A Multi-Level Attention Model for Evidence-Based Fact Checking
Canasai Kruengkrai Junichi Yamagishi Xin Wang National Institute of Informatics, Japan
Abstract
Evidence-based fact checking aims to verify the truthfulness of a claim against evidence extracted from textual sources. Learning a representation that effectively captures relations between a claim and evidence can be challenging. Recent state-of-the-art approaches have developed increasingly sophisticated models based on graph structures. We present a simple model that can be trained on sequence structures. Our model enables inter-sentence attentions at different levels and can benefit from joint training. Results on a large-scale dataset for Fact Extraction and VERification (FEVER) show that our model outperforms the graph-based approaches and yields 1.09% and 1.42% improvements in label accuracy and FEVER score, respectively, over the best published model.111 The code and model checkpoints are available at: https://github.com/nii-yamagishilab/mla.
中文速览
用维基百科作为知识来源、自动判断一条声明真假的任务（基于证据的事实核查，evidence-based fact checking）面临一个核心难题：如何让模型同时理解声明与多条证据句子之间的关系。当前最优方法普遍采用复杂的图神经网络来建模这些关系，但本文提出，图结构其实并非必要——一个基于序列结构、在词级和句级都施加注意力机制的多级注意力模型（Multi-Level Attention，MLA），配合"混合真实与预测证据"来缩小训练与测试时的数据分布差距，以及对句子选择和真假判断的联合训练，就能取得更好的效果。在大规模基准数据集 FEVER 上，MLA 在标签准确率和综合 FEVER 分数上分别比此前最佳发表模型高出 1.09% 和 1.42%，超越了所有已发表的图结构方法。这一结果表明，为事实核查任务精心设计复杂图结构可能是"杀鸡用牛刀"，简洁的序列注意力方案反而更实用、更高效。
原文 arXiv:2106.00950;中英对照 + 大白话阅读 https://aha.fim.ai/paper/2106.00950v1