Sentiment analysis on bangla and romanized bangla text using deep recurrent models

摘要

Sentiment Analysis (SA) is an opinion mining study analyzing people's opinions, sentiments, evaluations and appraisals towards societal entities such as products, services, individuals, organizations, events, etc. Of late, most of the research works on SA in natural language processing (NLP) are focused on English language. However, it is noted that Bangla does not have a proper dataset that is both large and standard. As a result, recent research works with Bangla in SA have fallen short to produce results that can be both comparable to works done by others in other languages and reusable for further prospective research. In this work, a substantial textual dataset of both Bangla and Romanized Bangla texts have been provided which is first of this kind and post-processed, multiple validated, and ready for SAimplementation and experiments. Further, this dataset have been tested in Deep Recurrent model, specifically, Long Short Term Memory (LSTM), using two types of loss functions - binary cross-entropy and categorical cross-entropy, and also some experimental pre-training were conducted by using data from one validation to pre-train the other and vice versa. Lastly, the results along with analysis are presented in this research.