命名實體過濾器使用於穩健的機器閱讀理解

Named Entity Filters for Robust Machine Reading Comprehension

彭于晏(Yu-Yan Peng)

指導教授 ： 許永真

國立臺灣大學/電機資訊學院/資訊網路與多媒體研究所/碩士(2018年)

https://doi.org/10.6342/NTU201801153

摘要

機器閱讀理解問題目的是從文章中抽取重要的資訊回答相關的問題。雖然有很多方法被提出，相似性干擾問題仍未被解決。相似性干擾問題指因為某些文章中的句子不包含答案卻跟問題很相似引起的錯誤。命名實體具有的獨特性可以用來區分這些相似的句子，讓模型不會遭受這些句子的干擾。在本論文中提出了命名實體過濾器。命名實體過濾器能善加利用命名實體所擁有的資訊減緩相似性干擾問題。論文中的實驗結果顯示命名實體過濾器能夠提升模型的穩健性，不減少SQuAD 上的 F1 分數，得到在兩個對抗式資料集 5% 到 10% F1 分數的提升。同時命名實體過濾器也能夠只損失不到 1% 原始資料集的 F1 分數簡單地提升其他現有的模型在對抗式資料集 5% F1 分數。

關鍵字

注意力機制 ； 命名實體 ； 類神經網路 ； 穩健性 ； 相似度

並列摘要

The machine reading comprehension problem aims to extract crucial information from the given document to answer the relevant questions.Although many methods regarding the problem have been proposed, the similarity distraction problem inside remains unsolved.The similarity distraction problem addresses the error caused by some sentences being very similar to the question but not containing the answer.Named entities have the uniqueness which can be utilized to distinguish similar sentences to prevent models from being distracted.In the thesis, named entity filters (NE filters) are proposed. NE filters can utilize the information of named entities to alleviate the similarity distraction problem.Experiment results in the thesis show that the NE filter can enhance the robustness of the used model. It increases 5% to 10% F1 score on two adversarial SQuAD datasets without decreasing the F1 score on the original SQuAD dataset.Besides, the NE filter easily increases 5% F1 score of other existing models on the adversarial datasets with less than 1% loss on the original one.

並列關鍵字

Attention mechanism ； Named entity ； Neural networks ； Robustness ； Similarity

參考文獻

[1] S. Bird. NLTK: the natural language toolkit. In Procedding of the 21st International Conference on Computational Linguistics and 44th Annual Meeting of the Association for Computational Linguistics, 2006.

[2] D. Chen, J. Bolton, and C. D. Manning. A Thorough Examination of the CNN/Daily Mail Reading Comprehension Task. In Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics, pages 2358–2367, 2016.

[3] Y. Cui, T. Liu, Z. Chen, S. Wang, and G. Hu. Consensus Attention-based Neural Networks for Chinese Reading Comprehension. In Proceedings of the 26th International Conference on Computational Linguistics, pages 1777—-1786, Osaka, Japan, 2016. The COLING 2016 Organizing Committee.

[4] M. Gardner, J. Grus, M. Neumann, O. Tafjord, P. Dasigi, N. F. Liu, M. Peters, M. Schmitz, and L. S. Zettlemoyer. Allennlp: A deep semantic natural language processing platform. 2017.

[5] Y. Gong and S. R. Bowman. Ruminating reader: Reasoning with gated multi-hop attention. CoRR, abs/1704.07415, 2017.