Two-Stage Text Feature Selection Method for Human Emotion Recognition

Abstract

In this paper, two-stage text feature selection method is proposed to identify significant features to effectively recognize the human emotions from the unstructured text documents. The proposed method employs two-stage feature filtering mechanism, namely, semantic, and statistical stage. The first stage consists of semantic-based method which extracts the meaningful words from the unstructured text data using parts of the speech (PoS) tagger. It identifies the noun, verb, adverb, and adjective as prospective words for detecting text-based human emotions. The second stage employs chi-square method to remove the weak semantic features with lower statistical score. The effectiveness of the two-stage feature selection method is evaluated and compared with existing methods with support vector machine (SVM) classifier on the publically available and widely accepted ISEAR dataset. The results obtained from the analysis indicate that the SVM classifier with two-stage method has achieved 10.6, 15.46, and 34.45 improvement in emotion recognition rate as compared with the single-stage methods such as PoS method, method, and baseline.