Multi-Hop Question Generation with Knowledge Graph-Enhanced Language Model
Abstract
:1. Introduction
- The utilization of a GPT-2 model for encoding the context and answer allows for better semantic understanding, leveraging the pre-trained knowledge from its large-scale corpus.
- The bidirectional attention mechanism in the answer-aware GAT enables the model to dynamically capture the interactions between the answer and the knowledge graph, thus leading to a more comprehensive and accurate aggregation of the relevant information. Furthermore, the Graph2Context encoding allows the context representations to be updated with the knowledge derived from the knowledge graph, providing the model with a better semantic understanding of the context and the answer. These innovations enable the model to have a more thorough understanding of the context and answer, and, therefore, generate higher quality questions.
- Additionally, the multi-head attention generation module adopts a multi-head attention mechanism to capture the relationships among the latent representations, resulting in a more comprehensive understanding of the context. Furthermore, the module generates questions based on the enhanced understanding of the context, ensuring the fluency and logicality of the generated questions. The technical details of the module and the entire approach are thoroughly described in Section 2 of the paper.
2. Related Work
3. Methodology
3.1. Context Understanding
3.2. Question Generation
4. Results
4.1. Dataset and Metrics
4.2. Baselines and Ablation Settings
4.3. Comparison with Baselines
4.4. Ablation Tests Analysis
4.5. Case Study
- The high fluency scores obtained by both KGEL and GPT-2 with a pre-trained language model highlight the ability of these models to generate grammatically correct and logical questions. However, the lower answerability and completeness scores of GPT-2 compared to human performance indicate its limitations in identifying key information from the context and determining “what to ask”.
- The improved completeness score of KGEL compared to GPT-2 highlights the effectiveness of our proposed reasoning module in recognizing multi-hop relationships and identifying relevant entities. This demonstrates the better capability of our model in generating questions with multi-hop relations.
- The better performance of KGEL in terms of answerability, compared to GPT-2, is a result of the incorporation of entity graph in our reasoning module. However, further improvement is required in the ability of the model to identify the correct target aspect to ask, in order to enhance answerability.
- The generation quality of multi-hop questions is far from desired;
- Incorporating external knowledge may help the model recognize the relationship between entities and answers;
- A copy mechanism has the potential to be introduced to help the model generate the shared semantic content in the target question for many words in question that come from context.
5. Conclusions
Author Contributions
Funding
Institutional Review Board Statement
Informed Consent Statement
Data Availability Statement
Conflicts of Interest
Abbreviations
MDPI | Multidisciplinary Digital Publishing Institute |
DOAJ | Directory of open access journals |
TLA | Three letter acronym |
LD | Linear dichroism |
Appendix A. The Illustrator of Data Processing
References
- Duan, N.; Tang, D.; Chen, P.; Zhou, M. Question generation for question answering. In Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, Copenhagen, Denmark, 7–11 September 2017; pp. 866–874. [Google Scholar]
- Mostafazadeh, N.; Brockett, C.; Dolan, W.B.; Galley, M.; Gao, J.; Spithourakis, G.; Vanderwende, L. Image-Grounded Conversations: Multimodal Context for Natural Question and Response Generation. In Proceedings of the Eighth International Joint Conference on Natural Language Processing (Volume 1: Long Papers), Taipei, Taiwan, 27 November–1 December 2017; pp. 462–472. [Google Scholar]
- Du, X.; Shao, J.; Cardie, C. Learning to Ask: Neural Question Generation for Reading Comprehension. In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), Vancouver, BC, Canada, 30 July–4 August 2017; pp. 1342–1352. [Google Scholar]
- Nema, P.; Mohankumar, A.K.; Khapra, M.M.; Srinivasan, B.V.; Ravindran, B. Let’s Ask Again: Refine Network for Automatic Question Generation. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), Hong Kong, China, 3–7 November 2019; pp. 3314–3323. [Google Scholar]
- Scialom, T.; Piwowarski, B.; Staiano, J. Self-attention architectures for answer-agnostic neural question generation. In Proceedings of the 57th annual meeting of the Association for Computational Linguistics, Florence, Italy, 28 July–2 August 2019; pp. 6027–6032. [Google Scholar]
- Zhao, Y.; Ni, X.; Ding, Y.; Ke, Q. Paragraph-level neural question generation with maxout pointer and gated self-attention networks. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, Brussels, Belgium, 31 October–4 November 2018; pp. 3901–3910. [Google Scholar]
- Pan, L.; Lei, W.; Chua, T.S.; Kan, M.Y. Recent advances in neural question generation. arXiv 2019, arXiv:1905.08949. [Google Scholar]
- Pan, L.; Xie, Y.; Feng, Y.; Chua, T.S.; Kan, M.Y. Semantic Graphs for Generating Deep Questions. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, Online, 5–10 July 2020; pp. 1463–1475. [Google Scholar]
- Gupta, D.; Chauhan, H.; Akella, R.T.; Ekbal, A.; Bhattacharyya, P. Reinforced Multi-task Approach for Multi-hop Question Generation. In Proceedings of the 28th International Conference on Computational Linguistics, Barcelona, Spain, 8–13 December 2020; pp. 2760–2775. [Google Scholar]
- Su, D.; Xu, Y.; Dai, W.; Ji, Z.; Yu, T.; Fung, P. Multi-hop Question Generation with Graph Convolutional Network. In Proceedings of the Findings of the Association for Computational Linguistics: EMNLP 2020, Online, 16–20 November 2020; pp. 4636–4647. [Google Scholar]
- Ji, H.; Ke, P.; Huang, S.; Wei, F.; Zhu, X.; Huang, M. Language Generation with Multi-Hop Reasoning on Commonsense Knowledge Graph. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), Online, 16–20 November 2020; pp. 725–736. [Google Scholar]
- Lin, B.Y.; Chen, X.; Chen, J.; Ren, X. KagNet: Knowledge-Aware Graph Networks for Commonsense Reasoning. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), Hong Kong, China, 3–7 November 2019; pp. 2829–2839. [Google Scholar]
- Ye, Z.X.; Chen, Q.; Wang, W.; Ling, Z.H. Align, mask and select: A simple method for incorporating commonsense knowledge into language representation models. arXiv 2019, arXiv:1908.06725. [Google Scholar]
- Radford, A.; Wu, J.; Child, R.; Luan, D.; Amodei, D.; Sutskever, I. Language models are unsupervised multitask learners. OpenAI Blog 2019, 1, 9. [Google Scholar]
- Lewis, M.; Liu, Y.; Goyal, N.; Ghazvininejad, M.; Mohamed, A.; Levy, O.; Stoyanov, V.; Zettlemoyer, L. BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, Online, 5–10 July 2020; pp. 7871–7880. [Google Scholar]
- Dong, L.; Yang, N.; Wang, W.; Wei, F.; Liu, X.; Wang, Y.; Gao, J.; Zhou, M.; Hon, H.W. Unified language model pre-training for natural language understanding and generation. In Proceedings of the Advances in Neural Information Processing Systems, Vancouver, BC, Canada, 8–14 December 2019; Volume 32. [Google Scholar]
- Heilman, M.; Smith, N.A. Good Question! Statistical Ranking for Question Generation. In Proceedings of the Human Language Technologies: The 2010 Annual Conference of the North American Chapter of the Association for Computational Linguistics, Los Angeles, CA, USA, 1–5 June 2010; Association for Computational Linguistics: Los Angeles, CA, USA, 2010; pp. 609–617. [Google Scholar]
- Chali, Y.; Hasan, S.A. Towards Topic-to-Question Generation. Comput. Linguist. 2015, 41, 1–20. [Google Scholar] [CrossRef]
- Zhou, Q.; Yang, N.; Wei, F.; Tan, C.; Bao, H.; Zhou, M. Neural question generation from text: A preliminary study. In Proceedings of the National CCF Conference on Natural Language Processing and Chinese Computing, Dalian, China, 8–12 November 2017; Springer: Berlin/Heidelberg, Germany, 2017; pp. 662–671. [Google Scholar]
- Liu, B.; Zhao, M.; Niu, D.; Lai, K.; He, Y.; Wei, H.; Xu, Y. Learning to generate questions by learningwhat not to generate. In Proceedings of the The World Wide Web Conference, San Francisco, CA, USA, 13–17 May 2019; pp. 1106–1118. [Google Scholar]
- Kipf, T.N.; Welling, M. Semi-supervised classification with graph convolutional networks. arXiv 2016, arXiv:1609.02907. [Google Scholar]
- Veličković, P.; Cucurull, G.; Casanova, A.; Romero, A.; Lio, P.; Bengio, Y. Graph attention networks. arXiv 2017, arXiv:1710.10903. [Google Scholar]
- Qiu, L.; Xiao, Y.; Qu, Y.; Zhou, H.; Li, L.; Zhang, W.; Yu, Y. Dynamically fused graph network for multi-hop reasoning. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, Florence, Italy, 28 July–2 August 2019; pp. 6140–6150. [Google Scholar]
- De Cao, N.; Aziz, W.; Titov, I. Question Answering by Reasoning Across Documents with Graph Convolutional Networks. In 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), Minneapolis, MN, USA, 2–7 June 2019; pp. 2306–2317. [Google Scholar]
- Welbl, J.; Stenetorp, P.; Riedel, S. Constructing datasets for multi-hop reading comprehension across documents. Trans. Assoc. Comput. Linguist. 2018, 6, 287–302. [Google Scholar] [CrossRef]
- Vaswani, A.; Shazeer, N.; Parmar, N.; Uszkoreit, J.; Jones, L.; Gomez, A.N.; Kaiser, Ł.; Polosukhin, I. Attention is all you need. In Proceedings of the Advances in Neural Information Processing Systems, Long Beach, CA, USA, 4–9 December 2017; Volume 30. [Google Scholar]
- Seo, M.; Kembhavi, A.; Farhadi, A.; Hajishirzi, H. Bidirectional attention flow for machine comprehension. arXiv 2016, arXiv:1611.01603. [Google Scholar]
- Yang, Z.; Qi, P.; Zhang, S.; Bengio, Y.; Cohen, W.; Salakhutdinov, R.; Manning, C.D. HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answering. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, Brussels, Belgium, 31 October–4 November 2018; pp. 2369–2380. [Google Scholar]
- Sharma, S.; El Asri, L.; Schulz, H.; Zumer, J. Relevance of Unsupervised Metrics in Task-Oriented Dialogue for Evaluating Natural Language Generation. Comput. Res. Repos. 2017, arXiv:1706.09799. [Google Scholar]
- Bahdanau, D.; Cho, K.; Bengio, Y. Neural machine translation by jointly learning to align and translate. Comput. Sci. 2014, arXiv:1409.0473. [Google Scholar]
- Kim, Y.; Lee, H.; Shin, J.; Jung, K. Improving neural question generation using answer separation. In Proceedings of the AAAI Conference on Artificial Intelligence, Honolulu, HI, USA, 27 January–1 February 2019; Volume 33, pp. 6602–6609. [Google Scholar]
Parameters | Values |
---|---|
Learning epoch | 50 |
Dropout | 0.1 |
Optimizer | AdamW |
Batch size | 32 |
Learning rate | to |
Warm up | reach max lr at 10 epoch |
Heads in GPT-2 | 12 |
Layers of GPT-2 | 12 |
Layers of GAT | 2 |
Embeddding size | 768 |
B1 | B2 | B3 | B4 | R-L | M | |
---|---|---|---|---|---|---|
seq2seq+Attn | 32.97 | 21.11 | 15.41 | 11.81 | 18.19 | 33.48 |
NQG++ | 35.51 | 22.12 | 15.53 | 11.50 | 16.96 | 32.01 |
ASs2s | 34.60 | 22.77 | 15.21 | 11.29 | 16.78 | 32.88 |
S2sa-at-mp-gsa | 35.36 | 22.38 | 15.88 | 11.85 | 17.63 | 33.02 |
GPT-2 | 39.00 | 24.46 | 17.21 | 12.71 | 16.81 | 32.06 |
KGEL | 41.93 | 28.04 | 20.83 | 16.13 | 19.70 | 35.28 |
Model | B1 | B2 | B3 | B4 | R-L | M |
---|---|---|---|---|---|---|
KGEL | 41.93 | 28.04 | 20.83 | 16.13 | 19.70 | 35.28 |
KGEL-AT | 41.14 | 27.32 | 20.00 | 15.24 | 20.12 | 35.53 |
KGEL-MHA | 38.84 | 25.00 | 18.03 | 13.66 | 17.46 | 32.98 |
KGEL-EL | 39.13 | 25.10 | 18.11 | 12.86 | 16.41 | 32.76 |
Support Fact A: | Ulysses is a decommissioned robotic space probe whose primary mission was to orbit the Sun and study it at all latitudes. |
Support Fact B: | His professional career has included work on the Voyager, Galileo, and Ulysses space missions, and is currently working as a consultant in aerospace power and propulsion systems. |
Answer: | Sun |
GPT-2: | What is the name of the support space of the space named after the physicist who has work on the voyager? |
KGEL: | David B. Mcdonald worked on the casting of the decommissioned spacecraft that are setting what landform? |
Ground Truth: | Gary L. Bennett was a part of the space missions that have a primary destination of what celestial body? |
Support Fact A: | In 2000, “Linux Format” called Qvwm “an unusually impressive imposter”. |
Support Fact B: | Linux Format was the UK’s first Linux-specific magazine, and as of 2013 was the best-selling Linux title in the UK. |
Answer: | An unusually impressive imposter |
GPT-2: | What was the first Linux-such first linux-specific magazine? |
KGEL: | What type of magazine called qvwm “an unusually impressive imposter”? |
Ground Truth: | What did the UK’s first Linux-specific magazine call Qvwm? |
Support Fact A: | SpaceDev, a part of the “Space Systems Business” of Sierra Nevada Corporation, is prominent for its spaceflight and microsatellite work. |
Support Fact B: | Sierra Nevada Corporation operates under the leadership of Chief Executive Officer, Fatih Ozmen and President, Eren Ozmen. |
Answer: | Fatih Ozmen |
GPT-2: | Who is the president of the company that operates Russian Li: Nobleman? |
KGEL: | Who is the president of the corporation that is a member of the space systems business? |
Ground Truth: | Who is the Chief Executive Officer of the corporation that owns Space Dev? |
Support Fact A: | Anne of Ingleside is a children’s novel by Canadian author Lucy Maud Montgomery. It was first published in July 1939 by McClelland and Stewart (Toronto) and the Frederick A. Stokes Company (New York). |
Support Fact B: | Lucy Maud Montgomery ‘1’:, ‘2’:, ‘3’:, ‘4’: (30 November 1874–24 April 1942) published as L.M. Montgomery, was a Canadian author best known for a series of novels beginning in 1908 with “Anne of Green Gables” The central character, Anne Shirley, an orphaned girl, made Montgomery famous in her lifetime and gave her an international following. |
Answer: | Lucy Maud Montgomery |
GPT-2: | Who wrote the central character’s novel in which the fictional female protagonist of the novel was first published in July 1939? |
KGEL: | Who is the author best known for a series of novels about Anne Shirley which was first published in July 1939, by Mcclelland and Stewart? |
Ground Truth: | Wich children’s novelist whow was first published in 1939 gained an internaional following writting about an orphaned girl named Anne Shirley? |
Model | Fluency | Answerability | Completeness |
---|---|---|---|
GPT-2 | 3.85 (±0.30) | 2.14 (±0.97) | 2.68 (±0.71) |
KGEL | 4.43 (±0.41) | 2.92 (±0.76) | 3.28 (±0.34) |
Human | 4.87 (±0.13) | 4.96 (±0.21) | 4.86 (±0.26) |
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |
© 2023 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).
Share and Cite
Li, Z.; Cao, Z.; Li, P.; Zhong, Y.; Li, S. Multi-Hop Question Generation with Knowledge Graph-Enhanced Language Model. Appl. Sci. 2023, 13, 5765. https://doi.org/10.3390/app13095765
Li Z, Cao Z, Li P, Zhong Y, Li S. Multi-Hop Question Generation with Knowledge Graph-Enhanced Language Model. Applied Sciences. 2023; 13(9):5765. https://doi.org/10.3390/app13095765
Chicago/Turabian StyleLi, Zhenping, Zhen Cao, Pengfei Li, Yong Zhong, and Shaobo Li. 2023. "Multi-Hop Question Generation with Knowledge Graph-Enhanced Language Model" Applied Sciences 13, no. 9: 5765. https://doi.org/10.3390/app13095765