网络表示学习在学者科研合作预测中的应用研究

Application of Network Representation Learning in the Prediction of Scholar Academic Cooperation

在大数据环境下,科研合作是提高科研水平、促进科研产出的重要途径。如何在浩如烟海的学者、机构、领域信息中准确地找到与自身研究方向相近的合作对象是近年来科研合作预测的研究重点。本文通过科学学领域科学文献的记录数据,构建作者-作者、机构-机构、作者-机构、作者-关键词、机构-关键词的共现网络,接着通过网络表示方法学习作者、机构、关键词在所处网络中的语境信息,将信息实体表示成相同空间的低维稠密向量,最后根据表示向量的相似度计算实现合作对象、合作领域挖掘。通过网络表示学习方法能实现多种异质信息融合,定量计算各信息实体间的关联强度,可以很好地捕捉科研网络中学者-学者、学者-机构、学者-关键词的关系,准确地为学者挖掘潜在合作者、合作机构和关键词。

In the context of Big Data, scientific cooperation has become an important means to improve the level of scientific research and output. A research focus in recent years includes accurately identifying the cooperation objects that are suitable for scholars, institutions, and fields among the vast numbers of these entities. This study constructs a co-occurrence network of author-author, institution-institution, author-institution, author-keyword, and institution-keyword through the recorded data of scientific literature in the field of science of science. The network representation method is used to learn the context information of authors, institutions, and keywords in a network, and the information entity is represented as a lowdimensional dense vector of the same space. Finally, the mining of the cooperation object is achieved based on the similarity calculation of representation vector. The network representation learning method can realize a variety of heterogeneous information fusion, quantitatively calculate the correlation strength between each information entity, capture the relationship between scholars-scholars, scholars-institutions, and scholars-keywords in the research network, accurately explore potential collaborators, and partner institutions and keywords for scholars.