Mining Twitter Messages for Software Evolution
Emitzá Guzmán, Mohamed Hamza Ibrahim, Martin Glinz · 2017
Twitter is a widely used social network. Previous research showed that users engage in Twitter to communicate about software applications via short messages, referred to as tweets, and that some of these tweets are relevant for software evolution. However, a manual analysis is impractical due to the large number of tweets - in the range of thousands per day for popular apps. In this work we present ALERTme, an approach to automatically classify, group and rank tweets about software applications. We apply machine learning techniques for automatically classifying tweets requesting improvements, topic modeling for grouping semantically related tweets and a weighted function for ranking tweets according to their relevance for software evolution. We ran our approach on 68,108 tweets from three different software applications and compared the results against practitioners' assessments. Our results are promising and could help incorporate short, informal user feedback with social components into the software evolution process.