Abstract
Nowadays, people pay more and more attention to privacy protection with the continuous occurrence of data breaches. In location-based services, dummy trajectory generation is a popular location privacy protection method. However, this method is used by some malicious users for benefits, which results in economic losses and the waste of resources of the locationbased services provider. For this problem, dummy trajectory identification has been proposed by researchers. Nevertheless, with the continuous development of dummy trajectory generation algorithms, the existing dummy trajectory identification methods are unsuitable. In this paper, we propose a hybrid neural network framework for dummy trajectory identification, called trajectory similarity hybrid networks (TSHN). The main idea of TSHN is to identify whether a target trajectory is a virtual trajectory according to the similarity score between historical trajectories and the target trajectory. For each historical trajectory of the user, the mobility and individual features are extracted to train TSHN. The trajectory similarity score generated by the TSHN is used to identify dummy trajectories. The experimental results show that our proposed TSHN can identify the dummy trajectory with an accuracy of 0.97, which significantly outperforms the existing dummy trajectory identification methods.