ACA: Attention-Based Context-Aware Answer Selection System

Abstract

Listen

Translate article

The main goal of question answering system is to develop chatbots capable of answering the questions irrespective of the domain. The system has to provide appropriate answers according to the user queries. The challenge of a question answering system lies in analyzing the question to retrieve the accurate answers from a large amount of data present. The main aim of this paper is to propose a question answering system that analyzes the questions properly and answers according to the type of the question in a precise way. The model uses similarity measures to find the candidate set which are the most relevant answers from the content given. The generated candidate set is then analyzed further using the bidirectional long short-term memory (BiLSTM) to retrieve the appropriate answer. Based on the type of question, we retrieve the exact word/phrase as a response to the user. This is done using an attention mechanism which _nds the exact answer to the query. Results of models without memory and with an attention-based BiLSTM are being compared. Using the attention-based BiLSTM, we found that the retrieved answer had increased accuracy rate by 17% when compared with without-attention models like convolution neural networks (CNN) and recurrent neural networks (RNN).