Fake News Detection with the New German Dataset 'GermanFakeNC'

摘要

The spread of misleading information and "alternative facts" on the internet gained in the last decade considerable importance worldwide. In recent years, several attempts have been made to counteract fake news based on automatic classification via machine; learning models. These, however, require labeled data. The scarcity of available corpora for predictive modeling is a major stumbling block in this field, especially in other languages than English. Our contribution is twofold. First, we introduce a new publicly available German dataset "German Fake News Corpus" (GermanFakeNC) for the task of fake news detection which consists of 490 manually fact-checked articles. Every false statement in the text is verified claim-by-claim by authoritative sources. Our ground truth for trustworthy news consists of 4,500 news articles from well-known mainstream news publishers. With regard to the second contribution, we choose a Convolutional Neural Network (CNN) (k = 0.89) and the widely used SVM (k = 0.72) technique to detect fake news. Thus we hope that our approach will stimulate the progress in fake news detection and claim verification across languages.