Authors: Mohammad Reza Keyvanpour; Mehrnoush Barani Shirzad; Maryam Ghaderi
Addresses: Department of Computer Engineering, Alzahra University, Tehran, Iran ' Data Mining Laboratory, Department of Computer Engineering, Alzahra University, Tehran, Iran ' Department of Computer Engineering, Alzahra University, Tehran, Iran
Abstract: Anomaly detection in social networks as a challenging task has gained great attention. Every unusual behavioural pattern in a social network can be spotted as an anomaly which provides useful information. In this paper, a new method is proposed to identify anomaly based on community detection (AD-C) for the social network graph. Our model is made up of weighting in pre-processing step and three principle processes, including community detection, auxiliary community detection and node filtering. AD-C method offers a flexible framework for anomaly detection, which can be employed in different stages of its related algorithms. The experiments are conducted on two social media datasets, including Facebook and Flickr datasets. Experimental results indicate more efficiency in comparison to other anomaly methods as baselines in terms of the F-score. Also, the results indicate that applying the proposed steps lead to increased accuracy of the community detection methods.
Keywords: anomaly; anomaly detection; social networks; community detection; social media mining; network structure; network mining; weighted graph; clustering; outlier detection.
International Journal of Electronic Business, 2020 Vol.15 No.3, pp.199 - 222
Received: 22 Aug 2018
Accepted: 02 Jan 2019
Published online: 13 Aug 2020 *