Detecting Anomalous Events over Time Using RDF Triple Extraction and a Dynamic Implementation of OddBall

This paper shows a new approach for anomaly detection by combining the extraction of so-called triples consisting of a subject, predicate, and object using dynamic anomaly-detection. First, the methods used to extract triples and general principles of anomaly detection and event detection are discussed. Next, a novel approach is presented where extracted triples are converted into time-lapsed networks of triples on which anomaly and event detection methods from social network analysis are applied. Subsequently, the results of the experiments are presented together with the evaluation method used. Considering the results of the tested methods, the dynamic variation of the OddBall algorithm, which measures network changes over time, displays the connection between the predictions of our model and real-life events accurately.