Fusion-based Model for Detection and Classification of Human Sentiments from Bengali Text

Emotions have a big impact on human connection in daily life. Emotions can be expressed through spoken language, written text, and facial expressions. Emotional expression on blogs and social media has become increasingly common in recent years. On any political or international issue, people write down their opinions and sentiments. Everything is there for us to gather and analyze human feelings from the text in these societal occurrences. Although extensive research has been done on emotion recognition, most research has been thoroughly examined in the English language. The field of the Bangla language is still in its early stages. The main goals of the research are to develop an emoji prediction system and a Bangla text-based emotion identification system. To address these problems, we have utilized several neural networks and machine learning algorithms. In this regard, a new Bangla corpus dataset is generated containing six classes: happiness, sadness, fear, anger, disgust and guilt. Moreover, We have also developed a hybrid model by using model fusion with CNN, BiLSTM, and CNN + BiLSTM. This model outperforms some works while showing an accuracy of 84%.