LAUSR

Scientific talents are the cornerstone of science and technology development. The current method to find out the scientific talent is almost based on the scientists’ achievement, less considering the interrelationships hidden in the objects. In this paper, we propose a centrality-adjusted graph attention network to discover active scientific talents. This graph network can find out the internal correlation among scientific papers by using a predicate-based attention mechanism and in-degree adjustment strategy on the node. We adopt two training ways for experimentation and validation of the proposed model: in-domain and out-of-domain estimation. The experiment result shows that our model performs better than existing models on the Database of Chinese Science Citation on Normalized Discounted Cumulative Gain (NDCG) metrics.