DEFENSE: Enhancing Fake News Detection on COVID Through Transformer Based Feature Engineering and Sentence Embedding Approach

EasyChair Preprint 15828

Abstract

The current global COVID-19 pandemic has wreaked havoc in our daily lives, both physically and mentally. An enormous
amount of fake news and misinformation about COVID-19 has spread fast across social media platforms, as people rely
heavily on them for current updates. Inestimable harm on human lives can be caused by the surmise, misconceptions,
fear and the spread of rumours. Detecting such fake news and blocking their spread is of predominant importance
and an influential research problem as well. Some primary challenges in fake news detection involve lack of contextual
understanding of the social media post and the absence of a concrete feature engineering mechanism in analysing
the contents of the post. In this article, we present DEFENSE, a Transformer-based model for fake news detection
in social media posts. We focus on constructing a precise and concrete feature engineering model to extract the
textual and sentimental features like sentiment polarity and sentiment subjectivity, of a post. Moreover, we use an
efficient mechanism to extract the contextual meaning of the post using various sentence embedding methods. In order
to reduce overfitting and increase accuracy, our model is trained to remove multi-collinearity through dimensionality
reduction, before classifying with an extensive set of classifiers. Comprehensive experiments on the benchmark dataset
namely Contraint@AAAI 2021 COVID-19 Fake News Detection Dataset(20) are performed to evaluate our method. The
results of our experiments demonstrate the efficacy of DEFENSE in detecting fake news, which significantly outperforms
a few of the state-of-the-art baselines with an Accuracy of 0.9472, increase in Precision by 8% and Recall by 3%, and
an F1-score of above 0.9.

Keyphrases: Accuracy, COVID-19, Classification, Fake News Detection, Sentence Embedding, social media

@booklet{EasyChair:15828,
  author    = {C Oswald and Allen Puthenparambil Alex},
  title     = {DEFENSE: Enhancing Fake News Detection on COVID Through Transformer Based Feature Engineering and Sentence Embedding Approach},
  howpublished = {EasyChair Preprint 15828},
  year      = {EasyChair, 2025}}