Query Graph Generation for Answering Multi-hop Complex Questions from Knowledge Bases

ACL, pp. 969-974, 2020.

Cited by: 0|Bibtex|Views47
EI
Other Links: dblp.uni-trier.de|academic.microsoft.com
Weibo:
In this paper we proposed a modified staged query graph generation method to deal with complex questions with both multi-hop relations and constraints

Abstract:

Previous work on answering complex questions from knowledge bases usually separately addresses two types of complexity: questions with constraints and questions with multiple hops of relations. In this paper, we handle both types of complexity at the same time. Motivated by the observation that early incorporation of constraints into quer...More
0
Introduction
  • Knowledge base question answering (KBQA) aims at answering factoid questions from a knowledge base (KB).
  • In the question “Who was the first president of the U.S.?” there is a single relation “president of” between the answer entity and the entity “U.S.,” but the authors have the constraint “first” that needs to be satisfied
  • For this type of complex questions, a staged query graph generation method has been proposed, which first identifies a single-hop relation path and adds constraints to it to form a query graph (Yih et al, 2015; Bao et al, 2016; Luo et al, 2018).
  • Little work has been done to deal with both types of complexity together
Highlights
  • Knowledge base question answering (KBQA) aims at answering factoid questions from a knowledge base (KB)
  • Work on Knowledge base question answering focused on simple questions containing a single relation (Yih et al, 2014; Bordes et al, 2015; Dong et al, 2015; Hao et al, 2017)
  • Real questions are often more complex and recently some studies looked into complex Knowledge base question answering
  • We propose to modify the staged query graph generation method by allowing longer relation paths
  • 10The knowledge base can be downloaded from https: //developers.google.com/freebase/
  • In this paper we proposed a modified staged query graph generation method to deal with complex questions with both multi-hop relations and constraints
Methods
  • Methods for Comparison

    3 Experiments

    3.1 Implementation Details

    The authors' method requires entities to be identified from the questions and linked to their corresponding entries in the KB.
  • The authors make use of the training questions and their answers to learn a linking model.
  • For the hyper-parameters in BERT model, the authors set the dropout ratio as 0.1, the hidden size as 768.
  • The authors use the standard BERT model (Devlin et al, 2019) to process the entire sequence and derive a score at the top layer.
  • Note that the authors fine-tune the pre-trained BERT parameters during learning
Results
  • The authors show the overall comparison in Table 1b. The authors can see that on the CWQ dataset, the method clearly achieves the best performance in terms of

    10The KB can be downloaded from https: //developers.google.com/freebase/.

    11Note that the authors treat 2-hop relation paths with CVT nodes as 1-hop paths.

    12We note that on the leaderboard of CWQ the best Prec@1 was achieved by Sun et al (2019).
  • The authors can see that on the CWQ dataset, the method clearly achieves the best performance in terms of.
  • Their method uses annotated topic entities and is not comparable here.
Conclusion
  • In this paper the authors proposed a modified staged query graph generation method to deal with complex questions with both multi-hop relations and constraints.
  • Experiments showed the method substantially outperformed existing methods on the ComplexWebQuestions dataset and outperformed the previous state of the art on two other KBQA datasets
Summary
  • Introduction:

    Knowledge base question answering (KBQA) aims at answering factoid questions from a knowledge base (KB).
  • In the question “Who was the first president of the U.S.?” there is a single relation “president of” between the answer entity and the entity “U.S.,” but the authors have the constraint “first” that needs to be satisfied
  • For this type of complex questions, a staged query graph generation method has been proposed, which first identifies a single-hop relation path and adds constraints to it to form a query graph (Yih et al, 2015; Bao et al, 2016; Luo et al, 2018).
  • Little work has been done to deal with both types of complexity together
  • Methods:

    Methods for Comparison

    3 Experiments

    3.1 Implementation Details

    The authors' method requires entities to be identified from the questions and linked to their corresponding entries in the KB.
  • The authors make use of the training questions and their answers to learn a linking model.
  • For the hyper-parameters in BERT model, the authors set the dropout ratio as 0.1, the hidden size as 768.
  • The authors use the standard BERT model (Devlin et al, 2019) to process the entire sequence and derive a score at the top layer.
  • Note that the authors fine-tune the pre-trained BERT parameters during learning
  • Results:

    The authors show the overall comparison in Table 1b. The authors can see that on the CWQ dataset, the method clearly achieves the best performance in terms of

    10The KB can be downloaded from https: //developers.google.com/freebase/.

    11Note that the authors treat 2-hop relation paths with CVT nodes as 1-hop paths.

    12We note that on the leaderboard of CWQ the best Prec@1 was achieved by Sun et al (2019).
  • The authors can see that on the CWQ dataset, the method clearly achieves the best performance in terms of.
  • Their method uses annotated topic entities and is not comparable here.
  • Conclusion:

    In this paper the authors proposed a modified staged query graph generation method to deal with complex questions with both multi-hop relations and constraints.
  • Experiments showed the method substantially outperformed existing methods on the ComplexWebQuestions dataset and outperformed the previous state of the art on two other KBQA datasets
Tables
  • Table1: a) Some statistics of CWQ and WQSP. CONS stands for constraints. (b) Comparison between our method and existing work. † denotes our re-implementation. (c) Ablation study on the CWQ dataset
Download tables as Excel
Funding
  • This research is supported by the National Research Foundation, Singapore under its International Research Centres in Singapore Funding Initiative
Reference
  • Ghulam Ahmed Ansari, Amrita Saha, Vishwajeet Kumar, Mohan Bhambhani, Karthik Sankaranarayanan, and Soumen Chakrabarti. 2019. Neural program induction for kbqa without gold programs or query annotations. In Proceedings of the IJCAI, pages 4890– 4896.
    Google ScholarLocate open access versionFindings
  • Yunshi Lan, Shuohang Wang, and Jing Jiang. 2019b. Multi-hop knowledge base question answering with an iterative sequence matching model. In Proceedings of the ICDM, pages 359–368.
    Google ScholarLocate open access versionFindings
  • Junwei Bao, Nan Duan, Zhao Yan, Ming Zhou, and Tiejun Zhao. 2016. Constraint-based question answering with knowledge graph. In Proceedings of the COLING, pages 2503–2514.
    Google ScholarLocate open access versionFindings
  • Nikita Bhutani, Xinyi Zheng, and H V Jagadish. 2019. Learning to answer complex questions over knowledge bases with query composition. In Proceedings of the CIKM, pages 739–748.
    Google ScholarLocate open access versionFindings
  • Antoine Bordes, Nicolas Usunier, Sumit Chopra, and Jason Weston. 201Large-scale simple question answering with memory networks. CoRR, abs/1506.02075.
    Findings
  • Antonie Bordes, Sumit Chopra, and Jason Weston. 2014. Question answering with subgraph embeddings. In Proceedings of the EMNLP, pages 615– 620.
    Google ScholarLocate open access versionFindings
  • Zi-Yuan Chen, Chih-Hung Chang, Yi-Pei Chen, Jijnasa Nayak, and Lun-Wei Ku. 2019. UHop: An unrestricted-hop relation extraction framework for knowledge-based question answering. In Proceedings of the NAACL-HLT, pages 345–356.
    Google ScholarLocate open access versionFindings
  • Rajarshi Das, Shehzaad Dhuliawala, Manzil Zaheer, Luke Vilnis, Ishan Durugkar, Akshay Krishnamurthy, Alex Smola, and Andrew McCallum. 201Go for a walk and arrive at the answer: Reasoning over paths in knowledge bases using reinforcement learning. In International Conference on Learning Representations.
    Google ScholarLocate open access versionFindings
  • Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 201Bert: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of the NAACL, pages 4171– 4186.
    Google ScholarLocate open access versionFindings
  • Li Dong, Furu Wei, Ming Zhou, and Ke Xu. 2015. Question answering over freebase with multicolumn convolutional neural networks. In Proceedings of the ACL-IJCNLP, pages 260–269.
    Google ScholarLocate open access versionFindings
  • Yanchao Hao, Yuanzhe Zhang, Kang Liu, Shizhu He, Zhanyi Liu, Hua Wu, and Jun Zhao. 2017. An endto-end model for question answering over knowledge base with cross-attention combining global knowledge. In Proceedings of the ACL, pages 221– 231.
    Google ScholarLocate open access versionFindings
  • Chen Liang, Jonathan Berant, Quoc Le, Kenneth D. Forbus, and Ni Lao. 2017. Neural symbolic machines: learning semantic parsers on freebase with weak supervision. In Proceedings of the ACL, pages 23–33.
    Google ScholarLocate open access versionFindings
  • Kangqi Luo, Fengli Lin, Xusheng Luo, and Kenny Zhu. 2018. Knowledge base question answering via encoding of complex query graphs. In Proceedings of the EMNLP, pages 2185–2194.
    Google ScholarLocate open access versionFindings
  • Michael Petrochuk and Luke Zettlemoyer. 2018. Simple questions nearly solved: a new upperbound and baseline approach. In Proceedings of the EMNLP, pages 554–558.
    Google ScholarLocate open access versionFindings
  • Haitian Sun, Tania Bedrax-Weiss, and William W. Cohen. 2019. Pullnet: Open domain question answering with iterative retrieval on knowledge bases and text. In Proceedings of the EMNLP, pages 2380– 2390.
    Google ScholarLocate open access versionFindings
  • Alon Talmor and Jonathan Berant. 2018. The web as a knowledge-base for answering complex questions. In Proceedings of the NAACL-HLT, pages 641–651.
    Google ScholarLocate open access versionFindings
  • Kun Xu, Siva Reddy, Yansong Feng, Songfang Huang, and Dongyan Zhao. 2016. Question answering on Freebase via relation extraction and textual evidence. In Proceedings of the ACL, pages 2326–2336.
    Google ScholarLocate open access versionFindings
  • Wen-tau Yih, Ming-Wei Chang, Xiaodong He, and Jianfeng Gao. 2015. Semantic parsing via staged query graph generation: Question answering with knowledge base. In Proceedings of the ACLIJCNLP, pages 1321–1331.
    Google ScholarLocate open access versionFindings
  • Wen-tau Yih, Xiaodong He, and Christopher Meek. 2014. Semantic parsing for single-relation question answering. In Proceedings of the ACL, pages 643– 648.
    Google ScholarLocate open access versionFindings
  • Mo Yu, Wenpeng Yin, Kazi Saidul Hasan, Cicero dos Santos, Bing Xiang, and Bowen Zhou. 2017. Improved neural relation detection for knowledge base question answering. In Proceedings of the ACL, pages 571–581.
    Google ScholarLocate open access versionFindings
  • Sen Hu, Lei Zou, and Xinbo Zhang. 2018. A state-transition framework to answer complex questions over knowledge base. In Proceedings of the EMNLP, pages 2098–2108.
    Google ScholarLocate open access versionFindings
  • Yunshi Lan, Shuohang Wang, and Jing Jiang. 2019a. Knowledge base question answering with topic units. In Proceedings of the IJCAI, pages 5046–5052.
    Google ScholarLocate open access versionFindings
Full Text
Your rating :
0

 

Tags
Comments