Sentiment Analysis on Reviews of Mobile Users

Abstract

In recent years, the dramatic increase of smartphone and tablet applications has allowed users to comment on various service platforms at any time through mobile internet, social media, cloud computing, and etc. While unfortunately, up to now, very few studies of classification methods have been applied in such area. In this paper, we concluded the following unique characteristics through more than 1,400,000 real mobile application reviews: (1) Short average length; (2) Large span of length; (3) Power-law distribution and (4) Significant difference in polarity. Based on above mentioned characteristics, a series of comparative experiments have been done for emotion classifications through classification algorithms, feature representations and review length, respectively.