کنفرانس مهندسی برق ایران

صفحه اصلی / سی و یکمین کنفرانس بین المللی مهندسی برق

Single- and Multi-Hop BERT Question Classifier for Open-Domain Question Answering (SiMQC)

نویسندگان :

Faeze Zakaryapour Sayyad¹ Mahdi Bohlouli²

1- Institute for Advanced Studies in Basic Sciences 2- Institute for Advanced Studies in Basic Sciences

کلمات کلیدی :

Text-based Question Answering،Information Retrieval،Deep Learning،Natural Language Processing

چکیده :

Multi-hop Question Answering has received particular attention in research and practice in recent years, especially in the context of conversational systems and answering complex questions. Various architectures have been proposed to answer these complex multi-hop questions. However, in real-world scenarios, a conversational system should be able to respond to both simple (single-hop), and complex (multi-hop) questions and particular attention to one shouldn't result in not being able to respond to another. Furthermore, some multi-hop questions can be easily processed by splitting them into multiple single-hop questions. In this work, we propose an efficient BERT question classifier that supports retrievers in the question answering systems to process both single- and multi-hop questions. Another contribution of this work is the release of a mixed dataset that consists of both single- and multi-hop questions. We show that utilizing our classifier inside the QA system can improve the accuracy of these systems and enables them to answer both kinds of questions considering their complexity.