Online Behavior Pattern Recognition Based On Attention Features And Time-Frequency Features

The identification of online Web user is critical in many fields,such as business,public security and so on.Using Web browsing log information of users,the thesis intends to explore the identification of online users based on the integration of the attention features and time-frequency features of Web browsing behaviors of users.The research contents include:(1)Attention feature extraction of Web browsing behaviors of users.Regarding Web browsing behaviors of users as their attention to specific resources,an attention representation model for Web browsing behaviors is proposed.Compared with the current methods,based on this model,the features of user attention,such as intensity,stability,breadth and its distribution can be extracted from Web browsing logs of users.It can provide a new perspective for the recognition of Web user behavior patterns.(2)Time-frequency feature extraction of Web browsing behaviors of users.The records of Web browsing of users are abstracted as a kind of non-stationary random signals.Then,based on wavelet analysis,the attention signal pattern of Web browsing behaviors of users can be extracted.Moreover,the multi-scale spatial energy distribution approach is used to extract the behavior features of Web users.Experiments show that this can perceive the randomness,periodicity and trend of user behaviors.(3)Online behavior pattern recognition of Web users.Combining the attention features and time-frequency features of Web user's browsing behaviors,a matching algorithm of user's behavior is proposed,which can enable online user identification.On a dataset about Web access of students from a vocational and technical college,the approach proposed is evaluated.Experiments show that the accuracy of Web user recognition approach proposed based on the integration of attention and time-frequency features is higher than that based on attention or time-frequency features independently.