Abstract:

Abstract The state-of-the-art misinformation detection techniques mainly focused on supervised learning approach, however, it requires a huge amount of labeled dataset resulting into manual efforts and delays in detecting misinformation. Thus, an unsupervised approach to misinformation detection is in demand. The researchers with unsupervised misinformation detection show average performance as they lack in generating important textual and user-specific features. Further, since the data in the real world is time- sensitive, a large amount of data is generated over a period of time and the models need to adapt to this newly arriving chunk of data. To tackle the above problems, the authors have proposed a first-of-its-kind unsupervised misinformation detection model using an incremental learning approach that can handle newly arriving data without needing to label the data. To evaluate the model's performance, the authors have used various metrics like silhouette score, purity, and importance of various features in cluster formation. The model showed a purity score of 0.92 % and average silhouette score of 0.57%.

Keywords: