Detection Of User Abnormal Interaction Behavior Based On Social Network Content Mining

Detecting the abnormal interaction behavior of social network users is an important research content of social network data mining.The main challenge is how to use the appropriate anomaly detection algorithm in data mining to analyze and detect the abnormal interaction behavior of social network users.In recent years,the traditional anomaly detection algorithm has made outstanding achievements in the detection of abnormal social network users' abnormal account behavior.However,when analyzing large-scale complex social networks,the detection target cannot be effectively completed.The thesis will detect the abnormal interaction behavior of users in social networks as the research goal.Firstly,a social network multi-user anomaly interaction behavior detection algorithm based on community similarity is designed and implemented to detect the community with multi-user abnormal interaction behavior.A single-user abnormal interaction behavior detection algorithm based on graph embedded in social network is implemented to complete the single-user abnormal interaction behavior detection in social network.Finally,a single-user abnormal interaction behavior detection optimization algorithm for large-scale social network is designed and implemented to solve large-scale social interaction.The device performance limitations brought by the network to the algorithm.The specific research work is as follows:(1)Firstly,the social network graph is established according to the user interaction relationship by using the official social network data set;then the Louvain community discovery algorithm is used to divide the social network graph into multiple sub-communities according to the user connection relationship of the social network graph;then each is calculated The quality scores of the user nodes and edges in the community graph,using the local hash to filter the quality scores of the user nodes and edges in each community graph as the equal length binary features of each group of progressive comparisons,and finally based on the similarity Score to compare thresholds to obtain a community graph with multi-user anomalous interactions.The feasibility and effectiveness of the algorithm in detecting communities with multi-user anomalous interactions are verified by experiments.(2)Firstly,the user nodes in the community are embedded into the matrix space.The weight between the user nodes in the figure represents the interaction strength between the users.Then the final embedded model of the user nodes is completed by the gradient descent method,and according to the user node model.Define a model of user anomaly interaction behavior.Through the published social network dataset,the other anomaly detection algorithms and the accuracy and efficiency of the algorithm are compared and analyzed.(3)Firstly,the user node model is embedded for dimensionality optimization,and the end user node embedding model constructed by gradient descent is optimized by the uncertain line search.The abnormal user node is inferred according to the user node model substituting the optimized single user abnormal interaction behavior model.The test scale is expanded on the dataset to form large-scale social network graph data for comparative analysis and verification.The experimental verification model can effectively optimize the algorithm efficiency.