Emerging Product Topics Prediction in Social Media without Social Structure Information

Abstract

Social media provides a vast continuous supply of dynamic and diverse information contents from the crowd, which serves as useful resources for predictive analytical applications. Although there exist already a number of studies on emerging topics detection, they focused on modelling of textual contents and emerging detection mechanism over topic popularity. To meet the real-life demands, prediction of emerging product topic, rather than detection, in the early stage is required. Besides, despite that some relevant studies considered social structure information, they suffer from the assumption that the complete network is available and the diffusion process only depends on social influence among members of networks. Moreover, not all social media sites provide the functionality to facilitate the development of online social networks. In this paper, we tackle the problem of emerging product topics prediction in social network with implicit networks. Two tasks, one for long-term forecast in pre-production stage and the other for short-term forecast in post-release stage, are investigated. We present a novel framework named Emerging Topics Predictor (ETP). Two novel features, namely author diversity and competition features, are also proposed to accommodate the diffusion process with implicit networks based on the rationale of product marketing. Through empirical evaluation on movie reviews from two real social media sites, ETP is shown to provide effective and efficient performance in predicting the emerging topics as early as possible. In particular, the experiment results show the promising effect of author diversity in emerging prediction. To the best of our knowledge, this work is among the very first studies on emerging product topic prediction in social media with considerations of implicit networks.