First-Impression-Based Unreliable Web Pages Detection – Does First Impression Work?

Considering the continuous increase in the number of web pages worldwide, detecting unreliable pages, such as those containing fake news, is indispensable. Natural language processing and social-information-based methods have been proposed for web page credibility evaluation. However, the applicability of the former to web pages is limited because a model is required for each language, while the latter is poorly adapted to changes, owing to its dependence on external services that can be discontinued. To solve these problems, herein we propose a first-impression-based web credibility evaluation method. Our experimental evaluation of a fake news corpus gave an accuracy of 0.898, which is superior to those of existing methods.