An Account Matching Method Based on Hyper Graph
Zhiwei Tang,Xuemeng Zhai,Gaolei Fei,Junyi Gao,Jianwei Ding,Keren Wang,Qiang Li,Guangmin Hu
DOI: https://doi.org/10.1007/978-981-97-5028-3_22
2024-01-01
Abstract:Character attribute recognition refers to the process of extracting character features and applying machine learning, deep learning and other technical means to recognize character attributes such as position, work unit, field of study, etc., by using data such as text and social relations in multi-source networks, which plays an important role in the fields of character profiling and user recommendation. In this paper, for the problems of strong sparsity of single-source networks in the Internet and insufficient ability to characterize character attributes in text, we propose a hyper graph-based account matching method from constructing multi-source network data associations, which achieves the characterization learning of multiple social network data in the same embedding space, avoids the noise introduced by the existing graph characterization methods based on the topology, and then utilizes the user profile similarity and node proximity of cross-social networks to identify character attributes. profile similarity and node proximity information across social networks to train an unsupervised matching model, which realizes the integration of multi-source network character attributes by matching account information from multiple social networks. Experiments show that this method greatly improves the accuracy of account matching and effectively establishes the connection between multi-source social network data.