Introduction to the special issue on social web mining

Social web platforms such as Facebook, Twitter, and Flickr are used by millions of users daily, leading to a vast volume of semantically rich and dynamically updated social data. This data richness represents an unprecedented opportunity for any researcher interested in modeling human interactions. Making sense of the data generated from social web platforms can also be important for developing new technologies and applications including sentiment analysis for elections, ad serving, and social recommender systems. Various methods have been developed at commercial and academic laboratories for these purposes. This special issue consists of eight articles accepted by ACM Transactions on Intelligent Systems and Technology Special Issue on Social Web Mining. The articles have been selected according to a strict review process. There were in total 33 submissions to this special issue, and each submission received peer reviews by three reviewers. The accepted articles cover topics in mining the social web, including new approaches for prediction tasks (e.g., link predictions, tag recommendations, user preferences) and community detection for social networks. They also cover new approaches for topic modeling in social media including dynamically modeling sentiments and topics for social networks considering the underlying dynamics and personalized emerging topic detection. The complex, highly dynamic temporal behavior of topics is clearly one of the distinguishing aspects of social media. Not only it is important to determine topics and users' views at any given moment, but also to track and describe their evolution over time in an interpretable manner. Four of the contributions to the special issue deal with this temporal aspect. The first such article, " Dynamic Joint Sentiment-Topic Model " , by He, Lin, Gao, and Wong, describes methods for dynamically building and incrementally adjusting the joint sentiment-topic distributions, allowing for a multiscale perspective. The second article, " Personalized Emerging Topic Detection Based on a Term Aging Model " , by Cataldi, Di Caro, and Schifanella, proposes a method for retrieving emerging topics, built on aging models for terms and topics, in real time and recommending them at the user level. Arias, Arratia, and Xuriguera in their article, " Forecasting with Twitter Data " , study how much mood analysis in Twitter helps to improve prediction in time series when used as a complement to predictors that use standard information, in particular for predicting volatility in stock markets and movie box office revenues. Finally, Lee, Caverlee, Cheng, and Sui in " Campaign …