A multi-channel multi-tower GNN model for job transfer prediction based on academic social network

Abstract

The study of scholar job transfer is beneficial to both individual career success and talent recruitment of institutions. For such a complicated problem, the diversity and objectivity of the obtained information are essentially significant. However, previous works focus on the general job-hopping problems and utilize the information from the recruitment or social platform, which is somehow limited by data privacy. In this paper, we define the scholar job transfer prediction task by introducing more diversified information (e.g., career sequence, collaboration graph), which is obtained from their publications, for more comprehensive modeling without privacy data. Moreover, we design a Multi-channel Multi-tower Enhanced Framework (MMEF) to integrate the heterogeneous inputs in a complementary manner, which can capture the temporal pattern from career trajectories, leverage the academic collaboration information considering the influence from co-authors, and deal with extra descriptions and estimate the relevance scores between the scholars and institutions. Extensive experiments on two real-world datasets demonstrate the superiority of the proposed framework, which outperforms the state-of-the-art approaches by about 20% overall for making better use of the potential patterns in heterogeneous data. More intensive studies explore how and the degree to different collaborators impact scholar job transfer.