A Headline-Centric Graph-Based Dual Context Matching Approach for Incongruent News Detection

The prevalence of incongruent news has demonstrated its significant role in propagating fake news, which catalyzes the dissemination of both misinformation and disinformation. Consequently, detecting incongruent news articles is an important research problem to counter early spreading of misinformation. In the literature, researchers have explored various bag-of-word-based features, news body-centric and news headline-centric encoding methods for incongruent news article detection. However, headline-centric and body-centric approaches in the literature fail to detect partially incongruent articles efficiently. Motivated by the above limitations, this study proposes graph-based dual context matching (GDCM), which first represents headlines and news bodies as a
bigram
network to capture contextual relations between words and document structure. For every word in the headline, GDCM extracts dual contexts (positive and negative) from the
bigram
network representing news body and estimates similarity between dual contexts and the headline for incongruent news detection. We conduct extensive experiments on three publicly available benchmark datasets and compare its performance with 16 baseline models. Our experimental results suggest that the proposed model outperforms existing state-of-the-art models and efficiently detects partially incongruent news. We further validate the performance of the proposed model through several ablation studies. The following key observations can be made from the ablation studies: 1) extracting dual
bigram
context of words in the headline from different segments of news body and then estimating the similarity between dual
bigram
contexts from news body and the headline helps in incongruent news detection and also helps in detecting partial incongruent news efficiently; and 2) representing news headlines and bodies in the form of a network based on
bigram
context helps to capture better nonlinear and contextual relationships between headline and body.