Gated Recursive and Sequential Deep Hierarchical Encoding for Detecting Incongruent News Articles

Anonymous

Abstract: With the increase in misinformation across digital platforms, incongruent news detection is becoming an important research problem. Earlier, researchers have exploited various feature engineering approaches and deep learning models with embedding to capture incongruity between news headlines and the body. Recent studies have also shown the advantages of capturing structural properties of the body using hierarchical encoding. Hierarchical encoding decomposes the body of a news article into smaller segments such as sentences or paragraphs. However, the existing hierarchical methods have not considered two important aspects; (i) deeper hierarchical level, and (ii) importance of different paragraphs in generating document encoding. Motivated by this, in this paper, we propose a Gated RecursiveAnd Sequential Deep HierarchicalEncoding (GRASHE) method for detectingincongruent news articles by extends hierarchicalencoding upto word level and incorporatingincongruently weight of each paragraph. Experimental results show that the proposed models outperform the bag-of-word features, sequential and hierarchical encoding-based counterparts. We also perform various ablation analysis to support the proposed models.

Paper Type: long