FDHFUI: Fusing Deep Representation and Hand-Crafted Features for User Identification

User identification across multiple social networks is an important task for cross-social network recommendation, network security, and other related fields. Usernames belonging to the same individual are rich in information that is valuable for user identification. However, existing user identification methods based on username similarity rely on hand-crafted features that cannot mine semantic features. Although this problem can be overcome by adopting deep learning, end-to-end models have poor interpretability. To address these limitations, we propose a hybrid approach that fuses deep semantic and hand-crafted features to identify users. We mined semantic features from usernames using a pretrained multilingual Bidirectional Encoder Representations from Transformers algorithm to improve similarity discrimination. Subsequently, we constructed 15 hand-crafted features for usernames from the Chinese user community to enhance the interpretability of the deep model. We also propose a user identification model that fuses the semantic and hand-crafted features for usernames. We evaluated our proposed approach using four real-world datasets from Chinese social networks, and our method was more accurate than the state-of-the-art methods. Moreover, our method was effective even with limited positive sample data.