Suspicious news detection through semantic and sentiment measures

Abstract

Misinformation has always existed in society. Nowadays, the technological development and the appearance of social networks, pseudo-newspapers and blogs, have aggravated this problem by facilitating the rapid spread of malicious news. This fact makes it easier to use disinformation as an attack vector for huge communities. This has led to the development of procedures that detect the appearance of this type of news and mitigate its influence. This article presents the Knowledge Recovering Architecture based on Keywords Extraction from Narratives for Suspicious News Detection (KRAKEN-SND) system. Its main goal is to support human experts to detect suspicious news articles that should be verified. In order to achieve this objective, it gathers narratives from multiple reliable information sources. Then, it extracts the semantic and sentiment relevant features from these narratives. This information is structured by date using a conceptual graph to generate trustworthy knowledge. The system includes a novel similarity measure that combines three specific components. This measure uses the stored knowledge to detect the peculiarity of a reported narrative that may contain suspicious information. Several experiments using relevant topics as Brexit and the COVID-19 pandemic among others have been carried out to validate the proposal, obtaining promising results.