摘要： 基于图的随机游走算法在预测论文影响力时，仅利用学术网络的全局结构信息而未考虑局部结构信息，对预测准确率造成影响。针对该问题，提出一种基于异构学术网络表示学习和多变量随机游走的论文影响力预测算法。通过构建异构学术网络表示模型，将网络中的论文、作者和期刊/会议等不同类型的节点表征到同一个低维向量空间中，同时保留网络的局部结构信息，将节点的向量相似度应用于多变量随机游走方法，实现对论文影响力的准确预测。在AMiner网站公开数据集上的实验结果表明，相比于PageRank、FutureRank等算法，该算法的预测准确性较高。

关键词: 网络表示学习, 影响力预测, 异构学术网络, 多变量随机游走, 局部结构信息

Abstract: The graph-based random walk algorithm for paper influence prediction exploits only global structural information of academic network,and local structural information is usually ignored,which influence the prediction accuracy.To address the problem,this paper proposes a paper influence prediction algorithm based on heterogeneous academic network representation learning and multivariate random walk.By constructing a heterogeneous academic network representation model,different kinds of nodes of paper,authors,and journals/conferences in the network are represented into the same low-dimensional vector space,and the local structural information of network is kept.Similarity between vectors is applied to multivariate random walk to implement the accurate paper influence prediction.Experimental results on public datasets of AMiner Website show that the proposed method is more accurate in prediction than PageRank,FutureRank and other algorithms.

Key words: network representation learning, influence prediction, heterogeneous academic network, multivariate random walk, local structural information

中图分类号: 

• TP311