Predicting Personality Based On Social Media Behavior Of Mobile

Research shows that there is a strong relationship between a person's character and its behavior in the real world.With the development of mobile Internet,in particular the development of mobile social software,the role of smart phones has grown from a simple communication tool into an essential social tool.Predicting user's personality with its social data on mobile terminal is a viable and meaningful research.In recent years,foreign scholars have done some attempts on personality prediction with internet social data,and achieved some results.In domestic,the study in this field is still relatively few.This paper attempts to analysis and forecast users' personality with their social data on mobile terminal.The main work is as follows:First,experiment recruitment.We recruited 231 users from Sina Weibo,one of the most popular social media in China,and do some filtrations on these users.In this study,our criteria of user filtrations are: 1)the number of articles published are not less than 100 in Weibo;2)utilization of mobile terminals accounted for not less than 90%.After the above filtrations,we got 86 qualified users.Second,user personality labeling.For 86 qualified users,we carried out the Big Five personality test.Big Five personality is the most widely accepted personality model in the field of psychology of personality.We use the "personality factor" to analyzing user responses to personality questionnaires,and accordingly labeled personality of users.Third,user data acquisition.There are some differences between getting Weibo user behavior information and normal access web pages.In order to achieve the user's data acquisition,we have implemented a web crawler program.The program implements simulating user's login,crawling user's behavior data and some other functions.Fourth,feature extraction.To be able to predict the user's personality,we extracted two kinds of features for each user.1)behavior features.We collected user's social behavior on Weibo,including user registration information,the number of status posts,the number of @ use,the number of followers and followings,and so on.2)text features.We did some pretreatment on text information which is posted by users on Weibo,and converted text information into personality-related features with psychology dictionary.In this process,we also analyzed the characteristics of mobile user behavior and extracted some features.Fifth,personality prediction.We expressed all the features into vectors.Then we used logistic regression and SVM to predict user's personality.Sixth,analysis and comparison of prediction results.After comparison and analysis,we found out that the accuracy of our predictions on all personality dimensions is about 70%.Furthermore,we analyzed the impact of behavioral features and text features on prediction results.Finally,we concluded our research and discussed future work.