NBLex: emotion prediction in Kannada-English code-switch text using naïve bayes lexicon approach

Abstract

Emotion analysis is a process of identifying the human emotions derived from the various data sources. Emotions can be expressed either in monolingual text or code-switch text. Emotion prediction can be performed through machine learning (ML), or deep learning (DL), or lexicon-based approach. ML and DL approaches are computationally expensive and require training data. Whereas, the lexicon-based approach does not require any training data and it takes very less time to predict the emotions in comparison with ML and DL. In this paper, we proposed a lexicon-based method called non-binding lower extremity exoskeleton (NBLex) to predict the emotions associated with Kannada-English code-switch text that no one has addressed till now. We applied the One-vs-Rest approach to generate the scores for lexicon and also to predict the emotions from the code-switch text. The accuracy of the proposed model NBLex (87.9%) is better than naïve bayes (NB) (85.8%) and bidirectional long short-term memory neural network (BiLSTM) (84.7%) and for true positive rate (TPR), the NBLex (50.6%) is better than NB (37.0%) and BiLSTM (42.2%). From our approach, it is observed that a simple additive model (lexicon approach) can also be an alternative model to predict the emotions in code-switch text.