Research On Predictive Behavior Based On User Access Behavior

With the popularization and application of Web2.0 technology,various types of online social networking sites have sprung up.With the help of these social networking sites,users can share content,expressing opinions,establish intimate relationship,thus social networking sites play a good role on enriching peoples' emotion,culture and entertainment.At the same time,users left a large number of traces on the social networking sites.Based on these behavioral information,it is very important to excavate the online behavior regular of users,predict the online behavior of users,which is of great significance to public opinion analysis,network security,social security,information recommendation and commodity marketing.This paper takes advantage of a college social network users to access data,analyzes the user's social network access to behavioral characteristics and reveals the inherent mechanism behind the behavior of users online.The specific researches and contributions include:(1)Paper analyzes the time interval distribution of the two consecutive online access behaviors based on the individual level and the population level,studies the correlation and activeness of the user time interval sequence,and reveals the internal mechanism behind it;(2)This paper analyzes the memory characteristics of user access behavior and finds that the online behavior of users have a strong short memory,and its distribution is subject to Gaussian distribution.Then establishes the Markov process model,uses to explain the user access behavior in the memory characteristics;(3)Based on the characteristics of access behavior that has found,this paper conducts a time series prediction study of user access behavior.Aim at the data of users access history,uses the ARIMA model and Holt-Winters three-parameter exponential smoothing method to predict the time series of the click flow,by establishing a model to predict the future trend of the data,then compares and analyses the advantages and disadvantages of two model.