Semantic Improvement of Question-Answering System with Multi-Layer LSTM

Hanif Bhuiyan, Md. Abdul Karim, Faria Benta Karim, Jinat Ara

Abstract: Question answering (QA) is a popular method to extract information from a large pool of data. However, effective feature extraction and semantic understanding, and interaction between question and answer pairs are the main challenges for the QA model. Mitigating these issues, here, we propose a Long-Short-Term-Memory (LSTM) model using the recurrent neural network. This model aims to extract the effective interaction between questions-answers pairs using the Softmax and Max-pooling function. The experiment performs on a publicly available Wiki-QA dataset to identify the effectiveness of the proposed model. The evaluation performs by comparing the result with other existing models. The comparison shows that the proposed approach is significant and competitive for QA and achieves 83% training and 81% testing accuracy on the Wiki-QA dataset.