Viscovery: Trend Tracking in Opinion Forums based on Dynamic Topic\n Models

Opinions in forums and social networks are released by millions of people due\nto the increasing number of users that use Web 2.0 platforms to opine about\nbrands and organizations. For enterprises or government agencies it is almost\nimpossible to track what people say producing a gap between user\nneeds/expectations and organizations actions. To bridge this gap we create\nViscovery, a platform for opinion summarization and trend tracking that is able\nto analyze a stream of opinions recovered from forums. To do this we use\ndynamic topic models, allowing to uncover the hidden structure of topics behind\nopinions, characterizing vocabulary dynamics. We extend dynamic topic models\nfor incremental learning, a key aspect needed in Viscovery for model updating\nin near-real time. In addition, we include in Viscovery sentiment analysis,\nallowing to separate positive/negative words for a specific topic at different\nlevels of granularity. Viscovery allows to visualize representative opinions\nand terms in each topic. At a coarse level of granularity, the dynamic of the\ntopics can be analyzed using a 2D topic embedding, suggesting longitudinal\ntopic merging or segmentation. In this paper we report our experience\ndeveloping this platform, sharing lessons learned and opportunities that arise\nfrom the use of sentiment analysis and topic modeling in real world\napplications.\n