A Research Of SNS Users' Posting Behavior And Interest Prediction

Billions of users share their ideas through posting photos and texts on social network services(SNSs).They are interested in various topics,usually have different sentiment tendencies and posting activities.This paper proposed a method to characterize SNS users' posting activities with users' post type distributions instead of unique user types.As an application of our latent post types,we investigate how the last post types can improve the accuracy of predicting a user being a member of interest groups.Our research is based on Facebook and Twitter,two popular social media sites,and we build the datasets of Facebook users and Twitter users.In this paper,we propose two models,post type discovering model which characterizes users' posting behavior and user interest prediction model which predicts users' interest.In the the post type discovering model,we encode three scores(polarity,subjectivity and word count)of each post into one tuple and apply Latent Dirichlet Allocation(LDA)on the discrete score tuples of users' posts,so as to find latent patterns of users' posting activities.In this way,one user's posting activities are represented as probability distribution on the post types.In interest prediction model,we investigate how post types can improve the accuracy of predicting a user being a member of interest groups.We integrate user profiles,user behavior features,and per-user post type distribution,and combine them with semantic features extracted from users' likes pages to build an interest prediction model.The result shows that using all features including post type distribution performs better than only using likes page features,which indicates that although user behavior features are independent from the semantics of interests or topics,they can improve the prediction accuracy and post type distribution can effectively improve the accuracy of user interest presiction.The method proposed in this paper can effectively help analyze users' posting behavior and predict users' interest.It can be applied to areas such as user analysis,interest prediction and personalized recommendation system.