Automatic Estimation of Web Page Credibility

Abstract

This paper proposes a computational method for estimating Web page credibility based on the properties of a Web page or its text. The proposed method consists of two processes: the process of judging whether a given Web page has each of the properties influencing Web page credibility and the process of assessing the degree of Web credibility using the properties judged to be possessed by the given Web page. Concerning what properties really influence people's judgment of Web page credibility, we conducted a questionnaire survey and obtained 68 properties and their degree of influence on credibility. We then developed a method for judging the presence or absence of 40 properties. As a method for assessing the degree of Web credibility, we also developed an aggregative method (i.e., calculating the sum of the degree of influence) and a machine learning method using Support Vector Machine (SVM). We then conducted an evaluation experiment for binary classification of credible or incredible Web pages and verified the effectiveness of the proposed method.