Detecting Malicious Social Bots Based on Clickstream Sequences

Abstract

The crucial part development in amount, pace, and range of user data (such Information created by users) web-based social networks has led to attempts to develop new techniques for acquiring and reviewing such massive data. For instance, social bots till now used to offer users higher-quality customer service while carrying out automated analytical tasks. Harmful social bots, such as fake news, have circulated false information, which has had an effect on the real world. Consequently, it is crucial to identify and get rid of risky social bots from online social networks.Based on their characteristics, the bulk of currently employed detection methods for risky social robots focus concerning the numeral aspects. Low analytical accuracy is caused by the ease with which social bots can mimic these characteristics. This article provides a word of art feature method based to the shifting likelihood of clickstream sequences and semi-supervised clustering for the detection of harmful social bots. This method considers both the transition probability of stream of click by the user analysis and the temporal component of activity. The detection accuracy for various types of malicious social bots based on transition probabilities of clickstreams increases by an average of 12.8%, according to the results of our tests on actual platforms for online social networks. This is in contrast to detection methods based on examination of user behavior by employing stats.