A Comparative study for sentiment analysis of raw and translated text

Abstract

Sentiment analysis is a type of opinion mining study analyzing people's thoughts, feelings, and assessments evaluations of society things such as products, services, and institutions individuals, organizations, events and etc. This paper focuses on an effective machine learning model, Recurrent Neural Network (RNN), for sentiment analysis of both Bengali and English text. For sentiment analysis of Bengali text, we construct a novel dataset accumulated from different open source sites and social media. The dataset comprises over 17,000 texts, which are labelled as neutral, positive, and negative. Our study provides a rigorous comparative analysis on the performance of RNN in terms of sentiment analysis of both raw text and corresponding translated text (for both English and Bengali language). In summary, we achieve 82.9% and 84.5% accuracies for sentiment analysis of Bengali and English text respectively using RNN.