整合腳本知識的機器常識閱讀理解

Modeling Script Knowledge for Machine Commonsense Reading Comprehension

劉宏國(Hung-Kuo Liu)

指導教授 ： 陳信希

國立臺灣大學/電機資訊學院/資訊工程學研究所/碩士(2019年)

https://doi.org/10.6342/NTU201900781

摘要

近年來，在機器閱讀理解的任務中引入常識知識是廣泛被討論的研究議題。 以往大多數的研究都使用ConceptNet來推斷抽象概念，並幫助他們的模型來回答閱讀理解中的問題。然而，很少有研究採用腳本知識來改進或增強機器閱讀理解的模型。本文透過結合腳本知識對各種類型的常識進行建模，提出了一種新的機器閱讀理解的模型。而實驗結果顯示, 我們的模型在MCScript的數據集上達到了在SemEval-2018 Task 11中最佳的性能也提升了在COIN數據集的效能。

關鍵字

機器閱讀理解 ； 腳本知識 ； 常識知識

並列摘要

Introducing commonsense knowledge to the machine reading comprehension (MRC) task attracts attention in recent years. Most studies use ConceptNet to inference the abstract concepts and help their models answer the questions in reading comprehension. However, few studies employ Script knowledge to improve their MRC models. This thesis proposes a novel model for MRC by incorporating Script knowledge for modeling the various types of commonsense. Experimental results show that our model achieves the best performance on the MCScript dataset in the SemEval-2018 Task 11 and it increases the accuracy on the COIN dataset.

並列關鍵字

Machine Reading Comprehension ； Script Knowledge ； Commonsense Knowledge

參考文獻

Chen, Z., Cui, Y., Ma, W., Wang, S., Liu, T., & Hu, G. (2018). HFL-RC system at SemEval-2018 task 11: hybrid multi-aspects model for commonsense reading comprehension. arXiv preprint arXiv:1803.05655.

Chung, J., Gulcehre, C., Cho, K., & Bengio, Y. (2014). Empirical evaluation of gated recurrent neural networks on sequence modeling. arXiv preprint arXiv:1412.3555.

Cui, Y., Chen, Z., Wei, S., Wang, S., Liu, T., & Hu, G. (2016). Attention-over-attention neural networks for reading comprehension. arXiv preprint arXiv:1607.04423.

González, J.-Á., Hurtado, L.-F., Segarra, E., & Pla, F. (2018). ELiRF-UPV at SemEval-2018 Task 11: Machine Comprehension using Commonsense Knowledge. Paper presented at the Proceedings of The 12th International Workshop on Semantic Evaluation.

Hochreiter, S., & Schmidhuber, J. (1997). Long short-term memory. Neural computation, 9(8), 1735-1780.