Unsupervised clustering model for misinformation detection using knowledge base triples and word embedding techniques

In the modern upscaling technology age, wide spread of false information especially via social media and other mediums, poses a significant challenge. Existing systems use supervised models for misinformation classification which presents robust performance on labeled data but, demonstrate lower performance compared to unsupervised models on unlabeled data. In scenarios where unlabeled data is to be handled, the Clustering model is the best choice. Hence, to classify misinformation, the work proposes a Clustering model which examines the authenticity of knowledge base triples before they are incorporated in the information systems. In addition to TF-IDF and Count Vectorizers, Word2Vec and GloVe are used for feature vectorization. Empirical results showcase the robust performance of clustering models with accuracy of 89 % for unlabeled data, thereby offering a viable approach for improving the reliability of knowledge bases.