低资源语言假新闻检测器的跨语言迁移学习
计算与语言
2022-08-29 v1 人工智能
摘要
由于缺乏训练数据，开发用于检测低资源语言中假新闻（FN）的方法一直受到阻碍。在本研究中，我们仅使用来自高资源语言的训练数据来解决该问题。我们的假新闻检测系统通过应用对抗学习在不同语言间迁移检测知识，从而实现了这一策略。为辅助知识迁移，我们的系统利用来源信息来判断文章的可信度，来源信息是一种代表说话者可信度的跨语言特征。在实验中，我们的系统比使用机器翻译训练数据集的系统准确率高出 3.71%。此外，我们提出的用于假新闻检测的跨语言特征利用使准确率提高了 3.03%。
引用
@article{arxiv.2208.12482,
title = {Cross-lingual Transfer Learning for Fake News Detector in a Low-Resource Language},
author = {Sangdo Han},
journal= {arXiv preprint arXiv:2208.12482},
year = {2022}
}
备注
I've withdraw this paper from a journal during revision. There were two reasons. First, data was not verified enough. Data verification steps are required. Second, even the average accuracy was higher than baseline, but it was not stable enough. However, I think the news embedding truely represents credibility of the speakers. I hope that the knowledge I've got would help for other researchers