Exploring Word2vec Embedding for Sentiment Analysis of Bangla Raw and Romanized Text

In this era of rapid technological advancement, every individual’s daily life has become a routine of sharing their perspectives, opinions, emotions, and experiences through social networking sites and platforms on the Internet. These viewpoints can be used to establish strategies that can improve efficiency in a variety of areas such as business, politics, research, and analysis. Sentiment analysis (SA) is used in natural language processing (NLP) to automatically monitor, analyze, and categorize individuals’ thoughts and opinions in order to acquire a sense of general sentiment. To date, a significant amount of research has been conducted on SA of English language with remarkable successes. Unfortunately, there has been relatively insufficient research in the field of SA with the Bangla language. Despite the fact that romanized Bangla has gained in popularity among Bangla speakers as a result of the recent surge in social networks, there is even less research on romanized Bangla text. Therefore, this research has concentrated on the analysis of sentiment for both Bangla raw and romanized texts. In this study, a corpus of romanized Bangla texts has been constructed from a raw Bangla sentiment corpus. Furthermore, both of these corpora have been tested for SA using the deep recurrent neural network with continuous bag of words and skip-gram word2vec word embeddings for both binary and multi-label classifications. Finally, this study concludes with the comparative results and analysis of SA of both forms of Bangla texts, where SA of romanized Bangla texts outperforms its raw form.