Question-Answer System on Episodic Data Using Recurrent Neural Networks (RNN)

Data comprehension is one of the key applications of question-answer systems. This involves a closed-domain answering system where a system can answer questions based on the given data. Previously people have used methods such as part of speech tagging and named entity recognition for such problems but those methods have struggled to produce accurate results since they have no information retention mechanisms. Deep learning and specifically recurrent neural networks based methods such as long short-term memory have been shown to be successful in creating accurate answering systems. This paper focuses on episodic memory where certain facts are aggregated in the form of a story and a question is asked related to a certain object in the story and a single fact present is given as answer. The paper compares the performance of these algorithms on benchmark dataset and provides guidelines on parameter tuning to obtain maximum accuracy. High accuracy (80% and above) was achieved on three tasks out of four.