Abstract:

In the research of discipline collaboration network, different schemes for extracting networks will be adopted according to different research purposes. The changes of experimental network structure will affect the ranking robustness of link prediction results. This paper selects Library and Information Science from CSSCI database, and constructs the collaboration networks. Then, it builds an experimental environment of network structure perturbation according to different network extraction schemes. By comparing the link prediction results and their ranking, the overlapping degree on the same scale of different link predictors, it analyzes the ranking robustness in the environment of network structure perturbation. The study finds that the perturbation of network structure has the least influence on the prediction results of the Katz index, and has the greatest impact on the prediction results of the Rooted PageRank index. It compares the link prediction effect of different link predictors in the environment of network structure perturbation. The study finds that AA, Katz and SimRank are relatively stable in terms of prediction effects, and they are relatively less affected by network structure perturbation.

Key words: Collaboration network, Link prediction, Network structure, Robustness