融入三维语义特征的常识推理问答方法

doi:10.11772/j.issn.1001-9081.2023010063

《计算机应用》唯一官方网站 ›› 2024, Vol. 44 ›› Issue (1): 138-144.DOI: 10.11772/j.issn.1001-9081.2023010063

• 人工智能 • 上一篇

融入三维语义特征的常识推理问答方法

王红斌¹^,²^,³, 房晓¹^,²^,³, 江虹¹^,²^,³()

^1.昆明理工大学信息工程与自动化学院, 昆明 650500
^2.云南省人工智能重点实验室(昆明理工大学), 昆明 650500
^3.云南省计算机技术应用重点实验室(昆明理工大学), 昆明 650500

收稿日期:2023-01-30 修回日期:2023-05-10 接受日期:2023-05-12 发布日期:2023-06-06 出版日期:2024-01-10
通讯作者: 江虹
作者简介:王红斌（1983—），男，云南曲靖人，教授，博士，CCF会员，主要研究方向：自然语言处理、信息检索、机器学习；
房晓（1997—），女，山东烟台人，硕士研究生，CCF会员，主要研究方向：智能信息处理；
第一联系人：江虹（1965—），男，云南昆明人，讲师，硕士，主要研究方向：智能信息处理。
基金资助:
国家自然科学基金资助项目(61966020);云南省基础研究计划项目(202201AT070157)

Commonsense reasoning and question answering method with three-dimensional semantic features

Hongbin WANG¹^,²^,³, Xiao FANG¹^,²^,³, Hong JIANG¹^,²^,³()

^1.Faculty of Information Engineering and Automation，Kunming University of Science and Technology，Kunming Yunnan 650500，China
^2.Yunnan Key Laboratory of Artificial Intelligence （Kunming University of Science and Technology），Kunming Yunnan 650500，China
^3.Yunnan Key Laboratory of Computer Technology Application （Kunming University of Science and Technology），Kunming Yunnan 650500，China

Received:2023-01-30 Revised:2023-05-10 Accepted:2023-05-12 Online:2023-06-06 Published:2024-01-10
Contact: Hong JIANG
About author:WANG Hongbin， born in 1983， Ph. D.， professor. His research interests include natural language processing， information retrieval， machine learning.
FANG Xiao， born in 1997， M. S. candidate. Her research interests include intelligent information processing.
Supported by:
National Natural Science Foundation of China(61966020);Basic Research Program of Yunnan Province(202201AT070157)

摘要/Abstract

摘要：

现有使用预训练语言模型和知识图谱的常识问答方法主要集中于构建知识图谱子图及跨模态信息结合的研究，忽略了知识图谱自身丰富的语义特征，且缺少对不同问答任务的知识图谱子图节点相关性的动态调整，导致预测准确率低。为解决以上问题，提出一种融入三维语义特征的常识推理问答方法。首先提出知识图谱节点的关系层级、实体层级、三元组层级三维语义特征量化指标；其次，通过注意力机制动态计算关系层级、实体层级、三元组层级三种维度的语义特征对不同实体节点间的重要性；最后，通过图神经网络进行多层聚合迭代嵌入三维语义特征，获得更多的外推知识表示，更新知识图谱子图节点表示，提升答案预测精度。与QA-GNN常识问答推理方法相比，所提方法在CommonsenseQA数据集上的验证集和测试集的准确率分别提高了1.70个百分点和0.74个百分点，在OpenBookQA数据集上使用AristoRoBERTa数据处理方法的准确率提高了1.13个百分点。实验结果表明，所提出的融入三维语义特征的常识推理问答方法能够有效提高常识问答任务准确率。

关键词: 常识问答, 知识图谱, 图神经网络, 语义特征, 注意力机制

Abstract:

The existing commonsense question answering methods based on pre-trained language model and knowledge graph mainly focus on the construction of subgraphs of knowledge graph and combination of cross-modal information， ignoring the rich semantic features of knowledge graph itself， and lack dynamic adjustment of correlation among knowledge graph subgraph nodes to different question answering tasks， thus they do not achieve satisfactory prediction accuracies. To solve these above problems， a commonsense reasoning and question answering method integrating three-dimensional semantic features was proposed. Firstly， the quantitative indicators of three-dimensional semantic features at relation level， entity level and triple level for knowledge graph nodes were proposed. Secondly， the importance of semantic features of three dimensions of relation level， entity level and triple level to different entity nodes was dynamically calculated through attention mechanism. Finally， multi-layer aggregation iterative embedding of three-dimensional semantic features was carried out through graph neural network， to obtain more extrapolated knowledge representation， update subgraph node representation of knowledge graph， and improve the accuracy of answer prediction. Compared with QA-GNN commonsense question answering and reasoning method， the accuracy of proposed method in verification set and test set of CommonsenseQA dataset was improved by 1.70 percentage points and 0.74 percentage points， and the accuracy of the proposed method by AristoRoBERTa data processing method on OpenBookQA dataset was improved by 1.13 percentage points. Experimental results show that the proposed commonsense reasoning and question answering method integrating three-dimensional semantic features can effectively improve the accuracy of commonsense question answering tasks.

Key words: commonsense question answering, knowledge graph, graph neural network, semantic feature, attention mechanism

中图分类号:

TP391.1

王红斌, 房晓, 江虹. 融入三维语义特征的常识推理问答方法[J]. 计算机应用, 2024, 44(1): 138-144.

Hongbin WANG, Xiao FANG, Hong JIANG. Commonsense reasoning and question answering method with three-dimensional semantic features[J]. Journal of Computer Applications, 2024, 44(1): 138-144.

图/表 8

图1 CommonsenseQA数据集示例

Fig. 1 Example in CommonsenseQA dataset

图2 融入三维语义特征的常识推理问答方法的整体框架

Fig. 2 Overall framework of commonsense reasoning and question answering method with three-dimensional semantic features

图3 三维语义特征

Fig. 3 Three-dimensional semantic features

表1 实验数据集统计

Tab. 1 Statistics for experimental datasets

数据集	样本数
数据集	训练集	验证集	测试集
CommonsenseQA	18 241	2 442	2 381
OpenBookQA	4 957	500	500

表2 不同方法在CommonsenseQA数据集的准确率对比 ( %)

Tab. 2 Accuracy comparison among different methods on CommonsenseQA dataset

方法	Dev-Acc	Test-Acc
RoBERTa-large （w/o KG）	73.07	68.69
R-GCN	72.69	68.41
GconAttn	72.61	68.59
KagNet	73.47	69.01
RN	74.57	69.08
MHGRN	74.45	71.11
QA-GNN	76.54	73.41
DRGN	78.20	74.00
本文方法	78.24	74.15

表3 不同方法在OpenBookQA数据集下的准确率对比 ( %)

Tab. 3 Accuracy comparison among different methods on OpenBookQA dataset

方法	RoBERTa-Large	AristoRoBERTa
RoBERTa-large （w/o KG）	64.80	78.40
R-GCN	62.45	74.60
GconAttn	64.75	71.80
RN	65.20	75.35
MHGRN	66.85	80.60
QA-GNN	70.58	82.77
DRGN	70.10	81.80
本文方法	70.63	83.90

表4 CommonsenseQA数据集的三维语义特征消融实验结果 ( %)

Tab. 4 Ablation experiment results of three-dimensional semantic features on CommonsenseQA dataset

方法	Dev-Acc
w/o关系级语义特征	77.84
w/o实体级语义特征	77.73
w/o三元组级语义特征	77.74
w/o关系级&实体级语义特征	77.44
w/o关系级&三元组级语义特征	77.24
w/o实体级&三元组级语义特征	76.94
本文方法	78.24

表5 不同GNN层数的准确率对比 ( %)

Tab. 5 Comparison of accuracy with different GNN layers

GNN层数	Dev-Acc	GNN层数	Dev-Acc
3	75.93	6	77.68
4	77.18	7	77.20
5	78.24

参考文献 30

1	RAJPURKAR P， ZHANG J， LOPYREV K， et al. SQuAD： 100000+ questions for machine comprehension of text ［C］// Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing. Stroudsburg， PA： ACL， 2016： 2383-2392. 10.18653/v1/d16-1264
2	MIN S， ZHONG V， ZETTLEMOYER L， et al. Multi-hop reading comprehension through question decomposition and rescoring ［C］// Proceedings of the 57th Conference of the Association for Computational Linguistics. Stroudsburg， PA： ACL， 2019： 6097-6109. 10.18653/v1/p19-1613
3	YANG Z， QI P， ZHANG S， et al. HotpotQA： A dataset for diverse， explainable multi-hop question answering ［C］// Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing. Stroudsburg， PA： ACL， 2018： 2369-2380. 10.18653/v1/d18-1259
4	TALMOR A， HERZIG J， LOURIE N， et al. CommonsenseQA： A question answering challenge targeting commonsense knowledge ［C］// Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics： Human Language Technologies， Volume 1（Long and Short Papers）. Stroudsburg， PA： ACL， 2019： 4149-4158.
5	YASUNAGA M， REN H， BOSSELUT A， et al. QA-GNN： Reasoning with language models and knowledge graphs for question answering ［C］// Proceedings of the 2021 Conference on North American Chapter of the Association for Computational Linguistics： Human Language Technologies. Stroudsburg， PA： ACL， 2021： 535-546. 10.18653/v1/2021.naacl-main.45
6	SUN Y， SHI Q， QI L， et al. JointLK： Joint reasoning with language models and knowledge graphs for commonsense question answering ［C］// Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics： Human Language Technologies. Stroudsburg， PA： ACL， 2022： 5049-5060. 10.18653/v1/2022.naacl-main.372
7	ZHENG C， KORDJAMSHIDI P. Dynamic relevance graph network for knowledge-aware question answering ［C］// Proceedings of the 29th International Conference on Computational Linguistics. ［S.l.］： International Committee on Computational Linguistics， 2022： 1357-1366.
8	白铂，刘玉婷，马驰骋，等.图神经网络［J］.中国科学：数学， 2020， 50（3）： 367-384. 10.1360/n012019-00133
	BAI B， LIU Y T， MA C C， et al. Graph neural network ［J］. SCIENTIA SINICA Mathematica， 2020， 50（3）： 367-384. 10.1360/n012019-00133
9	MIHAYLOV T， CLARK P， KHOT T， et al. Can a suit of armor conduct electricity？ a new dataset for open book question answering ［C］// Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing. Stroudsburg， PA： ACL， 2018： 2381-2391. 10.18653/v1/d18-1260
10	DEVLIN J， CHANG M W， LEE K， et al. BERT： Pre-training of deep bidirectional transformers for language understanding ［C］// Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics： Human Language Technologies， Volume 1（Long and Short Papers）. Stroudsburg， PA： ACL， 2019： 4171-4186. 10.18653/v1/n18-2
11	FAGHIHI H R， KORDJAMSHIDI P. Time-stamped language model： teaching language models to understand the flow of events ［C］// Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics： Human Language Technologies. Stroudsburg， PA： ACL， 2021： 4560-4570. 10.18653/v1/2021.naacl-main.362
12	LIN B Y， CHEN X， CHEN J， et al. KagNet： Knowledge-aware graph networks for commonsense reasoning ［C］// Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing. Stroudsburg， PA： ACL， 2019： 2829-2839. 10.18653/v1/d19-1282
13	KIPF T N， WELLING M. Semi-supervised classification with graph convolutional networks ［EB/OL］. （2017-02-22）［2022-06-20］. . 10.48550/arXiv.1609.02907
14	FENG Y， CHEN X， LIN B Y， et al. Scalable multi-hop relational reasoning for knowledge-aware question answering ［C］// Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing. Stroudsburg， PA： ACL， 2020： 1295-1309. 10.18653/v1/2020.emnlp-main.99
15	FANG Y， SUN S， GAN Z， et al. Hierarchical graph network for multi-hop question answering ［C］// Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing. Stroudsburg， PA： ACL， 2020： 8823-8838. 10.18653/v1/2020.emnlp-main.710
16	ZHENG C， KORDJAMSHIDI P. SRLGRN： Semantic role labeling graph reasoning network ［C］// Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing. Stroudsburg， PA： ACL， 2020： 8881-8891. 10.18653/v1/2020.emnlp-main.714
17	SCHLICHTKRULL M S， KIPF T N， BLOEM P， et al. Modeling relational data with graph convolutional networks ［C］// Proceedings of the 2018 European Semantic Web Conference， LNCS 10843. Cham： Springer， 2018： 593-607.
18	VASHISHTH S， SANYAL S， NITIN V， et al. Composition-based multi-relational graph convolutional networks ［EB/OL］. （2020-01-18）［2022-06-20］. .
19	SPEER R， CHIN J， HAVASI C. ConceptNet 5.5： An open multilingual graph of general knowledge ［C］// Proceedings of the 31st AAAI Conference on Artificial Intelligence. Palo Alto， CA： AAAI Press， 2017： 4444-4451. 10.1609/aaai.v31i1.11164
20	LI R， CAO Y， ZHU Q， et al. How does knowledge graph embedding extrapolate to unseen data： a semantic evidence view ［C］// Proceedings of the 36th AAAI Conference on Artificial Intelligence. Palo Alto， CA： AAAI Press， 2022： 5781-5791. 10.1609/aaai.v36i5.20521
21	CHOUDHURY A， SHARMA S， MITRA P， et al. SimCat： An entity similarity measure for heterogeneous knowledge graph with categories ［C］// Proceedings of the 2nd ACM IKDD Conference on Data Sciences. New York： ACM， 2015： 112-113. 10.1145/2732587.2732604
22	ZHU G， IGLESIAS C A. Computing semantic similarity of concepts in knowledge graphs ［J］. IEEE Transactions on Knowledge and Data Engineering， 2017， 29（1）： 72-85. 10.1109/tkde.2016.2610428
23	HARRIS Z S. Distributional structure ［J］. Word， 1954， 10（2/3）： 146-162. 10.1080/00437956.1954.11659520
24	SRIVASTAVA N， HINTON G， KRIZHEVSKY A， et al. Dropout： a simple way to prevent neural networks from overfitting ［J］. Journal of Machine Learning Research， 2014， 15： 1929-1958.
25	LIU L， JIANG H， HE P， et al. On the variance of the adaptive learning rate and beyond ［EB/OL］. （2021-10-26）［2022-06-20］. .
26	PASCANU R， MIKOLOV T， BENGIO Y. On the difficulty of training recurrent neural networks ［C］// Proceedings of the 30th International Conference on Machine Learning. New York： JMLR.org， 2013： 1310-1318.
27	SANTORO A， RAPOSO D， BARRETT D G， et al. A simple neural network module for relational reasoning ［C］// Proceedings of the 31st International Conference on Neural Information Processing Systems. Red Hook， NY： Curran Associates Inc.， 2017： 4974-4983.
28	WANG X， KAPANIPATHI P， MUSA R， et al. Improving natural language inference using external knowledge in the science questions domain ［C］// Proceedings of the 33rd AAAI Conference on Artificial Intelligence. Palo Alto， CA： AAAI Press， 2019： 7208-7215. 10.1609/aaai.v33i01.33017208
29	LIU Y， OTT M， GOYAL N， et al. RoBERTa： A robustly optimized BERT pretraining approach ［EB/OL］. （2019-07-26）［2022-06-20］. .
30	CLARK P， ETZIONI O， KHASHABI D， et al. From F to A on the New York regents science exams： An overview of the aristo project ［J］. AI Magazine， 2020， 41（4）： 39-53. 10.1609/aimag.v41i4.5304

[1]	王春雷, 王肖, 刘凯. 多模态知识图谱表示学习综述[J]. 《计算机应用》唯一官方网站, 2024, 44(1): 1-15.
[2]	张秋余, 温永旺. 用于语音检索的三联体深度哈希方法[J]. 《计算机应用》唯一官方网站, 2023, 43(9): 2910-2918.
[3]	郑浩东, 马华, 谢颖超, 唐文胜. 融合遗忘因素与记忆门的图神经网络知识追踪模型[J]. 《计算机应用》唯一官方网站, 2023, 43(9): 2747-2752.
[4]	杨昊, 张轶. 基于上下文信息和多尺度融合重要性感知的特征金字塔网络算法[J]. 《计算机应用》唯一官方网站, 2023, 43(9): 2727-2734.
[5]	马国帅, 钱宇华, 张亚宇, 李俊霞, 刘郭庆. 动态异构信息融合的科研合作潜力预测[J]. 《计算机应用》唯一官方网站, 2023, 43(9): 2775-2783.
[6]	袁国龙, 张玉金, 刘洋. 基于残差反馈和自注意力的图像篡改取证网络[J]. 《计算机应用》唯一官方网站, 2023, 43(9): 2925-2931.
[7]	王宏, 钱清, 王欢, 龙永. 融合大核注意力卷积的轻量化图像篡改定位算法[J]. 《计算机应用》唯一官方网站, 2023, 43(9): 2692-2699.
[8]	潘润超, 虞启山, 熊泓霏, 刘智慧. 基于深度图神经网络的协同推荐算法[J]. 《计算机应用》唯一官方网站, 2023, 43(9): 2741-2746.
[9]	梁美佳, 刘昕武, 胡晓鹏. 基于改进YOLOv3的列车运行环境图像小目标检测算法[J]. 《计算机应用》唯一官方网站, 2023, 43(8): 2611-2618.
[10]	刘源, 董永权, 贾瑞, 杨昊霖. 面向个性化课程推荐的分层分期注意力网络模型[J]. 《计算机应用》唯一官方网站, 2023, 43(8): 2358-2363.
[11]	马胜位, 黄瑞章, 任丽娜, 林川. 基于多层语义融合的结构化深度文本聚类模型[J]. 《计算机应用》唯一官方网站, 2023, 43(8): 2364-2369.
[12]	段升位, 程欣宇, 王浩舟, 王飞. 基于改进的YOLOv5的大坝表面病害检测算法[J]. 《计算机应用》唯一官方网站, 2023, 43(8): 2619-2629.
[13]	王静红, 周志霞, 王辉, 李昊康. 双路自编码器的属性网络表示学习[J]. 《计算机应用》唯一官方网站, 2023, 43(8): 2338-2344.
[14]	衡红军, 杨鼎诚. 知识增强的方面词交互图神经网络[J]. 《计算机应用》唯一官方网站, 2023, 43(8): 2412-2419.
[15]	崔雨萌, 王靖亚, 刘晓文, 闫尚义, 陶知众. 融合注意力和裁剪机制的通用文本分类模型[J]. 《计算机应用》唯一官方网站, 2023, 43(8): 2396-2405.

融入三维语义特征的常识推理问答方法

Commonsense reasoning and question answering method with three-dimensional semantic features

RichHTML

PDF

可视化

摘要/Abstract

引用本文

使用本文

图/表 8

参考文献 30

相关文章 15

编辑推荐

Metrics