Multi-Task Learning Framework for Detecting Hashtag Hijack Attack in Mobile Social Networks
Zheng Qu, Chen Lyu, Chi‐Hung Chi · 2022 IEEE 19th International Conference on Mobile Ad Hoc and Smart Systems (MASS) · 2022
As a refinement of a microblog, hashtag supports the services of search and recommendation for mobile social networks. However, it's vulnerable to hashtag hijack attack. To solve this problem, previous works either only direct at some specific hijack or extract features to match the microblog's content to the hashtag. However, these solutions cannot distinguish between non-malicious use of hashtags by normal users and real hashtag hijack by attackers. By analyzing the behavioral intentions of users, we propose a multi-task learning framework to address this issue, which includes three relevant learning tasks. First, we introduce the task of text matching by combining a pre-trained language model with LSTM, to improve the accuracy and generalization of our model. Second, we design the task of informative text classification to classify informative and non-informative microblogs, which excludes the effect of the non-malicious use of hashtags. Third, we propose the task of hashtag hijack identification with the cooperation of the other two tasks. Hence, our hard-parameter sharing multi-task learning framework is constructed by utilizing the internal connections among three tasks. Finally, we design a dynamic weighting method to optimize multi-task loss function for our multi-task learning framework. Extensive experiments have demonstrated that our model can significantly improve the detection accuracy of hashtag hijack attack.