If You are Happy and Know It ... Tweet

Abstract

Extracting sentiment from Twitter data is one of the fundamental problems in Social Media Analytics. The length constraint of Twitter, an average of about six words per message, renders determining the positive or negative sense of a tweet difficult even for a human judge. In this work we present a general framework for single tweet (in contrast with batches of tweets) sentiment analysis which consists of three steps: extracting tweets about a desired target, separating tweets with sentiment, and discriminating positive and negative tweets. We study the performance of a number of classical and new machine learning algorithms for classification in each step. We also show that the intrinsic sparsity of tweets enables us to perform classification with their low dimensional random projections without losing accuracy. In addition, we present weighted variants of all employed algorithms which exploit available labeling uncertainty in order to further improve classification accuracy. Finally, we show that aggregating per-tweet sentiment analysis results, improve the accuracies and make our approach a good candidate for batch tweet sentiment analysis.