A Metaphorical Study Of Variants Of Recurrent Neural Network Models For A Context Learning Chatbot

Abstract

In this paper, different types of recurrent neural networks such as GRU, bi-directional LSTM and a single forward pass LSTM available for question answering systems are explored, so that it can be used in a high level API to be able to create a chatbot interface for a web application. The networks are compared on the Facebook bAbi dataset to test the question answering functionality on 20 different types tasks available in the dataset. The results from existing models were compared and it is observed that the bi-directional LSTM and the GRU performed better in few tasks, however the single forward pass LSTM performed the best for majority of tasks in the dataset.