学术合作是科研模式的第四纪元，学术合作网络正在成为网络科学、数据科学和计算社会科学领域新兴的研究热点。然而，学术合作网络具有大规模、动态性、异构性等特征，对其分析挖掘的高效性和准确性提出了挑战，许多问题亟待解决。本项目研究适用于大规模学术合作网络挖掘的关键技术，以图学习作为基本技术框架，构建一套学术合作网络分析挖掘技术和方法体系，解决学术合作网络大规模网络结构刻画、动态网络演化规律建模、合作价值评估等关键科学问题。为了在高动态多重网络环境下实现隐式关系挖掘，建立基于时变多层网络理论的网络结构刻画算法；研究大规模动态网络环境下学术团队中学者合作模式的建模方法，探究学术团队演化规律；针对学者属性多样性和合作网络大规模性，提出基于动态属性网络表征的消失链路预测方法；针对合作价值的结构和非结构性特点，建立基于互惠性的学术合作价值评估方法。预期成果将为学术大数据创新服务与应用提供理论和技术支撑。

As scientific collaboration becomes the forth age of research, academic collaboration networks have become an emerging area that is attracting more and more attention from network science, data science and computational social science. However, academic collaboration networks are typically dynamic, heterogeneous, and in large scale. These characteristics give rise to a large number of significant challenges in efficiently and correctly mining academic collaboration networks. Among others, large-scale network topology building, dynamic network modelling, and collaboration value evaluation are three important challenges that remain to be addressed. With focus on these issues, this project will study key technologies that are suitable for the mining of academic collaboration networks, and build a set of academic collaboration network mining technologies and methods which take advantage of graph learning as the basic technical framework. Specifically, in order to realize the implicit relation mining in dynamic multi-layer network environments, a network structure depiction algorithm based on the theory of time-varying multi-layer network will be established. Considering the large-scale dynamic network environment, this project will propose a scalable collaboration pattern model to explore the rules and dynamics of academic teams. Based on the diversity of scholars' attributes and the large-scale property of collaboration networks, a method for disappearing link prediction based on dynamic attribute network embedding will be devised. According to the structural and non-structural characteristics of collaboration value, this project will develop a reciprocity-based collaboration value assessment method. Expected results will help provide theoretical and technical support for innovative services and applications built upon scholarly big data.