Machine reading comprehension with multi-hop reasoning always suffers from reasoning path breaking due to the lack of world knowledge, which always results in wrong answer detection. In this paper, we analyze what knowledge the previous work lacks, e.g., dependency relations and commonsense. Based on our analysis, we propose a Multi-dimensional Knowledge enhanced Graph Network, named MKGN, which exploits specific knowledge to repair the knowledge gap in reasoning process. Specifically, our approach incorporates not only entities and dependency relations through various graph neural networks, but also commonsense knowledge by a bidirectional attention mechanism, which aims to enhance representations of both question and contexts. Besides, to make the most of multi-dimensional knowledge, we investigate two kinds of fusion architectures, i.e., in the sequential and parallel manner. Experimental results on HotpotQA dataset demonstrate the effectiveness of our approach and verify that using multi-dimensional knowledge, especially dependency relations and commonsense, can indeed improve the reasoning process and contribute to correct answer detection.
Ying ZHANG
Beijing Jiaotong University
Fandong MENG
Tecent Inc
Jinchao ZHANG
Tecent Inc
Yufeng CHEN
Beijing Jiaotong University
Jinan XU
Beijing Jiaotong University
Jie ZHOU
Tecent Inc
The copyright of the original papers published on this site belongs to IEICE. Unauthorized use of the original or translated papers is prohibited. See IEICE Provisions on Copyright for details.
Copy
Ying ZHANG, Fandong MENG, Jinchao ZHANG, Yufeng CHEN, Jinan XU, Jie ZHOU, "MKGN: A Multi-Dimensional Knowledge Enhanced Graph Network for Multi-Hop Question and Answering" in IEICE TRANSACTIONS on Information,
vol. E105-D, no. 4, pp. 807-819, April 2022, doi: 10.1587/transinf.2021EDP7154.
Abstract: Machine reading comprehension with multi-hop reasoning always suffers from reasoning path breaking due to the lack of world knowledge, which always results in wrong answer detection. In this paper, we analyze what knowledge the previous work lacks, e.g., dependency relations and commonsense. Based on our analysis, we propose a Multi-dimensional Knowledge enhanced Graph Network, named MKGN, which exploits specific knowledge to repair the knowledge gap in reasoning process. Specifically, our approach incorporates not only entities and dependency relations through various graph neural networks, but also commonsense knowledge by a bidirectional attention mechanism, which aims to enhance representations of both question and contexts. Besides, to make the most of multi-dimensional knowledge, we investigate two kinds of fusion architectures, i.e., in the sequential and parallel manner. Experimental results on HotpotQA dataset demonstrate the effectiveness of our approach and verify that using multi-dimensional knowledge, especially dependency relations and commonsense, can indeed improve the reasoning process and contribute to correct answer detection.
URL: https://global.ieice.org/en_transactions/information/10.1587/transinf.2021EDP7154/_p
Copy
@ARTICLE{e105-d_4_807,
author={Ying ZHANG, Fandong MENG, Jinchao ZHANG, Yufeng CHEN, Jinan XU, Jie ZHOU, },
journal={IEICE TRANSACTIONS on Information},
title={MKGN: A Multi-Dimensional Knowledge Enhanced Graph Network for Multi-Hop Question and Answering},
year={2022},
volume={E105-D},
number={4},
pages={807-819},
abstract={Machine reading comprehension with multi-hop reasoning always suffers from reasoning path breaking due to the lack of world knowledge, which always results in wrong answer detection. In this paper, we analyze what knowledge the previous work lacks, e.g., dependency relations and commonsense. Based on our analysis, we propose a Multi-dimensional Knowledge enhanced Graph Network, named MKGN, which exploits specific knowledge to repair the knowledge gap in reasoning process. Specifically, our approach incorporates not only entities and dependency relations through various graph neural networks, but also commonsense knowledge by a bidirectional attention mechanism, which aims to enhance representations of both question and contexts. Besides, to make the most of multi-dimensional knowledge, we investigate two kinds of fusion architectures, i.e., in the sequential and parallel manner. Experimental results on HotpotQA dataset demonstrate the effectiveness of our approach and verify that using multi-dimensional knowledge, especially dependency relations and commonsense, can indeed improve the reasoning process and contribute to correct answer detection.},
keywords={},
doi={10.1587/transinf.2021EDP7154},
ISSN={1745-1361},
month={April},}
Copy
TY - JOUR
TI - MKGN: A Multi-Dimensional Knowledge Enhanced Graph Network for Multi-Hop Question and Answering
T2 - IEICE TRANSACTIONS on Information
SP - 807
EP - 819
AU - Ying ZHANG
AU - Fandong MENG
AU - Jinchao ZHANG
AU - Yufeng CHEN
AU - Jinan XU
AU - Jie ZHOU
PY - 2022
DO - 10.1587/transinf.2021EDP7154
JO - IEICE TRANSACTIONS on Information
SN - 1745-1361
VL - E105-D
IS - 4
JA - IEICE TRANSACTIONS on Information
Y1 - April 2022
AB - Machine reading comprehension with multi-hop reasoning always suffers from reasoning path breaking due to the lack of world knowledge, which always results in wrong answer detection. In this paper, we analyze what knowledge the previous work lacks, e.g., dependency relations and commonsense. Based on our analysis, we propose a Multi-dimensional Knowledge enhanced Graph Network, named MKGN, which exploits specific knowledge to repair the knowledge gap in reasoning process. Specifically, our approach incorporates not only entities and dependency relations through various graph neural networks, but also commonsense knowledge by a bidirectional attention mechanism, which aims to enhance representations of both question and contexts. Besides, to make the most of multi-dimensional knowledge, we investigate two kinds of fusion architectures, i.e., in the sequential and parallel manner. Experimental results on HotpotQA dataset demonstrate the effectiveness of our approach and verify that using multi-dimensional knowledge, especially dependency relations and commonsense, can indeed improve the reasoning process and contribute to correct answer detection.
ER -