Filtering and Polarity Detection for Reputation Management on Tweets
Viktor Hangya, Richárd Farkas · 2013
Abstract. In this paper we introduce our contribution to the RepLab 2013 – An evaluation campaign for Online Reputation Management Sys-tems challenge. We participated in the filtering and polarity detection subtasks. The task of filtering is to determine whether a tweet is related to an entity. Then we classify tweets into positive, negative or neutral classes from the entity point of view. To solve these problems we em-ployed supervised machine learning techniques. We applied several Twit-ter specific text preprocessing and features engineering methods. Besides supervised methods, we experimented with incorporating clustering in-formation as well. Our system was ranked 2nd in the filtering task and 1st in the polarity detection task.