Uneven Hybrid Clickstream Generation Based on Multi-Layer Perceptron

Online social networking is prevalent around the world, meanwhile, social bots damage users’ experience by stealing user privacy, sending spam, spreading malicious links, launching DDoS attacks, et al. In this paper, we use social bots to automatically manipulate social accounts based on generated clickstreams to demonstrate the vulnerability of social networking sites and the capability of social bots. Clickstreams are the sequential click events of users or bots browsing a website. Existing research on clickstream generation assumes that the intervals between all click events are equal, which makes it difficult to fully capture the relationship between uneven intervals. To this end, we propose an uneven hybrid clickstream generation algorithm based on multi-layer perceptron (UHCG-MLP), which can generate click events and intervals simultaneously by dealing with uneven intervals. UHCG-MLP firstly transforms each clickstream in the training set into a sequence of discrete elements separated by coded interval, and uses multi-layer perceptron to mine user behavior patterns from these sequences. Then it generates hybrid clickstreams according to the prediction of the trained model. Experiments on the Twitter platform using the clickstreams generated by the proposed model and the other five benchmark models show that UHCG-MLP achieves the best execution effect.