Predicting Single-tweet Popularity In Social Network

Centralized with users being the creators and propagators,social network tends to be an indispensable part of modern people's life,in the era of Web 2.0.Massive amount of users' thoughts and friendship are implied in social network,which becomes a promising source of big data.One of the most significant meanings for data mining is to analyze the underlined relations among data,and use it for future.In social network,the limitation of users' time and attention determines that users will only focus on what they are interested and what is popular for the time being.Predicting what is popular in time will not only improve the utilization of users' time and attention,but also benefit social websites to offer better service to their users.In this chapter,we intend to research on the popularity prediction of textual content,using big data in social network.We focus on methods and models of prediction,which are well classified by elements the models consider,such as user behaviors,the life cycles of information,and the social network topology.We also reveal researchers' work on classifying social networks,evaluating metrics,as well as feature selection,and what remains to be done.Although a few topic or event prediction models have been proposed in the past few years,researches that focus on the single tweet prediction just emerge recently.Therefore,we further dig into predicting the popularity of single tweet with STH-Bass,a Spatial and Temporal Heterogeneous Bass model derived from economic field,to predict the popularity of a single tweet.Leveraging only the first day's information after a tweet is posted,STH-Bass can not only predict the trend of a tweet with favorite count and retweet count,but also classify whether the tweet will be popular in the future.We perform extensive experiments to evaluate the efficiency and accuracy of STH-Bass based on real-world Twitter data.The evaluation results show that STH-Bass obtains much less APE than the baselines when predicting the trend of a single tweet,and an average of 24% higher precision when classifying the tweets popularity.