Anomaly in Online Social Network can be designated as an unusual or illegal activity of an individual. It can also be considered as an outlier or a surprising truth. Due to the emergence of social networking sites such as Facebook, Instagram, etc., the number of negative impacts of aggressive and bullying phenomena has increased exponentially. Anomaly detection is a problem of crucial importance which has attracted researchers since the 2000s. This problem is often carried out, thanks to deep learning, artificial intelligence and statistics. Several methods have been devoted to solving the problem of detecting abnormal behavior on social media, which are kept under three different types: structural methods which are based on the analysis of graphs of social networks, behavioral methods which are based on the extraction and analysis of user activities and hybrid methods which combine the two types of methods mentioned above. This survey reviews various methods of data mining for the detection of anomalies to provide a better assessment that can facilitate the understanding of this area.

可以将在线社交网络中的异常指定为个人的异常或非法活动。也可以将其视为离群值或令人惊讶的事实。由于诸如Facebook，Instagram等社交网站的出现，激进和欺凌现象的负面影响数量呈指数增长。自2000年代以来，异常检测是至关重要的问题，吸引了研究人员。由于深度学习，人工智能和统计，经常会出现此问题。致力于解决社交媒体异常行为问题的方法有几种，分别属于三种类型：基于社交网络图分析的结构化方法，基于对用户活动的提取和分析的行为方法以及结合了上述两种方法的混合方法。这项调查回顾了用于发现异常的各种数据挖掘方法，以提供更好的评​​估，从而可以促进对该领域的理解。