An Unsupervised Common Sense-based Learning Framework for Emotion Detection and Classification in Textual Social Data

Abstract

Abstract

Emotions are inherent to human behavior and generally expressed in response to the stimuli from the external environment or some internal processes. With the widespread use of online social platforms, people often express themselves using tweets, blogs and posts which generally include affective text. The topic of emotion detection and recognition has been widely studied in computational science and in fact a separate area “Affective Computing” is dedicated to its study. Although several techniques have been used for the purpose of detecting emotions from text, each technique has its own strengths and weaknesses. Our study focuses on employing lexical, semantic and real world information to detect six basic emotions from text. We propose a hybrid framework, for detecting and classifying emotions, which initially builds up the emotion lexicon, uses semantic information for grouping together related words and employs, a common sense platform, ConceptNet, to make the classification more accurate. The results obtained, utilizing this framework are quite encouraging and comparable to state-of-the-art techniques available.

Keywords: Emotion detection, hybrid learning, affective computing, social data, real-world knowledge

Cite this Article

Abid Hussain Wani, Rana Hashmy. An Unsupervised Common Sense-based Learning Framework for Emotion Detection and Classification in Textual Social Data. Journal of Artificial Intelligence Research & Advances. 2017; 4(3): 49–56p.