Analyzing, classifying, and interpreting emotions in software users' tweets

Abstract

Twitter enables software developers to track users' reactions to newly released systems. Such information, often expressed in the form of raw emotions, can be leveraged to enable a more informed software release process. However, automatically capturing and interpreting multi-dimensional structures of human emotions expressed in Twitter messages is not a trivial task. Challenges stem from the scale of the data available, its inherently sparse nature, and the high percentage of domain-specific words. Motivated by these observations, in this paper we present a preliminary study aimed at detecting, classifying, and interpreting emotions in software users' tweets. A dataset of 1000 tweets sampled from a broad range of software systems' Twitter feeds is used to conduct our analysis. Our results show that supervised text classifiers (Naive Bayes and Support vector Machines) are more accurate than general-purpose sentiment analysis techniques in detecting general and specific emotions expressed in software-relevant Tweets.