Deep learning based Bengali question answering system using semantic textual similarity

Recently, Question answering system is a major research area in language processing. Bengali isone of the most popular spoken languages in India. Still, it has faced difficulties in natural language processing.Among the semantic based systems, word mapping and keyword based approaches achieved the best results and got better attention on the user side. These systems are already implemented in various languages but not much in Indian language like Bengali. This work presents an efficient question answering system for retrieving Bengali language text. This system includes word embedding clustering and deep level feature representation for providing better grammatical similarities for retrieving the Bengali textual contents relevant to user queries. The pre-trained word embedding module is created by the help of a deep belief network. The modified density peak algorithm is employed to perform word embedding clustering.The presented work has been tested on a dataset from the Bengali corpus developed by TDIL and synthetic Bengali translated datasets accessible in English called SQuAD 2.0. This question answering system is implemented in python with NLTK tool kit and got good performance while retrieving the Bengali textual data.