Identifying Sentiment and Recognizing Emotion from Social Media Data in Bangla Language
Rifat Rahman, Sheikh Abir Hasan, Fardous Ahmed Rubel · 2022
Sentiment identification and emotion recognition play a significant role in many areas of natural language understanding because of their extensive usability in different applications like chatbots, text summarization, social media monitoring, customer support management, etc. In comparison with other languages, the studies done in the Bangla language are very few due to the scarcity of data and proper preprocessing tools. In this work, we present a benchmark study on different models to classify both emotion and sentiment from Bangla textual data. We collect our corpus from three online social media platforms (e.g., Youtube, Twitter, and Reddit) and perform a comparative study among several traditional machine learning and neural network-based approaches. We also perform comparative analysis among various informative feature extraction techniques. We find that the sequence model-based approach (Bidirectional GRU) with the “word embedding” feature extraction technique (word2vec) outperforms other approaches for emotion recognition and sentiment analysis.