Fake News Identification Based on Sentiment and Frequency Analysis

Abstract

The advent of social networks has changed how can be the thinking of the population influenced. Although the spreading of false information or false messages for personal or political benefit is certainly nothing new, current trends such as social media enable every individual to create false information easier than ever with the spread compared to the leading news portals. Fake news detection has recently attracted growing interest from the general public and researchers. The paper aims to compare basic text characteristics of fake and real news article types. We analysed two datasets that contained a total of 28 870 articles. The results were validated using the third data set consisting of 402 articles. The most important finding is the statistically significant difference in the news sentiment where it has been shown that fake news articles have a more negative sentiment. Also, an interesting result was the difference of average words per sentence. Finding statistically significant differences in individual text characteristics is a piece of important information for the future fake news classifier in terms of selecting the appropriate attributes for classification.