Bengali Text Classification Using Bi-LSTM and LSTM: Differentiating Between Saint and Common Forms of Bengali Text

In the realm of natural language processing, Bengali text classification is an impactful factor that significantly enhances the applications of the Bengali natural language processing research domain. Although there are several research efforts on Bengali text classification in various aspects, notably there is a scarcity of Bengali text classification in terms of focusing on its two linguistic forms: saint (Sadhu) and common (Cholito), with a limited dataset size and a lack of advanced technique usages. Thus, this proposed work aims to classify Bengali texts into saint and common forms using two deep learning models, utilizing 7350 data accumulated from diverse sources such as Bengali literature books, magazines, newspapers, articles, blogs, and social media posts. After the accumulation of data, several preprocessing techniques have been applied. Two deep learning algorithms were utilized, including Bi-LSTM and LSTM, on the preprocessed data, and the performances of these algorithms were measured across several matrices. Specifically, Bi-LSTM surpasses LSTM in terms of accuracy, recall, precision, and f1-score. Bi-LSTM attained 92.52% accuracy, whereas LSTM attained 91.43% accuracy, and compared to LSTM, Bi-LSTM has a lower misclassification rate and it converges fastly. Ultimately, this research opens up the scope for research in Bengali natural language processing in terms of text classification with more advanced techniques.