Implementation of a Bangla chatbot

Abstract

We propose a fully data driven retrieval based closed domain chatbot which can converse in Bengali with the user in a chat interface based on its knowledge base and through learning from interactions with the user. Bengali is the state language of Bangladesh and fourth most popular language in the world. Although the first chatbot Eliza was invented in 1964, the author is not aware of any other work where Bengali chatbot has been implemented. So it is the demand of time to build a chatbot in Bengali. This paper proposes the first Bengali chatbot named Golpo based onalanguage-independent natural language processing library with a learning mechanism. The experiment shows that our chatbot is able to give responses to the user inreal time.Atfirst, it matches the input with the existing queries in the database, then it calculates a confidence score for each matching sentences with the input, and finally, it selects the one with highest confidence score as the response to the input.We hypo the size that this implementation will help to build different goal oriented chatbot in Bengali, for example, customer care representatives, FAQ (Frequently Asked Questions) chatbot, online sales agent etc. Besides, one of the main contributions of this work is that Golpo will be able to provide a Bengali Corpus for research purpose in future. Experimental results show that Golpo outperforms the state-of-the-art model named Cleverbot (Saenz, 2010), and achieves reasonably good results when compared to Neural Conversational Model (NCM) (Vinyals, 2015), a generative based model. Based on the evaluation of users, we can say Golpo can produce syntactically correct and natural responses in Bengali.