利用链路预测揭示合著网络演化机制

Uncovering the Mechanism of Co-Authorship Network Evolution by Link Prediction

【目的/意义】掌握合著网络的最佳演化机制及其演变能够更好的进行合著关系预测和推荐,进而为研究团队的人员选择和搭配提供建议和参考。【方法/过程】以共同邻居、到达路径、优先连接和随机游走共4类16种相关性指标表示合著网络演化机制,并运用链路预测的理论和方法系统全面的定量比较不同演化机制的优劣和时序变化,揭示合著网络的最佳演化机制及其变化并解析其成因。【结果/结论】在图书情报领域的实验证实:描述合著网络演化机制的最佳指标为AA(Adamic-Adar);不同时间段的相关性指标的预测准确率具有一定差异但总体趋势保持一致,并且最佳指标所属类别并未改变,表明合著网络演化机制具有较强的稳定性;对多种类别的合著网络演化机制成因及其改进方向进行了分析。

[Purpose/significance]Understanding the optimum mechanism of co-authorship network evolution is helpful to co-authorship prediction and recommendation, which could give advice on researchers＇ selection for team building. [Meth- od/process]This paper applies 16 relatedness indicators that are further classified into four categories （i.e., common neigh- bor based, path based, preferential attachment based and random walk based indicators）, which are evaluated by R-Preci- sion by link prediction. This could compare mechanisms and their temporal variation quantitatively and uncover the best mechanism and its causing reasons. [Result/conclusion]The result in LIS （Library and Information Science） shows that the best indicator describing the mechanism of co-authorship network evolution is AA （Adamic-Adar）, the temporal variation of all indicators is consistent most of the times and the possible improvements of mechanisms is analyzed.