Deep Learning based Analysis on Code-Mixed Tamil Text for Sentiment Classification with Pre-Trained ULMFiT

Abstract

Sentiment Analysis is the process of getting people’s ideas on what they think or feel about a particular issue or product or a person and classifying the information expressed about an issue in a positive, negative or a neutral manner. This extracted information can be found very much useful in determining popularity of a person or a product and they are very much useful in ecommerce in suggesting products to buy and in social media like YouTube in suggesting videos to view. Mostly people express their views and share their thoughts in their mother Tongue along with usage of mixed language words is common nowadays. Hence Sentiment analysis of codemixed language plays a major role. There is only little work available in Tamil mixed Sentiment analysis. In order to know the opinion of people who speak Tamil in an effective manner an effective algorithm is needed. Since there are many algorithms available in machine learning and deep learning, this work aims to find sentiment in code mixed words. Deep learning based Bi-LSTM model with ULMFiT Embedding gives more promising results for code-mixed language than other existing algorithms.