期刊
NEURAL COMPUTING & APPLICATIONS
卷 35, 期 15, 页码 11337-11357出版社
SPRINGER LONDON LTD
DOI: 10.1007/s00521-023-08300-x
关键词
Fact checking; Deep learning; Sentence classification
This work focuses on identifying whether a sentence is factual, proposing a G2CW framework based on glove embedding and gated recurrent units. The framework detects if a sentence has check-worthy content and evaluates its importance from a fact-checking perspective. The proposed framework outperforms prior work on two datasets.
People are exposed to a lot of information daily, which is a mix of facts, opinions, and false claims. The rate at which information is created and spread has necessitated an automated fact-checking mechanism. In this work, we focus on the first step of the fact-checking system, which is to identify whether a given sentence is factual. We propose a glove embedding-based gated recurrent unit pipeline for check-worthy sentence detection, referred to as G2CW framework. It detects whether a given sentence has check-worthy content in it or not; furthermore, if it has check-worthy content, whether it is important or not, from a fact-checking perspective. We evaluate our proposed framework on two datasets: a standard ClaimBuster dataset commonly used by the research community for this problem and a self-curated IndianClaim dataset. Our G2CW framework outperforms prior work with 0.92 as F1-score. Furthermore, our G2CW framework, when trained on the ClaimBuster dataset, performs the best on the IndianClaims dataset.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据