Research On User Behavior Analysis Of Campus Network Based On Graph Theory Subspace

Campus network user behavior prediction is one of the important ways to study students’ mental health.By analyzing student users’ online behavior,the psychological changes of student users can be inferred,so as to analyze campus public opinion and provide intervention means for relevant departments to ensure campus safety.The user behavior data in campus network has many redundances,high dimensions and miscellaneous attributes.In the process of analysis,the traditional campus network user behavior analysis method is prone to problems such as insufficient samples after screening and single user dimension analysis,resulting in inaccurate behavior clustering results.To solve the above problems,this paper uses the optimized graph theory subspace clustering algorithm to identify campus network users’ online behaviors.For the campus network student users of Henan University of Technology,this paper carries out the user behavior analysis research based on deep sparse autoencoder,and introduces the graph theory method to carry the cutting information characteristic to represent the original manifold data.On this basis,a graph theory subspace clustering model based on deep sparse autoencoder is established to improve the reliability and robustness of campus network user behavior clustering,and to achieve accurate analysis of campus network user behavior.The main research contents are as follows:(1)To address the problem of using a single attribute of the dataset resulting in one-sided results of user behavior clustering,a multi-attribute fusion weighting MFW(Multi-attribute fusion Weighting)method is proposed to solve the problem of non-uniform nature among attributes.The data are pre-processed by normalization and transformed into unitless attributes,which improves the readability of the data and largely preserves the inherent characteristics of different attributes.The obtained data are manipulated in a weighted form,so that data with close but not comparable links between attributes in the behavior log can be better processed.(2)To address the problem of low accuracy rate of user behavior clustering in campus network,we propose a DGSC(Deep Graph Subspace Clustering)method based on deep sparse self-encoder graph theory subspace clustering,which innovatively solves the difficulties of too high data dimensionality and too large data volume affecting the clustering effect,and improves the accuracy rate of user behavior clustering.The deep sparse self-encoder four-layer model is used to process the behavioral data,extract the internal deep data structure,and integrate different attributes to achieve dimensionality reduction.The graph theory method is used to transform the data objects,and subspace clustering is performed on this basis to obtain the campus network user behavior clustering results.Based on the behavioral data provided by the information center of Henan University of Technology,we analyze the Internet behavior of campus network users.Experimental verification shows that the proposed DGSC algorithm can obtain accurate results of campus network user behavior clustering,and the accuracy of the algorithm can reach up to 92%.(3)To address the problem of inefficient processing time of campus network users’ online behavior habits,we propose a Key Attribute Frequent Pattern KAFP-Growth(Key Attribute Frequent Pattern Growth)association rule algorithm,which solves the problem of difficulty in determining the minimum support threshold in association rules,and improves the time efficiency in user behavior habit preference mining.The algorithm has improved the time efficiency of user behavior and preference mining.By mining the correlation of campus network users’ access to websites,we can obtain users’ online behavior habits and deeply grasp the online behavior patterns of campus network users.