Multi-hop reading comprehension based on reinforcement learning

Hongkai Wu

doi:10.1117/12.2635359

6 May 2022 Multi-hop reading comprehension based on reinforcement learning

Hongkai Wu

Proceedings Volume 12256, International Conference on Electronic Information Engineering, Big Data, and Computer Technology (EIBDCT 2022); 122561Q (2022) https://doi.org/10.1117/12.2635359
Event: 2022 International Conference on Electronic Information Engineering, Big Data and Computer Technology, 2022, Sanya, China

Abstract

Multi-hop machine reading comprehension (MRC) across multiple documents poses new challenges over single-hop MRC, because it requires reasoning several times to answer the given questions and being able to show the reasoning path to support its answer as explanation at the same time. In this paper, we propose a new model based on reinforcement learning for multi-hop MRC. Our model mainly consists of two parts: (i) a novel agent that decomposes the multi-hop question into several sub-questions implicitly, (ii) an incorporator incorporating external knowledge to enhance the answering ability of the sub-question answerer. Experimental results show that our model performs better compared to the path-based models and has higher interpretability compared to the graph-based models. In the last section, we discuss the development prospects of our model in the future.

Citation Download Citation

Hongkai Wu "Multi-hop reading comprehension based on reinforcement learning", Proc. SPIE 12256, International Conference on Electronic Information Engineering, Big Data, and Computer Technology (EIBDCT 2022), 122561Q (6 May 2022); https://doi.org/10.1117/12.2635359

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available