Leveraging One-Class SVM and Semantic Analysis to Detect Anomalous Content