| 摘 要: 针对任务导向型科研团队推荐中缺乏直接标签信息的问题,构建基于混合网络扩散的科研团队推荐方法。基于SciSciNet数据集构建作者-论文二部图,以共享参考文献确定种子论文,在网络上实施物质扩散、热传导及多轮迭代扩散,并通过线性混合机制融合传播结果。采用滚动时间窗进行评估。结果表明,参数约为0.8时,模型在排序准确性、新颖性和多样性之间取得较优平衡,可为面向主题任务的科研团队推荐提供数据支持。 |
| 关键词: 科研团队推荐 作者-论文二部图 物质扩散 热传导 线性混合 |
|
中图分类号: TP391
文献标识码:
|
|
| Research Team Recommendation Based on Hybrid Network Diffusion |
|
GU Jiarui1,2
|
1.School of Computer Science, Nanjing University of Information Science and Technology, Nanjing, 210044;2.China
|
| Abstract: A hybrid network diffusion-based method is developed to address the lack of direct label information in task-oriented research team recommendation. An author-paper bipartite graph is constructed based on the SciSciNet data set, and seed papers are identified through shared references. Mass diffusion, heat conduction, and multi-round iterative diffusion are performed on the network, and their propagation results are integrated by a linear hybrid mechanism. A rolling time-window strategy is adopted for evaluation. Experimental results show that the model achieves a favorable balance among ranking accuracy, novelty, and diversity when the parameter α is around 0.8, providing data-driven support for topic-oriented research team recommendation. |
| Keywords: research team recommendation author-paper bipartite graph mass diffusion heat conduction linear hybridization |
