Detecting Confirmation Biases Using Linguistic-based and User-based Approaches

Abstract

Logical fallacies and fake news are often intentionally created for the purpose of deception. Therefore, they are essentially similar in nature. In recent years, there has been a lot of research on detecting fake news, and many effective methods for detection are known. However, to the best of our knowledge, there is no attempt to detect logical fallacies. In this paper, we assume that logical fallacies can be detected by the same method as fake news, and attempt to detect them. The experiments summarized in this paper are: (1) four experiments to detect confirmation bias from linguistic features and user profile information using SVM and Random Forest, respectively; and (2) four experiments to detect confirmation bias by changing the acquisition method of distributed representation of words in the previous experiments using LSTM model to skip-gram and GloVe. The results of experiment are as follows. In experiment (1), the accuracy was about 62%, 61%, 69%, and 63%, respectively, and in experiment (2), the accuracy was about 73% and 76%, respectively.