Multi-Class Emotion Detection and Annotation in Malayalam Novels

Abstract

Sentiment analysis or opinion mining has been used widely in various applications like market analysis. Usually during sentiment detection the polarity of the sentiment either positive or negative is detected. Basically there are multiple classes of emotions and so emotion detection is different from sentiment analysis. Reading a novel for a visually impaired person with the help of a text to speech synthesizer is still a challenging task, since it was not possible to modulate the sound with respect to the emotion in the text or dialogue. Text to speech softwares can synthesize the speech signal to embrace the emotions if the emotion of that particular text was already annotated. Multi-Class emotion detection aims analyse different emotions hidden in the text data. Multi-class emotion classification in Indian languages was not experimented before. In this paper, an SVM classifier is used for sentence level multi-class emotion detection in Malayalam. The proposed approach uses different syntactic features such as n-gram, POS related, negation related, level related features etc, for better classification. The classifier classifies the Malayalam sentences into different emotion classes like happy, sad, anger, fear or normal etc. with level information such as high, low etc. It also states whether the sentence is dialogue, question or not for better hearing experience from a speech synthesiser while reading the novel.