Sentiment analysis and topic classification based on binary maximum entropy classifiers

Abstract

This paper presents a strategy based on binary maximum entropy classifiers for automatic sentiment analysis and topic classification over Spanish Twitter data. The developed system achieved the best results for topic classification, and the second place for sentiment analysis in a joint evaluation effort — the TASS challenge. Different configurations have been explored for both tasks, leading to the use of a cascade of binary classifiers for sentiment analysis and a one-vs-all strategy for topic classification, where the most probable topics for each tweet were selected.