A semantic approach in the lexicon-based feature selection for emotion detection

Emotion detection is a task of sentiment analysis that deals with the extraction and analysis of emotions in texts. Recognizing implicit emotions is one of the main challenges in keyword or lexicon-based approaches. This paper presents a hybrid emotion detection approach, which combines lexicon-based emotion-relevant feature selection with a classical learning-based approach to determine the emotion. The proposed feature selection process focuses on capturing the emotional meaning of the text by computing the semantic relationship between its content and the lexicon vocabulary, with the goal of increasing implicit emotion recognition. The proposed solution was evaluated on the classification of emotions in Spanish tweets included in the AIT corpus, with different alternatives to compute the semantic relation and several classification algorithms, obtaining very promising results.