Anomaly Detection in Academic Social Networks using Deep Clustering
Researchpedia Journal of Computing, Volume 2, Issue 1, Article 8, Pages 81-86, December 2021
Khurram Shehzad1, Safiullah Afzal1, Muhammad Waqar1, Malik Khizar Hayat2
1Department of Software Engineering, Foundation University Islamabad, Rawalpindi Campus, Pakistan
2Department of Information Technology, University of Haripur, Pakistan
Corresponding author: Malik Khizar Hayat (e-mail: khizerhayat92@gmail.com)
ABSTRACT Detection of anomalies that are evolutionary by nature has emerged as a trending research topic in many areas, such as security, bioinformatics, education, economy and so on. Although, most of the research has focused on detecting anomalies using evolutionary behavior among objects in a network. However, in the real-world heterogenous networks, multiple types of objects co-evolve together with their attributes. To understand the deviant co-evolution of multi-typed objects in heterogeneous information networks (HINs), a special approach is required that can capture abnormal co-evolution of multi-typed objects. Detecting co-evolution-based anomaly in heterogeneous bibliographic information network can portray better the object-oriented semantics than just analyzing the co-author or citation network alone. In this paper, we propose a deep clustering-based model for anomaly detection in Microsoft Academic Graph (MAG). The star-network schema is used to process the MAG data. Feature learning and clustering tasks are combined using deep learning. Experimentation on the MAG data shows the efficiency of the proposed model.
Keywords Anomaly detection, Clustering, Deep learning, Heterogenous networks, Microsoft academic graph
