for diagnosing anomalies in social media account operation data struggle to distinguish between short-term fluctuations and genuine anomalies, resulting in low diagnostic accuracy.
By collecting user and behavioral data from social media accounts across various sub-dimensions, dimensionality reduction and feature extraction are performed. Combined with ARIMA model prediction of residuals and feature value deviations between social media accounts, data regularity is determined for anomaly diagnosis.
It improves the accuracy of diagnosing anomalies in social media account operation data, enabling a better distinction between short-term fluctuations and genuine anomalies.
Smart Images
Figure CN121304367B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of data anomaly detection technology, specifically to an automatic diagnostic method and system for anomalies in the operational data of social media accounts. Background Technology
[0002] Social media account operation data includes user data (such as number of followers, follower growth rate, percentage of active followers, etc.) and behavioral data (content posting frequency, interaction frequency, etc.). This data helps reflect the account's basic status, user engagement, content performance, and external environment. During operation, data fluctuations may be caused by a variety of factors, so automatic diagnosis of data anomalies is very important.
[0003] Existing technologies typically use the Isolation Forest algorithm to diagnose anomalies based on the operational data of social media accounts, using anomaly scores to determine operational data anomalies. However, the Isolation Forest algorithm fails to fully incorporate the temporal changes characteristic of social media account operations, making it difficult to distinguish between short-term fluctuations and genuine anomalies. Especially in social media operations, many external or internal factors can cause short-term fluctuations, but these fluctuations do not necessarily indicate fundamental operational problems. For example, a sudden public opinion crisis may cause a large number of users to focus on or discuss a particular topic within a short period, resulting in instantaneous fluctuations in account data. These fluctuations are usually short-term and unrelated to the normal operation of the account. Therefore, the accuracy of existing technologies using the Isolation Forest algorithm for anomaly diagnosis of social media account operational data is relatively low. Summary of the Invention
[0004] To address the low accuracy of existing technologies using the Isolation Forest algorithm for anomaly diagnosis of social media account operational data, this application aims to provide an automatic anomaly diagnosis method for social media account operational data. The specific technical solution adopted is as follows:
[0005] The first aspect of this application provides an automatic diagnostic method for abnormal operational data of social media accounts, including:
[0006] At each sampling moment within the current time window, data sample values are collected for each social media account under each sub-dimension of each operational data dimension; the operational data dimension includes user data dimension and behavioral data dimension.
[0007] Data dimensionality reduction is performed based on the temporal fluctuation correlation of data sample values between user data dimension and behavioral data dimension under each sub-dimension within the current time window. The data feature value of each operational data dimension of each social account at each sampling time is determined. Based on the prediction error of the data feature value, the prediction residual of each operational data dimension of each social account at each sampling time within the current time window is determined.
[0008] At each sampling time, the importance of the feature data of each social account at each sampling time is determined based on the deviation of the data feature values of each social account from other social accounts in various operational data dimensions; the data regularity within the current time window is determined based on the distribution of the feature data importance of each social account at all sampling times and the relative deviation of the prediction residual.
[0009] Based on the aforementioned data patterns, operational data anomaly diagnosis is performed.
[0010] Furthermore, the process of obtaining the data feature values includes:
[0011] For each social media account, arrange all data sample values of each sub-dimension in each operational data dimension in chronological order within the current time window to determine the corresponding data sample value sequence;
[0012] The user data dimension and the behavioral data dimension are used as target data dimensions in sequence, and each sub-dimension in the target data dimension is used as a target sub-dimension in sequence; another operational data dimension other than the target data dimension is used as a reference data dimension.
[0013] Based on the fluctuation of the data sample value sequence corresponding to each sub-dimension of the reference data dimension and the trend correlation between the corresponding data sample value sequence and the data sample value sequence corresponding to the target sub-dimension, the corresponding weighted consistency is determined.
[0014] The dimensionality reduction importance of the target data dimension under the target sub-dimension is determined based on the cumulative weighted consistency value corresponding to all sub-dimensions of the reference data dimension.
[0015] PCA dimensionality reduction is performed by weighting the importance of each operational data dimension of each social account under each sub-dimension as the weight, and the data feature value of each operational data dimension of each social account at each sampling time is determined after dimensionality reduction.
[0016] Furthermore, the process of obtaining the weighted consistency includes:
[0017] The variance of all data sample values in the data sample value sequence corresponding to each sub-dimension of the reference data dimension is normalized to determine the corresponding contribution weight value; the Pearson correlation coefficient between the data sample value sequence corresponding to each sub-dimension of the reference data dimension and the data sample value sequence corresponding to the target sub-dimension is normalized to determine the corresponding data consistency; and the weighted consistency corresponding to each sub-dimension of the reference data dimension is determined based on the product of the contribution weight value and the data consistency.
[0018] Furthermore, the process of obtaining the predicted residual includes:
[0019] Based on the data feature values of each operational data dimension of each social account before each sampling time, the ARIMA model is used to predict the predicted feature values of each operational data dimension of each social account at each sampling time.
[0020] The difference between the data feature value and the predicted feature value is normalized to determine the prediction residual of each operational data dimension of each social account at each sampling time within the current time window.
[0021] Furthermore, the process of obtaining the importance of the feature data includes:
[0022] Each social media account is designated as the target account, and other social media accounts outside the target accounts are designated as reference accounts.
[0023] At each sampling moment within the current time window, the difference between the data feature values of the target account in the user data dimension and the data feature values of each reference account in the user data dimension is negatively correlated and normalized to determine the consistency of user features of each reference account in the user data dimension.
[0024] The differences between the data feature values of the target account in the behavioral data dimension and the data feature values of each reference account in the behavioral data dimension are normalized to determine the corresponding degree of behavioral deviation.
[0025] At each sampling moment within the current time window, the importance of the feature data of the target account at each sampling moment is determined based on the overall magnitude of the consistency of user characteristics and the degree of behavioral deviation among all reference accounts.
[0026] Furthermore, the process of determining the importance of the feature data of the target account at each sampling time based on the overall magnitude of the consistency of user characteristics and the degree of behavioral deviation among all reference accounts includes:
[0027] The importance of reference data for each reference account at each sampling time within the current time window is determined by multiplying the degree of behavioral deviation with the consistency of user characteristics. Within the current time window, the importance of feature data for the target account at each sampling time is determined by summing the reference data importance of all reference accounts at each sampling time.
[0028] Furthermore, the process of obtaining the data regularity includes:
[0029] At each sampling time, the difference between the prediction error under the user data dimension and the prediction error under the behavioral data dimension of each social account is negatively correlated and normalized to determine the corresponding data stability.
[0030] Based on the stability of the data and the importance of the feature data, the local regularity of each social account at each sampling time is determined;
[0031] Based on the cumulative value of the local regularity of each social account across all sampling times, the data regularity of each social account within the current time window is determined.
[0032] Furthermore, the process of obtaining the local regularity includes:
[0033] The importance of the feature data of each social account at each sampling time is normalized to determine the corresponding importance contribution value; the local regularity of each social account at each sampling time is determined based on the product between the importance contribution value and the data stability.
[0034] Furthermore, the process of diagnosing operational data anomalies based on the data regularity includes:
[0035] Within the current time window, social media accounts whose data regularity is less than a preset regularity threshold will be classified as accounts with abnormal operational data.
[0036] Secondly, this application provides an automatic diagnostic system for abnormal operational data of social media accounts, the system comprising:
[0037] The data acquisition module is used to collect data sample values for each social account under each sub-dimension of each operational data dimension at each sampling moment within the current time window; the operational data dimensions include user data dimensions and behavioral data dimensions.
[0038] The first determining module is used to perform data dimensionality reduction based on the temporal fluctuation correlation of data sample values between user data dimension and behavior data dimension in each sub-dimension within the current time window, and to determine the data feature value of each operational data dimension of each social account at each sampling time; and to determine the prediction residual of each operational data dimension of each social account at each sampling time based on the prediction error of the data feature value.
[0039] The second determination module is used to determine the importance of the feature data of each social account at each sampling time based on the deviation of the data feature values of each social account from other social accounts in various operational data dimensions; and to determine the data regularity within the current time window based on the distribution of the feature data importance of each social account at all sampling times and the relative deviation of the prediction residual.
[0040] The anomaly diagnosis module is used to diagnose operational data anomalies based on the regularity of the data.
[0041] Thirdly, this application provides a computer device including a memory and a processor. The memory is used to store computer program code, and the processor is used to call and run the computer program code from the memory to perform the method as described in the first aspect of this application or any embodiment of the first aspect.
[0042] Fourthly, this application provides a computer program product comprising computer program code, which, when executed, performs the method as described in the first aspect of this application or any embodiment thereof.
[0043] Fifthly, this application provides a computer-readable storage medium that stores computer program code, which, when executed, performs the method as described in the first aspect of this application or any embodiment thereof.
[0044] This application has the following beneficial effects:
[0045] This application first performs data dimensionality reduction by combining the fluctuation of data sampling values under multiple sub-dimensions for each operational data dimension, and determines data feature values that are more representative of the time-series changes in the data. Then, based on the regularity pattern of trends, periodicity, and inherent dependencies in the operational data of social accounts, the prediction residual is determined according to the prediction error of the data feature values. Furthermore, based on the prediction residual, the importance of feature data represented by the deviation of data feature values of different social accounts is combined to comprehensively characterize the data regularity of social accounts within the current time window, so as to improve the accuracy of operational data anomaly diagnosis based on data regularity. Attached Figure Description
[0046] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0047] Figure 1 A flowchart of an automatic diagnosis method for abnormal operation data of a social account provided in an embodiment of the present invention;
[0048] Figure 2 This is a structural diagram of an automatic diagnosis system for abnormal operational data of social media accounts provided in one embodiment of the present invention;
[0049] Figure 3This is a schematic diagram of a computer device structure provided in one embodiment of the present invention. Detailed Implementation
[0050] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of an automatic diagnosis method for abnormal operational data of social media accounts proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment, and specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined with "first" or "second" may explicitly or implicitly include one or more of that feature.
[0051] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.
[0052] The following description, in conjunction with the accompanying drawings, details the specific scheme of the automatic diagnosis method for abnormal operation data of social media accounts provided by the present invention.
[0053] This application provides an embodiment of a method for automatically diagnosing anomalies in the operational data of social media accounts. Please refer to [link / reference]. Figure 1 The diagram illustrates a flowchart of an automatic diagnosis method for abnormal operational data of a social media account according to an embodiment of the present invention. The method includes:
[0054] Step S101: At each sampling moment within the current time window, collect data sampling values for each social account under each sub-dimension of each operational data dimension; the operational data dimensions include user data dimension and behavioral data dimension.
[0055] Data samples are collected from each social media account under each sub-dimension of each operational data dimension through APIs officially authorized by the social media platform. The sampling period is the current time window. The operational data dimensions include user data and behavioral data. Sub-dimensions of user data include number of followers, follower growth rate, and percentage of active followers (active followers are those who have interacted with the social media account's content within the past week by viewing, commenting, liking, or forwarding). Sub-dimensions of operational data include content posting frequency, frequency of interaction with followers, and frequency of interaction with other social media accounts. These can be adjusted according to the specific implementation environment and will not be elaborated further here.
[0056] In one specific implementation of this invention, the current time window is set to the time window corresponding to the time period within one day prior to the current moment, and the sampling frequency is set to once per minute. This can be adjusted according to the specific implementation environment, and will not be elaborated further here. It should be noted that, in order to avoid the influence of units on subsequent analysis, the data sampling values under each sub-dimension of each operational data dimension collected in this embodiment of the invention are normalized values. The normalization method adopted is linear normalization, which will not be elaborated further here.
[0057] Step S102: Perform data dimensionality reduction based on the temporal fluctuation correlation of data sample values between user data dimension and behavior data dimension under each sub-dimension within the current time window, and determine the data feature value of each operational data dimension of each social account at each sampling time; determine the prediction residual of each operational data dimension of each social account at each sampling time within the current time window based on the prediction error of the data feature value.
[0058] When calculating the user feature dimensions and behavioral data feature dimensions of social accounts within each time scale and window, dimensionality reduction is needed to transform high-dimensional data into low-dimensional data to facilitate subsequent calculations, while retaining the most representative features. However, considering that each dimension of user feature dimensions and behavioral data dimensions has multiple sub-dimensions, and these sub-dimensions contribute differently to the subsequent analysis of regularity results—for example, the percentage of active fans may have different degrees of correlation with the average like rate, average comment rate, and average repost rate of content; for sub-dimensions with significant correlation features with other data dimensions, the information they contain is usually more important because the corresponding sub-dimensions are usually of higher importance—this embodiment of the invention performs data dimensionality reduction based on the temporal fluctuation correlation of the data sample values of user data dimensions and behavioral data dimensions under each sub-dimension within the current time window, determining the data feature values of each operational data dimension of each social account at each sampling time.
[0059] Preferably, in some possible implementations of the embodiments of the present invention, the process of obtaining data feature values includes: for each social account, arranging all data sample values of each sub-dimension in each operational data dimension in chronological order within the current time window to determine the corresponding data sample value sequence; sequentially using the user data dimension and behavioral data dimension as target data dimensions, and sequentially using each sub-dimension in the target data dimension as target sub-dimension; and using another operational data dimension other than the target data dimension as a reference data dimension. Since correlation analysis usually requires combining time-series changes, correlation analysis is performed using the data sample value sequence obtained after time-series arrangement.
[0060] Based on the fluctuation of the data sample value sequence corresponding to each sub-dimension of the reference data dimension and the trend correlation between the corresponding data sample value sequence and the data sample value sequence corresponding to the target sub-dimension, the corresponding weighted consistency is determined. In a specific implementation of this invention, the process of obtaining weighted consistency includes: normalizing the variance of all data sample values in the data sample value sequence corresponding to each sub-dimension of the reference data dimension to determine the corresponding contribution weight value; normalizing the Pearson correlation coefficient between the data sample value sequence corresponding to each sub-dimension of the reference data dimension and the data sample value sequence corresponding to the target sub-dimension to determine the corresponding data consistency; and determining the weighted consistency corresponding to each sub-dimension of the reference data dimension based on the product of the contribution weight value and the data consistency.
[0061] For the reference data dimension, the larger the variance of the data sample value sequence corresponding to its sub-dimension, the more significant its fluctuations, and the more information it contains. Therefore, it needs to be assigned a larger contribution weight value. The Pearson correlation coefficient can characterize the trend association between sequences. Therefore, the greater the data consistency and the larger the contribution weight value, the greater the dimensionality reduction weight of the target sub-dimension should be under the reference of the sub-dimension corresponding to the reference data dimension. Therefore, this embodiment of the invention further determines the dimensionality reduction importance of the target data dimension under the target sub-dimension based on the weighted consistency accumulation value of all sub-dimensions corresponding to the reference data dimension. The greater the dimensionality reduction importance, the greater the weight of the corresponding target sub-dimension during dimensionality reduction. Therefore, PCA dimensionality reduction is further performed by using the dimensionality reduction importance of each operational data dimension of each social account under each sub-dimension as the weight. The data feature value of each operational data dimension of each social account at each sampling time is determined based on the data sequence obtained after dimensionality reduction.
[0062] In one specific implementation of this invention, the process of obtaining the dimensionality reduction importance is expressed by the following formula: ;in, For social media accounts Target data dimensions In the target sub-dimension The importance of dimensionality reduction; For social media accounts Target data dimensions The number of sub-dimensions of the corresponding reference data dimension; For social media accounts Target data dimensions The corresponding reference data dimension The variance of all data sample values in the data sample value sequence corresponding to each sub-dimension; It is the softmax function; For social media accounts Target data dimensions The corresponding reference data dimension The contribution weight values corresponding to each sub-dimension, according to the definition of the softmax function, social accounts Target data dimensions The sum of the contribution weights of all sub-dimensions under the corresponding reference data dimension is 1. Normalization is performed by the softmax function to make the dimensionality reduction importance obtained by the subsequent summation more accurate. For social media accounts Target data dimensions In the target sub-dimension The data sample value sequence below and the corresponding reference data dimension are shown below. Pearson correlation coefficient between data sample value sequences of each sub-dimension; It is a minimum-maximum normalization function; For social media accounts Target data dimensions Target sub-dimension The first corresponding reference data dimension Data consistency across all sub-dimensions.
[0063] Social media account operation data contains a mixture of trends, periodicity, and inherent dependencies. The ARIMA model is a powerful time series forecasting tool that, based on historical data within a learning window at each time scale, can accurately capture and quantify these known regular patterns. If the predicted value at each point in time within the window has a small residual compared to the actual value, it indicates that the actual situation fully conforms to expectations based on historical patterns, suggesting strong regularity at that point in time. If the residual is large, it means that an unexpected event has occurred that the model has not learned from. This large deviation may stem from sudden public opinion events, an extremely successful (or unsuccessful) content release, or the actions of competitors, thus providing a basis for comparison when conducting regularity analysis of different social media accounts. Therefore, this embodiment of the invention further determines the prediction residual of each operational data dimension of each social media account at each sampling time within the current time window based on the prediction error of the data feature values.
[0064] Preferably, in some possible implementations of this invention, the process of obtaining the prediction residual includes: predicting the data feature values of each operational data dimension of each social account before each sampling time using an ARIMA model to determine the predicted feature values of each operational data dimension of each social account at each sampling time; normalizing the difference between the data feature values and the predicted feature values to determine the prediction residuals of each operational data dimension of each social account at each sampling time within the current time window. According to the characteristics of the ARIMA model, the smaller the deviation between the predicted feature value and the corresponding data feature value, the more consistent the data change state of the corresponding social account at the corresponding sampling time is with the historical data change state. That is, the evolution pattern of the corresponding operational data dimension is stable and conforms to historical patterns, and the lower the probability of sudden events or the less impact from sudden events.
[0065] In one specific implementation of this invention, the process of obtaining the predicted residual includes: ;in, For social media accounts Operational data dimensions Within the current time window Prediction residuals at each sampling time; For social media accounts Operational data dimensions Within the current time window Data feature values at each sampling time; For social media accounts Operational data dimensions Within the current time window Predicted feature values at each sampling time; It is the absolute value symbol; It is a minimum-maximum normalization function.
[0066] Step S103: At each sampling time, based on the deviation of data feature values between each social account and other social accounts in various operational data dimensions, determine the importance of feature data for each social account at each sampling time; based on the distribution of feature data importance for each social account at all sampling times and the relative deviation of prediction residuals, determine the data regularity within the current time window.
[0067] When monitoring the importance of user characteristics and behavioral data at each point in time within the current window for identifying patterns, it's crucial to compare the data within the same environment as the operational data of other social media accounts. A similar environment means that the social media accounts share similar characteristics in the user data dimension. By controlling for variables, this allows for a more accurate analysis of the characteristics contained in the behavioral data dimension. If, within the same environment, there are significant differences in the behavioral data characteristic values between the analyzed social media account and other social media accounts, the operational decisions of the analyzed social media account may have changed. In such cases, the higher the level of attention required, the higher the importance of the corresponding characteristic data should be. Therefore, further, at each sampling time, the importance of the characteristic data for each social media account at each sampling time is determined based on the deviation of the data characteristic values of each social media account from other social media accounts across various operational data dimensions.
[0068] Preferably, in some possible implementations of the embodiments of the present invention, the process of obtaining the importance of feature data includes: sequentially taking each social media account as the target account and other social media accounts outside the target account as reference accounts; at each sampling moment within the current time window, performing negative correlation normalization on the differences between the data feature values of the target account in the user data dimension and the data feature values of each reference account in the user data dimension to determine the consistency of user features of each reference account in the user data dimension; normalizing the differences between the data feature values of the target account in the behavioral data dimension and the data feature values of each reference account in the behavioral data dimension to determine the corresponding degree of behavioral deviation;
[0069] Based on the process of obtaining user feature consistency, the greater the user feature consistency, the more similar the target account and the reference account are in terms of user data dimensions at the corresponding sampling time. This means the analyzed environment is closer to being consistent, and the behavioral data dimension of the reference account is more valuable for the target account's behavioral data dimension at the corresponding sampling time. Conversely, for the target account, the greater the deviation in behavioral data feature values between it and the reference account at the corresponding sampling time—that is, the greater the degree of behavioral deviation—the greater the deviation between the target account's operational behavior and the reference account at the corresponding sampling time. This suggests that the target account's operational decisions at the corresponding sampling time may have deviated from those of the reference account.
[0070] To more accurately analyze the behavioral data deviation characteristics of the target account, and further combine all reference accounts, at each sampling time within the current time window, the importance of the target account's feature data at each sampling time is determined based on the consistency of user characteristics across all reference accounts and the overall magnitude of behavioral deviation. The process of determining the importance of the target account's feature data at each sampling time based on the consistency of user characteristics across all reference accounts and the overall magnitude of behavioral deviation includes:
[0071] The importance of reference data for each reference account at each sampling time point within the current time window is determined by multiplying the degree of behavioral deviation by the consistency of user characteristics. Within the current time window, the importance of the target account's feature data at each sampling time point is determined by summing the importance of the reference data for all reference accounts at each sampling time point. The greater the deviation in data feature values across the behavioral data dimension between the target account and reference accounts in similar environments, the more likely the target account is to experience changes in operational behavioral data. Therefore, the more attention needs to be paid to the data features of the target account at the corresponding sampling time point, and the greater the importance of the corresponding feature data. Weighting the degree of behavioral deviation by user characteristic consistency ensures that reference accounts with more similar user data dimension features have better reference value, improving the accuracy of the calculated feature data importance.
[0072] In one specific implementation of this invention, the process of obtaining the importance of feature data is expressed by the following formula: ;in, For social media accounts Within the current time window Importance of feature data at each sampling time; For social media accounts The corresponding number of reference accounts; For social media accounts The corresponding number The user data dimension of the reference account is the [number]th [period] within the current time window. Data feature values at each sampling time; For social media accounts The user data dimension within the current time window Data feature values at each sampling time; For social media accounts The corresponding number The behavioral data dimension of the reference account within the current time window Data feature values at each sampling time; For social media accounts The behavioral data dimension within the current time window Data feature values at each sampling time; The first in the current time window Social media accounts at each sampling time The corresponding number Consistency of user characteristics across all reference accounts within the user data dimension; normalization is performed using the softmax function to ensure consistency of user characteristics across all reference accounts within the current time window. The cumulative value of the consistency of user features across all reference accounts at each sampling time point is 1. The first in the current time window Social media accounts at each sampling time The corresponding number The degree of behavioral deviation of each reference account in the behavioral data dimension; The first in the current time window Social media accounts at each sampling time The corresponding number The importance of reference data for each reference account.
[0073] For each social media account, the similarity between the prediction error in the user data dimension and the prediction error in the behavioral data dimension can, to some extent, characterize whether there is a consistent pattern between the user data features and behavioral data features within the current time window. If, at a certain sampling moment, the prediction residuals between the user data dimension and the behavioral data dimension are highly similar, it indicates that user behavior is determined by the long-term patterns of the account itself rather than a single sudden factor, meaning it is more consistent with the data regularity under the historical operation model. Furthermore, the higher the importance of the feature data at the corresponding sampling moment, the higher the reference value of the regularity it represents.
[0074] Preferably, in some possible implementations of the embodiments of the present invention, the process of obtaining data regularity includes: at each sampling time, negatively normalizing the difference between the prediction error under the user data dimension and the prediction error under the behavioral data dimension for each social account to determine the corresponding data stability; determining the local regularity of each social account at each sampling time based on the data stability and the importance of the feature data; in a specific implementation of the embodiments of the present invention, the process of obtaining local regularity includes: normalizing the importance of the feature data of each social account at each sampling time to determine the corresponding importance contribution value; determining the local regularity of each social account at each sampling time based on the product between the importance contribution value and the data stability.
[0075] Greater data stability indicates that the relative changes in user and behavioral characteristics conform more closely to historical patterns, resulting in a more stable operational status at the corresponding sampling time and a lower probability of operational anomalies. Furthermore, the greater the importance of the feature data at the corresponding sampling time, the higher the reference value of the data stability at that sampling time. In other words, greater local regularity indicates a more regular operational status of the corresponding social media account at the corresponding sampling time, and a lower probability of operational anomalies. Further, by synthesizing all sampling times and calculating the cumulative value of the local regularity of each social media account across all sampling times, the data regularity of each social media account within the current time window is determined; thus, greater data regularity further reduces the likelihood of operational data anomalies.
[0076] In one specific implementation of this invention, the process of obtaining data regularity is expressed by the following formula: ;in, For social media accounts The regularity of data within the current time window; This represents the number of sampling moments within the current time window. For social media accounts Within the current time window Importance of feature data at each sampling time; For social media accounts Within the current time window The importance contribution value at each sampling time point is normalized using the softmax function to reflect the social media account's contribution. The cumulative value of the importance contribution at all sampling times within the current time window is 1; The first in the current time window Social media accounts at each sampling time Prediction error in the context of user data; The first in the current time window Social media accounts at each sampling time Prediction error in behavioral data dimension; For social media accounts Within the current time window Data stability at each sampling time; For social media accounts Within the current time window Local regularity at each sampling time.
[0077] Step S104: Diagnose operational data anomalies based on data patterns.
[0078] Since the greater the regularity of the data obtained through step S103, the lower the probability of operational data anomalies, operational data anomaly diagnosis is finally performed based on data regularity. Preferably, in some possible implementations of this invention, the process of diagnosing operational data anomalies based on data regularity includes: within the current time window, identifying social media accounts with data regularity less than a preset regularity threshold as operational data anomaly accounts. In a specific implementation of this invention, the preset regularity threshold is set to 0.6, which can be adjusted according to the specific implementation environment, and will not be further elaborated here.
[0079] In summary, an automatic diagnosis method for anomalies in social media account operation data first performs data dimensionality reduction by combining the fluctuation of data sampling values under multiple sub-dimensions for each operational data dimension, thus determining data feature values that are more representative of the time-series changes in the data. Then, based on the regular patterns exhibited by trends, periodicity, and inherent dependencies in the operational data of social media accounts, the prediction residuals are determined according to the prediction errors of the data feature values. Furthermore, based on the prediction residuals, the importance of the feature data represented by the deviation of data feature values for different social media accounts is combined to comprehensively characterize the data regularity of social media accounts within the current time window, resulting in higher accuracy in diagnosing operational data anomalies based on data regularity.
[0080] This application also provides an automatic diagnostic system for abnormal operational data of social media accounts. Please refer to [link / reference]. Figure 2 The diagram illustrates a structural diagram of an automatic diagnostic system for abnormal operational data of social accounts provided in an embodiment of the present invention. The system includes: a data acquisition module 201, a first determination module 202, a second determination module 203, and an abnormal diagnosis module 204.
[0081] Data acquisition module 201 is used to collect data sample values for each social account in each sub-dimension of each operational data dimension at each sampling moment within the current time window; the operational data dimensions include user data dimension and behavioral data dimension;
[0082] The first determining module 202 is used to perform data dimensionality reduction based on the temporal fluctuation correlation of data sample values between user data dimension and behavior data dimension under each sub-dimension within the current time window, and to determine the data feature value of each operational data dimension of each social account at each sampling time; and to determine the prediction residual of each operational data dimension of each social account at each sampling time based on the prediction error of the data feature value.
[0083] The second determining module 203 is used to determine the importance of the feature data of each social account at each sampling time based on the deviation of the data feature values of each social account from other social accounts in various operational data dimensions; and to determine the data regularity within the current time window based on the distribution of the feature data importance of each social account at all sampling times and the relative deviation of the prediction residual.
[0084] The anomaly diagnosis module 204 is used to diagnose operational data anomalies based on data patterns.
[0085] It should be noted that the system provided in the above embodiments is only an example of the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the computer device can be divided into different functional modules to complete all or part of the functions described above. In addition, the automatic diagnosis system for abnormal operation data of social media accounts and the automatic diagnosis method for abnormal operation data of social media accounts provided in the above embodiments belong to the same concept, and their specific implementation process can be found in the method embodiment, which will not be repeated here.
[0086] This application also provides a computer device; please refer to [link / reference]. Figure 3 The diagram illustrates a computer device structure according to an embodiment of the present invention. The computer device includes a memory 301, a processor 302, and a computer program 303 stored in the memory 301 and running on the processor 302. When the processor 302 executes the computer program 303, the computer device can execute any of the aforementioned automatic diagnostic methods for abnormal operation data of social accounts.
[0087] This application also provides a computer program product that, when run on a computer device, enables the computer device to execute any of the aforementioned methods for automatically diagnosing abnormal operational data of social accounts.
[0088] This application also provides a computer-readable storage medium storing computer program code. When the computer program code is run on a computer device, the computer device can execute any of the aforementioned methods for automatically diagnosing abnormal operational data of social accounts.
[0089] In the embodiments provided in this application, it should be understood that the computer device, computer program product and computer-readable storage medium provided are all used to perform the corresponding methods provided above, and therefore the beneficial effects they can achieve can be referred to the beneficial effects of the methods provided above, which will not be repeated here.
[0090] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0091] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.
Claims
1. A method for automatically diagnosing abnormal operational data of social media accounts, characterized in that, The method includes: At each sampling moment within the current time window, data sample values are collected for each social media account under each sub-dimension of each operational data dimension; the operational data dimension includes user data dimension and behavioral data dimension. Data dimensionality reduction is performed based on the temporal fluctuation correlation of data sample values between user data dimension and behavioral data dimension under each sub-dimension within the current time window. The data feature value of each operational data dimension of each social account at each sampling time is determined. Based on the prediction error of the data feature value, the prediction residual of each operational data dimension of each social account at each sampling time within the current time window is determined. At each sampling time, the importance of the feature data of each social account at each sampling time is determined based on the deviation of the data feature values of each social account from other social accounts in various operational data dimensions; the data regularity within the current time window is determined based on the distribution of the feature data importance of each social account at all sampling times and the relative deviation of the prediction residual. Diagnose operational data anomalies based on the aforementioned data patterns; The process of obtaining the prediction residual includes: predicting the data feature values of each operational data dimension of each social account before each sampling time using an ARIMA model to determine the predicted feature values of each operational data dimension of each social account at each sampling time; normalizing the difference between the data feature values and the predicted feature values to determine the prediction residual of each operational data dimension of each social account at each sampling time within the current time window. The process of obtaining the importance of the feature data includes: sequentially taking each social media account as the target account and other social media accounts outside the target account as reference accounts; at each sampling moment within the current time window, performing negative correlation normalization on the differences between the data feature values of the target account in the user data dimension and the data feature values of each reference account in the user data dimension to determine the consistency of user features of each reference account in the user data dimension; normalizing the differences between the data feature values of the target account in the behavioral data dimension and the data feature values of each reference account in the behavioral data dimension to determine the corresponding degree of behavioral deviation; at each sampling moment within the current time window, determining the importance of the feature data of the target account at each sampling moment based on the overall magnitude of the consistency of user features of all reference accounts and the degree of behavioral deviation.
2. The method for automatic diagnosis of abnormal operational data of a social media account according to claim 1, characterized in that, The process of obtaining the data feature values includes: For each social media account, arrange all data sample values of each sub-dimension in each operational data dimension in chronological order within the current time window to determine the corresponding data sample value sequence; The user data dimension and the behavioral data dimension are used as target data dimensions in sequence, and each sub-dimension in the target data dimension is used as a target sub-dimension in sequence; another operational data dimension other than the target data dimension is used as a reference data dimension. Based on the fluctuation of the data sample value sequence corresponding to each sub-dimension of the reference data dimension and the trend correlation between the corresponding data sample value sequence and the data sample value sequence corresponding to the target sub-dimension, the corresponding weighted consistency is determined. The dimensionality reduction importance of the target data dimension under the target sub-dimension is determined based on the cumulative weighted consistency value corresponding to all sub-dimensions of the reference data dimension. PCA dimensionality reduction is performed by weighting the importance of each operational data dimension of each social account under each sub-dimension as the weight, and the data feature value of each operational data dimension of each social account at each sampling time is determined after dimensionality reduction.
3. The method for automatic diagnosis of abnormal operational data of a social media account according to claim 2, characterized in that, The process of obtaining the weighted consistency includes: The variance of all data sample values in the data sample value sequence corresponding to each sub-dimension of the reference data dimension is normalized to determine the corresponding contribution weight value; the Pearson correlation coefficient between the data sample value sequence corresponding to each sub-dimension of the reference data dimension and the data sample value sequence corresponding to the target sub-dimension is normalized to determine the corresponding data consistency; and the weighted consistency corresponding to each sub-dimension of the reference data dimension is determined based on the product of the contribution weight value and the data consistency.
4. The method for automatic diagnosis of abnormal operational data of a social media account according to claim 1, characterized in that, The process of determining the importance of the target account's feature data at each sampling time based on the overall magnitude of the consistency of user characteristics and the degree of behavioral deviation across all reference accounts includes: The importance of reference data for each reference account at each sampling time within the current time window is determined by multiplying the degree of behavioral deviation with the consistency of user characteristics. Within the current time window, the importance of feature data for the target account at each sampling time is determined by summing the reference data importance of all reference accounts at each sampling time.
5. The method for automatic diagnosis of abnormal operational data of a social media account according to claim 1, characterized in that, The process of obtaining the data regularity includes: At each sampling time, the difference between the prediction error under the user data dimension and the prediction error under the behavioral data dimension of each social account is negatively correlated and normalized to determine the corresponding data stability. Based on the stability of the data and the importance of the feature data, the local regularity of each social account at each sampling time is determined; Based on the cumulative value of the local regularity of each social account across all sampling times, the data regularity of each social account within the current time window is determined.
6. The method for automatic diagnosis of abnormal operational data of a social media account according to claim 5, characterized in that, The process of obtaining the local regularity includes: The importance of the feature data of each social account at each sampling time is normalized to determine the corresponding importance contribution value; the local regularity of each social account at each sampling time is determined based on the product between the importance contribution value and the data stability.
7. The method for automatic diagnosis of abnormal operational data of a social media account according to claim 1, characterized in that, The process of diagnosing operational data anomalies based on the data regularity includes: Within the current time window, social media accounts whose data regularity is less than a preset regularity threshold will be classified as accounts with abnormal operational data.
8. An automatic diagnostic system for abnormal operational data of social media accounts, characterized in that, The system includes: The data acquisition module is used to collect data sample values for each social account under each sub-dimension of each operational data dimension at each sampling moment within the current time window; the operational data dimensions include user data dimensions and behavioral data dimensions. The first determining module is used to perform data dimensionality reduction based on the temporal fluctuation correlation of data sample values between user data dimension and behavior data dimension in each sub-dimension within the current time window, and to determine the data feature value of each operational data dimension of each social account at each sampling time; and to determine the prediction residual of each operational data dimension of each social account at each sampling time based on the prediction error of the data feature value. The second determination module is used to determine the importance of the feature data of each social account at each sampling time based on the deviation of the data feature values of each social account from other social accounts in various operational data dimensions; and to determine the data regularity within the current time window based on the distribution of the feature data importance of each social account at all sampling times and the relative deviation of the prediction residual. The anomaly diagnosis module is used to diagnose operational data anomalies based on the regularity of the data. The process of obtaining the prediction residual includes: predicting the data feature values of each operational data dimension of each social account before each sampling time using an ARIMA model to determine the predicted feature values of each operational data dimension of each social account at each sampling time; normalizing the difference between the data feature values and the predicted feature values to determine the prediction residual of each operational data dimension of each social account at each sampling time within the current time window. The process of obtaining the importance of the feature data includes: sequentially taking each social media account as the target account and other social media accounts outside the target account as reference accounts; at each sampling moment within the current time window, performing negative correlation normalization on the differences between the data feature values of the target account in the user data dimension and the data feature values of each reference account in the user data dimension to determine the consistency of user features of each reference account in the user data dimension; normalizing the differences between the data feature values of the target account in the behavioral data dimension and the data feature values of each reference account in the behavioral data dimension to determine the corresponding degree of behavioral deviation; at each sampling moment within the current time window, determining the importance of the feature data of the target account at each sampling moment based on the overall magnitude of the consistency of user features of all reference accounts and the degree of behavioral deviation.