The chatbot is designed to respond to users with automated responses with respect to the content provided by the user. But the question arises when the user provides a free text which is a synonym of required content or part of the content for the chatbot to understand. In this case most of the Chatbot models using huge libraries which has a large number of samples and require more computational time and storage. Keyword detection methods with a huge amount of data are suitable for most applications but chatbots were designed for specific tasks, for example, ordering food, customer support for the specific application, etc., so these types of chatbots don’t need huge training data. In this paper, we conducted a performance evaluation of different sets and sizes of samples based on certain keywords specifically used for the closed domain chatbot. In this research, we used Movielens 20M dataset which provides tag assignments between movies and unique tags. We used Deep Learning methods in this keyword extraction model.