Indonesian Hoax News Detection Using IndoBERT and Knowledge Graph

Abstract

The spread of hoax news in Indonesian online media has become an increasing concern, highlighting the need for accurate automatic detection systems. This study proposes an integrated hoax detection approach that combines IndoBERT, which captures semantic context in Indonesian text, with a Knowledge Graph that verifies relationships between entities. IndoBERT's performance is also compared with SVM+TF-IDF, FastText, 1D-CNN, and LSTM. Experimental results show that IndoBERT achieves the highest performance, reaching 98.05% accuracy on headline data and 96.35% on narrative data using a dataset of 8,433 news articles. Although the Knowledge Graph alone produces lower accuracy, it enhances information verification by validating entity relationships. Integrating IndoBERT and the Knowledge Graph using a 90:10 weighting scheme further improves overall performance, achieving 98.00% accuracy on headline data and 96.20% on narrative data. The proposed approach provides a more accurate and comprehensive solution for Indonesian hoax detection.