Question Classification on Question-Answer System using Bidirectional-LSTM

Abstract

The first step in the question-answer system is question analysis. This step was carried out with various processes, one of them was question classification. The question classification process can determine the accuracy of the answers generated by the system. Some approaches that have been used are Support Vector Machine (SVM), pattern matching, naive bayes classification and Latent Dirichlet Allocation (LDA). Research on question classification using that approach has been widely carried out but still only work on certain sentence patterns. These problems can be solved using deep learning method, one of them is Bidirectional Long Short Term Memory (Bi-LSTM). Bi-LSTM does not depend on certain sentence patterns. Bi-LSTM has good accuracy in text classification. This study uses Bi-LSTM in the question classification process. Questions divided into three classes, that is greeting, daily conversation, and meetings. The classification results show the accuracy of 0.909 with loss of 0.316. Bi-LSTM has higher accuracy compared to basic LSTM and Recurrent Neural Network (RNN). Based on the results, Bi-LSTM can be used as question classification method for Question Answer System (QA System).