Polarity analysis of micro reviews in foursquare

Abstract

On Foursquare, one of the currently most popular location-based social networks, users can not only share which places (venues) they visit but also leave short comments (tips) about their previous experiences at specific venues. Tips may provide a valuable feedback for business owners as well as for potential new customers. Sentiment or polarity classification provides useful tools for opinion summarization, which can help both parties to quickly obtain a predominant view of the opinions posted by users at a specific venue. We here present what, to our knowledge, is the first study of polarity of Foursquare tips. We start by characterizing two datasets of collected tips with respect to their textual content. Some inherent characteristics of tips, such as short sizes as well as informal and often noisy content, pose great challenges to polarity detection. We then investigate the effectiveness of four alternative polarity classification strategies on subsets of our dataset. Three of the considered strategies are based on supervised machine learning techniques and the fourth one is an unsupervised lexicon-based approach. Our evaluation indicates that effective polarity classification can be achieved even if the simpler lexicon-based approach, which does not require costly manual tip labeling, is adopted.