BanglaSenti: A Dataset of Bangla Words for Sentiment Analysis

Being the fifth most spoken language in the world, use of Bangla or Bengali language has spread its breadth into the world of social media. Huge volume of user-generated Bangla data are produced daily in various social media such as Facebook, Twitter, YouTube; online news portals and various websites. Hence, the importance of understanding the emotion and sentiment of these types of data has gain attraction to the researchers' recently. Bangla Natural Language Processing (BNLP) has emerged as a novel research domain because of these multidisciplinary scopes. In this paper, we have presented “BanglaSenti”, which is a lexicon based corpus or dataset generated solely to identify the sentiment analysis from textual data. This dataset contains 61582 Bangla words with positive, negative and neutral words. These polarities are very significant to understand the overall polarity of the sentences. Not only the corpus, but also a model simulation has been conducted in this paper to understand the usability of this dataset. The dataset is formalized as English SentiWordNet dataset so that researchers' may utilize it with the same format of codes. Though the dataset is developed for sentiment analysis, it could be utilized for emotion detection, opinion/review mining and such applications.