arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2001.03844cs.CL
收藏

再思考神经模型的泛化:一个命名实体识别的案例研究

Rethinking Generalization of Neural Models: A Named Entity Recognition Case Study

Fudan University(复旦大学);Carnegie Mellon University(卡内基梅隆大学)

Jinlan Fu, Pengfei Liu, Qi Zhang, Xuanjing Huang

摘要:

虽然基于神经网络的模型在大量的NLP任务上取得了令人印象深刻的性能,但不同模型的泛化行为仍然知之甚少:这种出色的性能是否意味着一个完美的泛化模型,或者仍然存在一些限制?本文以NER任务为测试平台,从不同角度分析了现有模型的泛化行为,并通过我们提出的指标来表征它们泛化能力的差异,从而指导我们更好地设计模型和训练方法。深入分析的实验诊断了现有神经NER模型在故障性能分析、注释错误、数据集偏差和类别关系方面的瓶颈,并提出了改进方向。我们已经发布了数据集:(ReCoNLL,PLONER),用于未来的研究,在我们的项目页面:http://pfliu.com/InterpretNER/。作为本文的副产品,我们开源了一个项目,该项目涉及最近NER论文的全面总结,并将其分类为不同的研究主题:https://github.com/pfliu-nlp/Named-Entity-Recognition-NER-Papers。

英文摘要:

While neural network-based models have achieved impressive performance on a large body of NLP tasks, the generalization behavior of different models remains poorly understood: Does this excellent performance imply a perfect generalization model, or are there still some limitations? In this paper, we take the NER task as a testbed to analyze the generalization behavior of existing models from different perspectives and characterize the differences of their generalization abilities through the lens of our proposed measures, which guides us to better design models and training methods. Experiments with in-depth analyses diagnose the bottleneck of existing neural NER models in terms of breakdown performance analysis, annotation errors, dataset bias, and category relationships, which suggest directions for improvement. We have released the datasets: (ReCoNLL, PLONER) for the future research at our project page: http://pfliu.com/InterpretNER/. As a by-product of this paper, we have open-sourced a project that involves a comprehensive summary of recent NER papers and classifies them into different research topics: https://github.com/pfliu-nlp/Named-Entity-Recognition-NER-Papers.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑