A method and system for evaluating the behavior of a whole network based on data analysis

Abstract

This invention discloses a method and system for evaluating online behavior based on data analysis, belonging to the field of data processing technology. The system comprises a behavior data acquisition module, a behavior data evaluation module, and a behavior profile fusion module. This invention enhances the privacy of user behavior data by combining optimization algorithms with differential privacy algorithms, preventing user information leakage, complying with data protection regulations, and helping to increase user trust in data privacy protection. The social behavior evaluation model enhances the features of user interaction behavior data, enabling more accurate analysis of user social interaction patterns, improving the accuracy of social behavior profiles, and thus achieving better personalized recommendations and social behavior prediction. Independent analysis of user non-social behavior data forms a complete user profile, avoiding the one-sidedness of a single dimension and covering comprehensive user behavior characteristics in different scenarios.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and system for evaluating network-wide behavior based on data analysis. Background Technology

[0002] Evaluating users' overall online behavior by analyzing multi-dimensional data such as browsing, consumption, and interactions can build personalized, multi-dimensional user profiles, which can be widely applied in fields such as commerce, public services, and healthcare. However, current evaluations of users' overall online behavior face several challenges: First, insufficient data quality is a major challenge. Much data is biased, incomplete, or inaccurate, due to issues such as limited data sources and delayed updates, which can distort user profiles. Second, algorithmic bias can amplify deviations in the original data, such as biases in results caused by uneven distribution of training data, thus affecting the fairness and credibility of user profiles. Furthermore, data silos between different platforms limit data integration and sharing, making it difficult to build comprehensive and accurate user profiles.

[0003] Therefore, it is necessary to establish a precise analysis system based on users' behavior data across the entire network, to reasonably evaluate and classify user behavior, and then build highly matched user profiles to better serve the application needs of different fields, while protecting users' data privacy when the data is publicly available. Summary of the Invention

[0004] The present invention aims to provide a data analysis-based method and system for evaluating network-wide behavior, which can reasonably evaluate and classify user behavior, and then construct a highly relevant user profile.

[0005] A data analysis-based method for evaluating network-wide behavior includes the following steps:

[0006] The model acquires user social behavior datasets and user non-social behavior datasets based on the whole network behavior data acquisition model. The whole network behavior data acquisition model uses a combination of optimization algorithms and differential privacy algorithms to encrypt the acquired user social behavior datasets and user non-social behavior datasets, thereby improving the privacy of user whole network behavior data.

[0007] Data analysis is performed based on user social behavior datasets and social behavior evaluation models to obtain user social behavior evaluation profiles. The social behavior evaluation model is used to extract interactive behavior data features from the user social behavior dataset for feature enhancement, and profile feature matching is performed based on social behavior data features and enhanced interactive behavior data features to improve the accuracy of profile feature matching.

[0008] Data analysis is conducted based on user non-social behavior datasets and non-social behavior assessment models to obtain user non-social behavior assessment profiles.

[0009] The user's overall online behavior assessment profile is obtained by fusing features from the user's social behavior assessment profile and the user's non-social behavior assessment profile; the user's overall online behavior assessment profile includes profile classifications of different online behavior domains.

[0010] As a preferred technical solution of the present invention, the whole network behavior data acquisition model includes a raw data acquisition layer, a data classification layer, a data encryption layer and a data output layer;

[0011] The raw data acquisition layer is used to acquire several pieces of user behavior data;

[0012] The data classification layer is used to assign social tags based on user behavior data, matching corresponding social tags to several pieces of user behavior data; among them, social tags include social behavior data and non-social behavior data;

[0013] The data classification layer is built upon a pre-trained data classification model, which is obtained by training the model on a validated user behavior dataset.

[0014] The data encryption layer is used to encrypt several pieces of user behavior data to obtain several pieces of user behavior data to be analyzed.

[0015] The data output layer is used to combine all user behavior data to be analyzed that are tagged with social behavior data to obtain a user social behavior dataset; and to combine all user behavior data to be analyzed that are tagged with non-social behavior data to obtain a user non-social behavior dataset.

[0016] As a preferred technical solution of the present invention, the data encryption layer includes a data splitting calculation layer, a privacy budget dynamic allocation layer, and a data encryption output layer;

[0017] The data splitting calculation layer is used to average the number of user behavior data to obtain user behavior data groups Hn to be encrypted, n=1,2,...,N; where N represents the total number of user behavior data groups to be averaged, and each user behavior data group Hn to be encrypted contains at least C user behavior data.

[0018] The privacy budget dynamic allocation layer is used to allocate the privacy budget to the user behavior data packet Hn to be encrypted, and obtain the encryption scheme Jn for the user behavior data to be encrypted.

[0019] Specific steps:

[0020] In the privacy budget dynamic allocation layer, the initial privacy budget is allocated based on the total number of user behavior data entries contained in the user behavior data group Hn to be encrypted;

[0021] Construct K privacy budget allocation schemes Gnk, where each privacy budget allocation scheme Gnk represents a scheme for allocating privacy budgets to the group Hn of encrypted user behavior data to be encrypted using differential privacy technology based on the initial privacy budget;

[0022] Combine the K privacy budget allocation scheme individuals Gnk to obtain the privacy budget allocation scheme iterative population; set the maximum number of iterations, where the fitness of the privacy budget allocation scheme individual Gnk is Dnk;

[0023] The specific steps for calculating fitness are as follows:

[0024] Based on the privacy budget allocation scheme, individual Gnk is simulated to allocate the privacy budget, and the user behavior data group Hn' to be evaluated is obtained. The formula Dnk=1 / εnk+log(1 / δnk) is used, where εnk represents the average data noise corresponding to the user behavior data group Hn' to be evaluated, and δnk represents the privacy leakage probability corresponding to the user behavior data group Hn' to be evaluated.

[0025] When the maximum number of iterations is reached, output the privacy budget allocation scheme individual Gnk corresponding to the maximum fitness, which is the optimal privacy budget allocation scheme individual; based on the optimal privacy budget allocation scheme individual, output the encryption scheme Jn for the user behavior data to be encrypted;

[0026] The data encryption output layer is used to encrypt the user behavior data in the user behavior data group Hn to be encrypted according to the encryption scheme Jn of all user behavior data to be encrypted, so as to obtain the user behavior data to be analyzed.

[0027] As a preferred technical solution of the present invention, the social behavior evaluation model includes a data feature extraction layer, an interaction data feature enhancement layer, a profile feature recognition layer, and a profile output layer.

[0028] The data feature extraction layer is used to extract features from the user behavior data to be analyzed in the user social behavior dataset, and obtain social behavior data features and interaction behavior data features.

[0029] The interactive data feature enhancement layer is used to enhance the features of interactive behavior data to obtain enhanced interactive behavior data features.

[0030] The profile feature recognition layer is used to perform profile feature recognition based on all social behavior data features and enhanced interaction behavior data features to obtain a user social behavior evaluation profile.

[0031] The specific steps for constructing the portrait feature recognition layer include:

[0032] Collect several sets of portrait feature recognition training samples; each set of portrait feature recognition training samples contains several user behavior features to be matched and corresponding feature portraits; combine several sets of portrait feature recognition training samples to obtain a portrait feature recognition training set;

[0033] The portrait feature recognition training set is input into the initial collaborative filtering model for model training to obtain the initial portrait feature recognition layer; the initial portrait feature recognition layer is evaluated to obtain the initial portrait feature recognition layer model evaluation result; if the initial portrait feature recognition layer model evaluation result is passed, the initial portrait feature recognition layer is used as the portrait feature recognition layer; otherwise, the model is trained again using the portrait feature recognition training set.

[0034] The profile output layer is used to output a profile of the user's social behavior.

[0035] As a preferred technical solution of the present invention, the interactive data feature enhancement layer includes a sentiment analysis layer, a graph structure interaction connection layer, and a feature enhancement layer.

[0036] The sentiment analysis layer is used to extract sentiment features from the interactive behavior data to obtain the sentiment features of the interactive behavior data.

[0037] The graph structure interaction layer is used to simultaneously extract social behavior data features and interaction behavior data features from two-layer graph convolutional networks to obtain social interaction behavior features.

[0038] The feature enhancement layer is used to fuse the emotional features and social interaction features of the interactive behavior data to obtain enhanced interactive behavior data features.

[0039] As a preferred technical solution of the present invention, the non-social behavior evaluation model is constructed based on the trained social behavior evaluation model, including a non-social data feature extraction layer, a non-social profile feature recognition layer, and a non-social profile output layer.

[0040] The non-social data feature extraction layer is used to extract features from the user behavior data to be analyzed in the user non-social behavior dataset to obtain non-social behavior data features.

[0041] The non-social profile feature recognition layer is used to perform profile feature recognition on all non-social behavior data features to obtain a user non-social behavior evaluation profile.

[0042] The non-social profile feature recognition layer is constructed based on the profile feature recognition layer. The training method is to change the profile feature recognition training samples in the profile feature recognition training set to profile feature recognition training samples that meet the training objectives.

[0043] The non-social profile output layer is used to output a user's non-social behavior evaluation profile.

[0044] As a preferred technical solution of the present invention, the specific steps for feature fusion based on user social behavior assessment profile and user non-social behavior assessment profile include:

[0045] Feature vectors are extracted based on user social behavior assessment profiles to obtain the network-wide social behavior feature vector Tsi, i=1,2,…,I; I represents the total number of different domains to which the feature vectors belong;

[0046] Feature vectors are extracted based on user non-social behavior assessment profiles to obtain the network-wide non-social behavior feature vector Tfi;

[0047] Using the formula Zi=A1 Tsi+A2 Tfi calculates the feature vector Zi of the entire network behavior domain. A1 represents the weight value of Tsi when performing feature fusion calculation, and A2 represents the weight value of Tfi when performing feature fusion calculation.

[0048] Based on the feature vector Zi of all network behavior domains, feature output is performed to obtain the user profile Zi' of the network behavior domain.

[0049] By combining all user profiles Zi' across all online behavior domains, a comprehensive online behavior assessment profile of the user is obtained.

[0050] A data analysis-based online behavior assessment system includes:

[0051] The behavior data acquisition module includes a data acquisition unit and a data encryption unit. The data acquisition unit is used to acquire user social behavior datasets and user non-social behavior datasets based on the whole network behavior data acquisition model. The data encryption unit is used to encrypt the acquired user social behavior datasets and user non-social behavior datasets based on the whole network behavior data acquisition model using a combination of optimization algorithms and differential privacy algorithms, thereby improving the privacy of user's whole network behavior data.

[0052] The behavioral data evaluation module includes a social data evaluation unit and a non-social data evaluation unit. The social data evaluation unit is used to perform data analysis based on the user's social behavior dataset and social behavior evaluation model to obtain the user's social behavior evaluation profile. The social behavior evaluation model is used to extract interactive behavior data features from the user's social behavior dataset for feature enhancement, and to perform profile feature matching based on the social behavior data features and enhanced interactive behavior data features to improve the accuracy of profile feature matching. The non-social data evaluation unit is used to perform data analysis based on the user's non-social behavior dataset and non-social behavior evaluation model to obtain the user's non-social behavior evaluation profile.

[0053] The behavior profile fusion module includes a segmented profile output unit. The segmented profile output unit is used to perform feature fusion based on the user's social behavior evaluation profile and the user's non-social behavior evaluation profile to obtain the user's overall network behavior evaluation profile. The user's overall network behavior evaluation profile includes profile classifications of different overall network behavior domains.

[0054] The present invention has the following advantages:

[0055] 1. This invention enhances the privacy of user behavior data by combining optimization algorithms with differential privacy algorithms, preventing the leakage of user information, complying with data protection regulations, and helping to enhance users' trust in data privacy protection. The social behavior evaluation model enhances the features of users' interaction behavior data, enabling more accurate analysis of users' social interaction patterns, improving the accuracy of social behavior profiles, and thus achieving better personalized recommendations and social behavior predictions. Furthermore, it independently analyzes users' non-social behavior data to form a complete user profile, avoiding the one-sidedness of a single dimension, making behavior evaluation more comprehensive, and able to cover all aspects of user behavior characteristics in different scenarios.

[0056] 2. This invention extracts and weights the feature vectors of social and non-social behavior assessment profiles, comprehensively considering behavioral characteristics from different domains, avoiding biases from a single dimension, and improving the overall accuracy of network-wide behavior assessment. Through feature vector extraction and fusion, a more personalized network-wide behavior assessment profile can be constructed for each user, fully reflecting the user's behavioral characteristics in different domains, making personalized recommendations, precision marketing, and other applications more in line with the user's actual needs. The feature fusion step effectively combines the features of social and non-social behaviors, simultaneously reflecting the user's social interaction patterns and other non-social behaviors, improving the comprehensiveness and diversity of user profiles, and supporting cross-domain analysis and decision-making. Attached Figure Description

[0057] Figure 1 This is a schematic diagram of the structure of a data analysis-based network behavior evaluation system used in an embodiment of the present invention. Detailed Implementation

[0058] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of this invention.

[0059] Example 1: A data analysis-based method for evaluating network-wide behavior, comprising the following steps:

[0060] The model acquires user social behavior datasets and user non-social behavior datasets based on the whole network behavior data acquisition model. The whole network behavior data acquisition model uses a combination of optimization algorithms and differential privacy algorithms to encrypt the acquired user social behavior datasets and user non-social behavior datasets, thereby improving the privacy of user whole network behavior data.

[0061] The whole network behavior data acquisition model includes a raw data acquisition layer, a data classification layer, a data encryption layer, and a data output layer;

[0062] The raw data acquisition layer is used to acquire several pieces of user behavior data. By collecting user behavior data through the raw data acquisition layer and processing it through steps such as classification and encryption, the data acquisition and processing process is optimized, data utilization efficiency and analysis effect are improved, and a reliable solution is provided for large-scale data processing. The network-wide behavior data is acquired based on publicly available and legal data.

[0063] The data classification layer is used to assign social tags based on user behavior data, matching corresponding social tags to several pieces of user behavior data; among them, social tags include social behavior data and non-social behavior data;

[0064] The data classification layer is built upon a pre-trained data classification model, trained using a validated user behavior dataset. This pre-trained model accurately categorizes user behavior data into social and non-social behaviors, providing a clear data structure for subsequent behavior analysis and feature extraction, and avoiding analytical biases caused by mixed data. By separating user behavior data according to social and non-social labels, more accurate and personalized analysis results can be provided when evaluating user behavior, supporting more detailed user profiles and behavior predictions.

[0065] The data encryption layer is used to encrypt several pieces of user behavior data to obtain several pieces of user behavior data to be analyzed.

[0066] The data output layer is used to combine all user behavior data to be analyzed that are tagged with social behavior data to obtain a user social behavior dataset; and to combine all user behavior data to be analyzed that are tagged with non-social behavior data to obtain a user non-social behavior dataset.

[0067] The data encryption layer includes a data splitting and computation layer, a privacy budget dynamic allocation layer, and a data encryption output layer;

[0068] The data splitting calculation layer is used to average and group several user behavior data into user behavior data groups Hn to be encrypted, n=1,2,...,N; where N represents the total number of user behavior data groups to be averaged and each user behavior data group Hn to be encrypted contains at least C user behavior data; the specific value of C is set by professional technicians according to the actual situation.

[0069] The privacy budget dynamic allocation layer is used to allocate the privacy budget to the user behavior data packet Hn to be encrypted, and obtain the encryption scheme Jn for the user behavior data to be encrypted.

[0070] Specific steps:

[0071] In the privacy budget dynamic allocation layer, the initial privacy budget is allocated based on the total number of user behavior data entries contained in the user behavior data group Hn to be encrypted; the initial privacy budget is set by professional technicians according to the actual situation, and is generally the total amount of privacy noise to be added;

[0072] Construct K privacy budget allocation schemes Gnk, where each privacy budget allocation scheme Gnk represents a scheme for allocating privacy budgets to the group Hn of encrypted user behavior data to be encrypted using differential privacy technology based on the initial privacy budget;

[0073] Combine the K privacy budget allocation scheme individuals Gnk to obtain the privacy budget allocation scheme iterative population; set the maximum number of iterations, where the fitness of the privacy budget allocation scheme individual Gnk is Dnk;

[0074] The specific steps for calculating fitness are as follows:

[0075] Based on the privacy budget allocation scheme, individual Gnk is simulated to allocate the privacy budget, and the user behavior data group Hn' to be evaluated is obtained. The formula Dnk=1 / εnk+log(1 / δnk) is used, where εnk represents the average data noise corresponding to the user behavior data group Hn' to be evaluated, and δnk represents the privacy leakage probability corresponding to the user behavior data group Hn' to be evaluated.

[0076] When the maximum number of iterations is reached, output the privacy budget allocation scheme individual Gnk corresponding to the maximum fitness, which is the optimal privacy budget allocation scheme individual; based on the optimal privacy budget allocation scheme individual, output the encryption scheme Jn for the user behavior data to be encrypted;

[0077] The data encryption output layer is used to encrypt the user behavior data in the user behavior data group Hn to be encrypted according to the encryption scheme Jn of all user behavior data to be encrypted, so as to obtain the user behavior data to be analyzed.

[0078] The dynamic privacy budget allocation layer can flexibly allocate the privacy budget based on the size and characteristics of each user behavior data group, ensuring that data protection is maximized while information loss is minimized during encryption, thus improving the efficiency and effectiveness of privacy protection. By constructing multiple privacy budget allocation schemes and iteratively optimizing these schemes, the optimal privacy budget allocation scheme can be selected based on the fitness evaluation results. This process ensures the best balance between privacy protection and data analysis needs, effectively reducing the risk of privacy leakage. The dynamic allocation strategy adopted in the privacy budget allocation process makes the allocation of the privacy budget more refined and adaptable, and can be customized according to the needs of different groups, ensuring the privacy protection effect in different scenarios. The use of differential privacy technology combined with dynamic privacy budget allocation not only reduces the impact of data noise, but also effectively reduces the probability of privacy leakage, providing strong privacy protection for user data and helping to comply with privacy protection regulations and standards.

[0079] Data analysis is performed based on user social behavior datasets and social behavior evaluation models to obtain user social behavior evaluation profiles. The social behavior evaluation model is used to extract interactive behavior data features from the user social behavior dataset for feature enhancement, and profile feature matching is performed based on social behavior data features and enhanced interactive behavior data features to improve the accuracy of profile feature matching.

[0080] The social behavior assessment model includes a data feature extraction layer, an interaction data feature enhancement layer, a profile feature recognition layer, and a profile output layer;

[0081] The data feature extraction layer is used to extract features from the user behavior data to be analyzed in the user social behavior dataset, and obtain social behavior data features and interaction behavior data features.

[0082] The interactive data feature enhancement layer is used to enhance the features of interactive behavior data to obtain enhanced interactive behavior data features.

[0083] The profile feature recognition layer is used to perform profile feature recognition based on all social behavior data features and enhanced interaction behavior data features to obtain a user social behavior evaluation profile.

[0084] The specific steps for constructing the portrait feature recognition layer include:

[0085] Collect several sets of portrait feature recognition training samples; each set of portrait feature recognition training samples contains several user behavior features to be matched and corresponding feature portraits; combine several sets of portrait feature recognition training samples to obtain a portrait feature recognition training set;

[0086] The portrait feature recognition training set is input into the initial collaborative filtering model for model training to obtain the initial portrait feature recognition layer; the initial portrait feature recognition layer is evaluated to obtain the initial portrait feature recognition layer model evaluation result; if the initial portrait feature recognition layer model evaluation result is passed, the initial portrait feature recognition layer is used as the portrait feature recognition layer; otherwise, the model is trained again using the portrait feature recognition training set.

[0087] The profile output layer is used to output user social behavior assessment profiles;

[0088] The data feature extraction layer extracts interaction behavior features and other relevant social behavior data features from the user's social behavior dataset, providing a foundation for subsequent feature enhancement and profile construction, and ensuring the comprehensiveness and completeness of data analysis. The interaction data feature enhancement layer optimizes and strengthens the interaction behavior data features, generating enhanced interaction behavior data features, which significantly improves the expressive power of the features and provides higher accuracy for feature matching and profile construction. The profile feature recognition layer, by combining all social behavior data features and enhanced interaction behavior data features, can accurately identify the main patterns of user behavior features, ensuring that the generated social behavior evaluation profile has high credibility and analytical value.

[0089] The interactive data feature enhancement layer includes a sentiment analysis layer, a graph structure interaction layer, and a feature enhancement layer;

[0090] The sentiment analysis layer is used to extract sentiment features from interactive behavior data to obtain the sentiment features of interactive behavior data. By extracting sentiment features from interactive behavior data through the sentiment analysis layer, we can deeply analyze the emotional state of users in social behavior. These sentiment features provide emotional dimension support for subsequent behavior analysis, which helps to better understand the motivations and emotional drives behind user behavior and improves the comprehensiveness and accuracy of behavior profiles.

[0091] The graph structure interaction layer is used to simultaneously extract interaction behavior features from social behavior data features and interaction behavior data features using a two-layer graph convolutional network, thus obtaining social interaction behavior features. By using a two-layer graph convolutional network to extract interaction behavior features from social behavior data, complex social behavior relationships and interaction patterns can be efficiently captured. The graph structure can reflect the interaction relationships between users and their position and influence in the social network. This process not only enhances the expressive power of interaction behavior data features, but also provides the model with richer contextual information.

[0092] The feature enhancement layer is used to fuse the emotional features and social interaction behavior features of the interaction behavior data to obtain enhanced interaction behavior data features. The feature enhancement layer fuses the emotional features and social interaction behavior features to form more comprehensive and profound interaction behavior data features. This feature fusion technology helps to make up for the limitations of single features, making the final enhanced interaction behavior data features more comprehensive and accurate, and improving the quality of social behavior profiles.

[0093] By combining sentiment analysis and graph structure interaction features, the model can capture users' social behavior patterns and emotional responses from multiple dimensions, providing a more nuanced social behavior profile. This multi-dimensional analysis helps to more accurately identify users' behavioral patterns, preferences, and emotional fluctuations, thus enhancing the effectiveness of personalized recommendations and precision marketing. When extracting social interaction behavior features, the dual-layer graph convolutional network can capture more complex user interaction relationships in social networks, such as users' social influence, relationship strength, and interaction density, providing a more comprehensive perspective for social network analysis. This method is applicable to various social behavior analysis scenarios, including social platform analysis, online social network recommendation, sentiment analysis, and user behavior prediction. By accurately extracting and fusing emotions and social connections from interaction behavior data, it can improve application effectiveness in complex social scenarios.

[0094] Data analysis is conducted based on user non-social behavior datasets and non-social behavior assessment models to obtain user non-social behavior assessment profiles.

[0095] The non-social behavior assessment model is built on the pre-trained social behavior assessment model and includes a non-social data feature extraction layer, a non-social profile feature recognition layer, and a non-social profile output layer.

[0096] The non-social data feature extraction layer is used to extract features from the user behavior data to be analyzed in the user non-social behavior dataset to obtain non-social behavior data features.

[0097] The non-social profile feature recognition layer is used to perform profile feature recognition on all non-social behavior data features to obtain a user non-social behavior evaluation profile.

[0098] The non-social profile feature recognition layer is constructed based on the profile feature recognition layer. The training method is to change the profile feature recognition training samples in the profile feature recognition training set to profile feature recognition training samples that meet the training objectives.

[0099] The non-social profile output layer is used to output a user's non-social behavior evaluation profile;

[0100] The non-social data feature extraction layer extracts key features from users' non-social behavior data, enabling in-depth capture of users' behavioral patterns in non-social contexts. This provides rich foundational data for subsequent profile construction, ensuring the comprehensiveness and accuracy of non-social behavior profiles. The non-social behavior evaluation model is built upon a pre-trained social behavior evaluation model. This cross-domain model building approach better combines the commonalities and differences between social and non-social behaviors, providing a more unified evaluation framework and improving the comprehensive analysis capability of user behavior. The non-social profile feature recognition layer effectively transforms non-social behavior data into meaningful profile features by recognizing its features. This allows for a more accurate depiction of users' behavioral patterns in non-social contexts, providing more detailed and layered information for user profiles.

[0101] The user's overall online behavior assessment profile is obtained by fusing features from the user's social behavior assessment profile and the user's non-social behavior assessment profile; the user's overall online behavior assessment profile includes profile classifications of different online behavior domains.

[0102] The specific steps for feature fusion based on user social behavior assessment profiles and user non-social behavior assessment profiles include:

[0103] Feature vectors are extracted based on user social behavior assessment profiles to obtain the network-wide social behavior feature vector Tsi, i=1,2,…,I; I represents the total number of different domains to which the feature vectors belong;

[0104] Feature vectors are extracted based on user non-social behavior assessment profiles to obtain the network-wide non-social behavior feature vector Tfi;

[0105] Using the formula Zi=A1 Tsi+A2 Tfi calculates the feature vector Zi of the entire network behavior domain. A1 represents the weight value of Tsi when performing feature fusion calculation, and A2 represents the weight value of Tfi when performing feature fusion calculation. The specific values ​​of A1 and A2 are set by professional technicians according to the actual situation.

[0106] Based on the feature vector Zi of all network behavior domains, feature output is performed to obtain the user profile Zi' of the network behavior domain.

[0107] Combine all user profiles Zi' across all online behavior domains to obtain a comprehensive online user behavior assessment profile;

[0108] By fusing features from social and non-social behavior assessment profiles, user performance in both social and non-social domains can be considered simultaneously. This allows the overall network behavior assessment profile to comprehensively reflect user behavior patterns in different contexts, providing richer and more multi-dimensional information. Extracting feature vectors from both social and non-social behavior profiles separately clearly depicts the behavioral characteristics of each domain. Weighted fusion allows for flexible adjustment of the influence weights of social and non-social behaviors, better meeting different business needs and improving the accuracy and adaptability of the profile. Setting the weight values ​​A1 and A2 allows for flexible adjustment of the importance of social and non-social behaviors according to different application scenarios. For example, in some applications… Social behavior may be more important, while in other applications, non-social behaviors such as consumption behavior or personal preferences may be more crucial. The flexibility of this feature fusion method makes the overall online behavior assessment profile highly customizable. For example, in the field of social behavior, it includes information such as users' social interactions, social activities participated in, communication frequency, and social circles. By analyzing users' social behavior, it is possible to reveal characteristics such as users' social influence, social activity, and social needs. It is suitable for applications such as social platform user analysis, advertising, and social interaction recommendations. It can combine social behavior data such as users' posts, comments, likes, and shares to analyze social patterns, thereby predicting users' social tendencies and potential interest groups, and providing support for personalized recommendations and group analysis.

[0109] Example 2: A data analysis-based system for evaluating network-wide behavior (see [link]). Figure 1 As shown, it includes:

[0110] The behavior data acquisition module includes a data acquisition unit and a data encryption unit. The data acquisition unit is used to acquire user social behavior datasets and user non-social behavior datasets based on the whole network behavior data acquisition model. The data encryption unit is used to encrypt the acquired user social behavior datasets and user non-social behavior datasets based on the whole network behavior data acquisition model using a combination of optimization algorithms and differential privacy algorithms, thereby improving the privacy of user's whole network behavior data.

[0111] The behavioral data evaluation module includes a social data evaluation unit and a non-social data evaluation unit. The social data evaluation unit is used to perform data analysis based on the user's social behavior dataset and social behavior evaluation model to obtain the user's social behavior evaluation profile. The social behavior evaluation model is used to extract interactive behavior data features from the user's social behavior dataset for feature enhancement, and to perform profile feature matching based on the social behavior data features and enhanced interactive behavior data features to improve the accuracy of profile feature matching. The non-social data evaluation unit is used to perform data analysis based on the user's non-social behavior dataset and non-social behavior evaluation model to obtain the user's non-social behavior evaluation profile.

[0112] The behavior profile fusion module includes a segmented profile output unit. The segmented profile output unit is used to perform feature fusion based on the user's social behavior evaluation profile and the user's non-social behavior evaluation profile to obtain the user's overall network behavior evaluation profile. The user's overall network behavior evaluation profile includes profile classifications of different overall network behavior domains.

[0113] It should be understood that those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims. Parts not described in detail in this specification are prior art known to those skilled in the art.

Claims

1. A method for network-wide behavior evaluation based on data analysis, characterized in that, Includes the following steps: Based on the whole network behavior data acquisition model, user social behavior dataset and user non-social behavior dataset are obtained; The whole network behavior data acquisition model uses a combination of optimization algorithms and differential privacy algorithms to encrypt the acquired user social behavior dataset and user non-social behavior dataset, thereby improving the privacy of the user's whole network behavior data. The whole network behavior data acquisition model includes a raw data acquisition layer, a data classification layer, a data encryption layer, and a data output layer. The data encryption layer is used to encrypt several pieces of user behavior data to obtain several pieces of user behavior data to be analyzed. Data analysis is performed based on user social behavior datasets and social behavior evaluation models to obtain user social behavior evaluation profiles. The social behavior evaluation model is used to extract interactive behavior data features from the user social behavior dataset for feature enhancement, and profile feature matching is performed based on social behavior data features and enhanced interactive behavior data features to improve the accuracy of profile feature matching. Data analysis is conducted based on user non-social behavior datasets and non-social behavior assessment models to obtain user non-social behavior assessment profiles. The user's overall online behavior assessment profile is obtained by fusing features from the user's social behavior assessment profile and the user's non-social behavior assessment profile; the user's overall online behavior assessment profile includes profile classifications of different online behavior domains. The data encryption layer includes a data splitting and computation layer, a privacy budget dynamic allocation layer, and a data encryption output layer; The data splitting calculation layer is used to average the number of user behavior data to obtain user behavior data groups Hn to be encrypted, n=1,2,...,N; where N represents the total number of user behavior data groups to be averaged, and each user behavior data group Hn to be encrypted contains at least C user behavior data. The privacy budget dynamic allocation layer is used to allocate the privacy budget to the user behavior data packet Hn to be encrypted, and obtain the encryption scheme Jn for the user behavior data to be encrypted. Specific steps: In the privacy budget dynamic allocation layer, the initial privacy budget is allocated based on the total number of user behavior data entries contained in the user behavior data group Hn to be encrypted; Construct K privacy budget allocation schemes Gnk, where each privacy budget allocation scheme Gnk represents a scheme for allocating privacy budgets to the group Hn of encrypted user behavior data to be encrypted using differential privacy technology based on the initial privacy budget; Combine the K privacy budget allocation scheme individuals Gnk to obtain the privacy budget allocation scheme iterative population; set the maximum number of iterations, where the fitness of the privacy budget allocation scheme individual Gnk is Dnk; The specific steps for calculating fitness are as follows: Based on the privacy budget allocation scheme, individual Gnk is simulated to allocate the privacy budget, and the user behavior data group Hn' to be evaluated is obtained. The formula Dnk=1 / εnk+log(1 / δnk) is used, where εnk represents the average data noise corresponding to the user behavior data group Hn' to be evaluated, and δnk represents the privacy leakage probability corresponding to the user behavior data group Hn' to be evaluated. When the maximum number of iterations is reached, output the privacy budget allocation scheme individual Gnk corresponding to the maximum fitness, which is the optimal privacy budget allocation scheme individual; based on the optimal privacy budget allocation scheme individual, output the encryption scheme Jn for the user behavior data to be encrypted; The data encryption output layer is used to encrypt the user behavior data in the user behavior data group Hn to be encrypted according to the encryption scheme Jn of all user behavior data to be encrypted, so as to obtain the user behavior data to be analyzed.

2. The method of claim 1, wherein, The raw data acquisition layer is used to acquire several pieces of user behavior data; The data classification layer is used to assign social tags based on user behavior data, matching corresponding social tags to several pieces of user behavior data; among them, social tags include social behavior data and non-social behavior data; The data classification layer is built upon a pre-trained data classification model, which is obtained by training the model on a validated user behavior dataset. The data output layer is used to combine all user behavior data to be analyzed that are tagged with social behavior data to obtain a user social behavior dataset; and to combine all user behavior data to be analyzed that are tagged with non-social behavior data to obtain a user non-social behavior dataset.

3. The method of claim 2, wherein, The social behavior assessment model includes a data feature extraction layer, an interaction data feature enhancement layer, a profile feature recognition layer, and a profile output layer; The data feature extraction layer is used to extract features from the user behavior data to be analyzed in the user social behavior dataset, and obtain social behavior data features and interaction behavior data features. The interactive data feature enhancement layer is used to enhance the features of interactive behavior data to obtain enhanced interactive behavior data features. The profile feature recognition layer is used to perform profile feature recognition based on all social behavior data features and enhanced interaction behavior data features to obtain a user social behavior evaluation profile. The specific steps for constructing the portrait feature recognition layer include: Collect several sets of portrait feature recognition training samples; each set of portrait feature recognition training samples contains several user behavior features to be matched and corresponding feature portraits; combine several sets of portrait feature recognition training samples to obtain a portrait feature recognition training set; The portrait feature recognition training set is input into the initial collaborative filtering model for model training to obtain the initial portrait feature recognition layer; the initial portrait feature recognition layer is evaluated to obtain the initial portrait feature recognition layer model evaluation result; if the initial portrait feature recognition layer model evaluation result is passed, the initial portrait feature recognition layer is used as the portrait feature recognition layer; otherwise, the model is trained again using the portrait feature recognition training set. The profile output layer is used to output a profile of the user's social behavior.

4. The method for evaluating network-wide behavior based on data analysis according to claim 3, characterized in that, The interactive data feature enhancement layer includes a sentiment analysis layer, a graph structure interaction layer, and a feature enhancement layer; The sentiment analysis layer is used to extract sentiment features from the interactive behavior data to obtain the sentiment features of the interactive behavior data. The graph structure interaction layer is used to simultaneously extract social behavior data features and interaction behavior data features from two-layer graph convolutional networks to obtain social interaction behavior features. The feature enhancement layer is used to fuse the emotional features and social interaction features of the interactive behavior data to obtain enhanced interactive behavior data features.

5. The method for evaluating network-wide behavior based on data analysis according to claim 4, characterized in that, The non-social behavior assessment model is built on the pre-trained social behavior assessment model and includes a non-social data feature extraction layer, a non-social profile feature recognition layer, and a non-social profile output layer. The non-social data feature extraction layer is used to extract features from the user behavior data to be analyzed in the user non-social behavior dataset to obtain non-social behavior data features. The non-social profile feature recognition layer is used to perform profile feature recognition on all non-social behavior data features to obtain a user non-social behavior evaluation profile. The non-social profile feature recognition layer is constructed based on the profile feature recognition layer. The training method is to change the profile feature recognition training samples in the profile feature recognition training set to profile feature recognition training samples that meet the training objectives. The non-social profile output layer is used to output a user's non-social behavior evaluation profile.

6. The method for evaluating network-wide behavior based on data analysis according to claim 5, characterized in that, The specific steps for feature fusion based on user social behavior assessment profiles and user non-social behavior assessment profiles include: Feature vectors are extracted based on user social behavior assessment profiles to obtain the network-wide social behavior feature vector Tsi, i=1,2,…,I; I represents the total number of different domains to which the feature vectors belong; Feature vectors are extracted based on user non-social behavior assessment profiles to obtain the network-wide non-social behavior feature vector Tfi; Using the formula Zi=A1 Tsi+A2 Tfi calculates the feature vector Zi of the entire network behavior domain. A1 represents the weight value of Tsi when performing feature fusion calculation, and A2 represents the weight value of Tfi when performing feature fusion calculation. Based on the feature vector Zi of all network behavior domains, feature output is performed to obtain the user profile Zi' of the network behavior domain. By combining all user profiles Zi' across all online behavior domains, a comprehensive online behavior assessment profile of the user is obtained.

7. A data analysis-based system for evaluating network-wide behavior, characterized in that, The system employs a data analysis-based network behavior assessment method as described in any one of claims 1-6, comprising: The behavior data acquisition module includes a data acquisition unit and a data encryption unit. The data acquisition unit is used to acquire user social behavior datasets and user non-social behavior datasets based on the whole network behavior data acquisition model. The data encryption unit is used to encrypt the acquired user social behavior datasets and user non-social behavior datasets based on the whole network behavior data acquisition model using a combination of optimization algorithms and differential privacy algorithms, thereby improving the privacy of user's whole network behavior data. The behavioral data evaluation module includes a social data evaluation unit and a non-social data evaluation unit. The social data evaluation unit is used to perform data analysis based on the user's social behavior dataset and social behavior evaluation model to obtain the user's social behavior evaluation profile. The social behavior evaluation model is used to extract interactive behavior data features from the user's social behavior dataset for feature enhancement, and to perform profile feature matching based on the social behavior data features and enhanced interactive behavior data features to improve the accuracy of profile feature matching. The non-social data evaluation unit is used to perform data analysis based on the user's non-social behavior dataset and non-social behavior evaluation model to obtain the user's non-social behavior evaluation profile. The behavior profile fusion module includes a segmented profile output unit. The segmented profile output unit is used to perform feature fusion based on the user's social behavior evaluation profile and the user's non-social behavior evaluation profile to obtain the user's overall network behavior evaluation profile. The user's overall network behavior evaluation profile includes profile classifications of different overall network behavior domains.

Images