11 - Emotion-based classification through fuzzy entropy-enhanced FCM clustering

Abstract

In the era of the social web, microblogging is a popular social media that enables users to share their feelings, ideas, and comments about everyday-life events, trends, and advertisements. Great emphasis has been placed on the investigation of human behavior in social networks by analyzing user-generated content to capture feelings and emotions from natural language. Delving into human reactions to events from the social message streams allows understanding collective moods and preferences of web users and supporting market strategies as well as political consensus analysis. This paper proposes a novel approach to the emotion-based classification of microblogging messages such as Twitter. The classification method is unsupervised and exploits the well-known fuzzy c-means (FCM) clustering algorithm, proposing an enhanced version called entropy weighted FCM (shortly, EwFCM) that overcomes the main drawback of the FCM, viz., the sensitivity to the random cluster initialization by leveraging a fuzzy measure to evaluate the entropy in the data distribution. The fuzzy entropy measure allows minimizing the fuzziness of clustering, speeding up the convergence of the clustering algorithm. The proposed approach aims at analyzing microblogging-based trends to detect the main emotions expressed in the textual content, according to a well-known computational analysis of emotions whose main models are attributable to Ekman and Plutchik. The experiments reveal that the presented approach offers high accuracy in the emotion-based classification of textual resources; comparisons with the FCM algorithm applied for emotion classification show that the proposed method converges faster and provides promising classification performance, as evaluated by common metrics such as accuracy, precision, and F1-score.