Smartphone continuous user identification via hybrid transfer learning: leveraging common semantic space and pre-trained models

Smartphone continuous user identification via hybrid transfer learning: leveraging common semantic space and pre-trained models

연구 분야: Strategies

학회: CCF Transactions on Pervasive Computing and Interaction

The widespread adoption of smartphones and their rapidly evolving functionalities have made these devices an integral part of everyday life for the general public. However, smartphones constantly face security threats due to storing personal information and confidential data. Traditional authentication methods are single-step processes and cannot continuously identify users after the initial authentication. In this paper, we investigate the sequence of mobile application usage to identify and profile individual users. The main challenge of different behavioural biometric methods is the change in user over time. This issue reduces the accuracy of the authentication system and consequently affects user security. To address this challenge, this paper introduces a hybrid method for continuous smartphone identification utilizing a transfer learning approach. Initially, a common semantic space is established in the first phase, enabling the proposed method to align the feature spaces of both the source and target data. This common semantic space ensures that the model remains unaffected by scenarios where users install, uninstall, or replace applications. Subsequently, in the second phase, the pre-trained model undergoes periodic updates to adjust to changes in user behaviour, ensuring its effectiveness in evolving user scenarios. Based on the experiments conducted, we have shown that the solution proposed in this paper continuously and implicitly identifies the user with reasonable accuracy and performance. The results indicate that this method surpasses alternative approaches.