Emotion Detection from text in Hindi

Lately, a large amount of text data is generated every day. We require a straightforward method to analyze and summarize this data to draw relevant insights. In order to assess a person’s interest in any topic, emotion detection (ED) is essential. Remarkable research has been done in this area using resource-intensive languages such as English, Spanish and others. However, due to a lack of annotated datasets, emotion identification work has fallen behind in low-resource languages like Hindi. In this study, we compare two models - a Naïve Bayes model and an SVM classifier, both trained on a labelled Hindi dataset in Devanagari script, classifying four emotions - happy, angry, sad and neutral. With a macro-average F1 score of 0.54—8% higher than the SVM classifier—experimental data demonstrates that the Naïve Bayes model outperforms the latter.