An unsupervised approach based on self-learning for the combination of sentiment analysis methods

Abstract

Sentiment analysis has become a very important tool for analysis of social media data. There are several methods developed for this research field, many of them working very differently from each other, covering distinct aspects of the problem and distinct strategies. Despite the large number of existent techniques, there is no single one which fits well in all cases and data sources. Moreover, in case of supervised approaches, it may be very hard to get labeled data for strategies that demand training, mainly for a new application. In this dissertation, we propose to combine several very popular and effective state-of-the-practice sentiment analysis methods, by means of an unsupervised bootstrapped strategy for classification of polarity. Our solution was thoroughly tested considering thirteen different datasets in several domains such as opinions, comments, and social media. The experimental results demonstrate that our combined method (aka, 10SENT) improves the effectiveness of the classification task, but more important, it solves a key problem in the field. It is consistently among the best methods in many data types, meaning that it can produce the best (or close to best) results in almost all considered contexts, without any additional costs. Our self-learning approach is also very independent of the base methods, which means that it is highly extensible to incorporate any new additional method that can be envisioned in the future. Finally, we investigate a transfer learning and active learning approach for sentiment analysis and show the potential of this technique to improve our results.