[4Xin1] Poster session 2

[4Xin1-54] Improvement of Explainability in Multimodal Fake News Detection Using External Knowledge and Visualization

〇Kota Tanabe¹, Itsuki Noda¹, Satoshi Oyama¹ (1.Hokkaido University)

Keywords:Fake News, Explainability, Multimodal, External Knowledge

The purpose of this study is to improve explainability and interpretability in multimodal fake news detection. We propose a model that utilizes external knowledge obtained from DBpedia in addition to image/text pairs in the Fakeddit dataset. We evaluated the performance of the model with external knowledge, and confirmed that the explainability and interpretability of the model can be improved by visualizing results of the fake news detection model using Grad-CAM and Attention, and by presenting external knowledge.