Exploration of robust features for multiclass emotion classification

Classification of emotion from sentences requires the classifier to be trained on relevant features. This paper focuses on different features (a) Bag-of-Words (b) Part-of-Speech tags (c) Sentence Length and (d) Lexical Emotion Features. Extensive evaluation on variable feature length for classifying textual emotions is carried out to understand their role in model performance. Experiments depict that the bag-of-words provide better accuracy as boolean representation of feature rather than as term-frequency.