Context-based Sentiment analysis of Indian Marathi Text using Deep Learning

Abstract

In Digital India, the Internet plays a crucial role in communication. The English language is widely used for such a process. The Internet has no language barrier. India is a multi-lingual country with boundless linguistic and social diversities. The most trending pattern observed in India is people intend to post their views, thoughts, feedback, and comments in their mother tongue over social media and blogs. Views posted by people is important for organization belonging to any category small, medium and large enterprises to improve their product or service. This data is hastily accumulated every day which should be necessary to identify and process. In terms of processing little work has been done for Indian languages where traditional approaches were used which are far away from the context of the text. In this research to perform sentiment analysis supervised algorithms that is Multinomial Naïve Bayes is implemented on the Marathi dataset. Along with this deep learning, Natural Language Processing approach Bidirectional Encoder Representations from Transformers (BERT) is utilized and fine-tuned for the specific work to evaluate more accuracy and State-of-the-Art results.