Bangla E-mail Body to Subject generation using sequence to sequence RNNs

Recent times have been difficult for deep learning and natural language processing in terms of topic generation. The production of a subject reduces the length of a lengthy email body comment. Our goal is to design a Bengali subject line generator that is both efficient and effective, and that can provide a subject line that is both understandable and illuminating based on the Bengali text of an email. In preparation for this, we gathered educational, commercial, and other email bodies. These writings will be used as the basis for themes generated by our algorithm. While bi-directional RNNs and LSTMs are used in the encoding layer, an attention model is utilized in the decoding layer. Our sequence-to-sequence paradigm develops topics. Text pre-processing, missing word counts, vocabulary counting, identifying new words, word embedding, and other tasks presented challenges during the construction of this model. This model's goals were to come up with a subject and cut down on training time lost. We were able to cut our study trial training loss down to 0.001 by generating a clear and coherent subject line from the text of an email.