A Classification Intelligent Question Answering Model for Retrieval-Based Chatbots

Abstract Intelligent question answering (QA) models or chatbots automatically provide appropriate responses to questions posed by users. In terms of generating continuous responses, they are divided into generative and retrieval-based approaches. For retrieval-based QA models, the key issue is how to reduce the search space. This research focuses on a retrieval-based approach and proposes a classification intelligent question answering (CIQA) model. The CIQA model contains two stages, namely a question classification stage and an answer prediction stage. The first stage consists of building a classification ensemble based on a training set. The second stage uses the first stage classification ensemble to determine the appropriate categories for a test set and selects an appropriate deep learning QA model based on a chosen category. A new benchmark dataset for chatbot, SQuAD (Stanford question answering dataset) 2.0, is used to evaluate performance. Based on the outcome of our experiments, the proposed CIQA model outperforms the baseline model and demonstrates the feasibility of the proposed approach. JEL classification numbers: M15, O35. Keywords: Question answering, Ensemble learning, Deep learning, Retrieval-based QA models.