Ensemble Learning Approach to Chatbot Design Based on Paraphrase Detection

Abstract

In this paper, we present a design for an ensemble chatbot based on paraphrase detection. Our proposed chatbot is intended to assist companies in reducing the need for costly call center operations by providing a 24-hour service to customers seeking information about products or services. Our algorithm is designed to work effectively on small data sets, such as an existing FAQ, and does not require a large number of instances. We evaluated the performance of our chatbot using publicly available data from the websites of major telecommunication companies and found that the ensemble model improved success rates by 6% compared to the single best model, with a top 3 accuracy of 84.54% and a top 1 accuracy of 70.10%.