Multi Linear Tensor and Graph Convoluted Attention Network Based Classifier for Fake News Detection

Abstract

The growing acceptance of social media platforms has streamlined the news articles sharing that have induced the boom in fake news. With the appearance of fake news at a very swift rate, a pressing distress has brought out in our society due to extensive fake content propagation. The quality of the news content is uncertain and there prevails a requirement for the detection at an early stage. However detection of fake news in a swift and accurate manner is a laborious and cumbersome task. However, in terms of knowledge extraction, most existing methods lack the mining of both textual and feature knowledge hidden in the news content and discard the mutual correlation between them. To address on these gaps, in this work, a method called, Multi-linear Tensor and Graph Convoluted Attention Network-based Classifier (MT-GCANC) for fake news detection is designed. The MT-GCANC method is split into two sections, namely preprocessing, feature engineering and classification. In the preprocessing stage, Multi-linear Tensor function is applied to the raw dataset to generate computationally efficient preprocessed sample results. Second with the preprocessed sample results as input are subjected to Graph Convoluted Attention Network-based Knowledge-aware feature engineering. Here, both relevant features are extracted and then with the engineered features classification is performed for accurate and precise fake news detection. The performance of our proposed MT-GCANC method has been validated on Fake News dataset. Classification results have revealed that the proposed MT-GCANC method outperforms existing and relevant onsets for fake news detection and accomplished a precision and recall rate of 19% and 58%. These results have shown significant advancements over the existing state-of-the-art methods in the domain of fake news detection and state the probable utilization of the method for classifying fake news.