Relationships between network's structure features and link prediction algorithms

The first essential thing on the application of link prediction is how to choose the right prediction algorithm. We experimented with five virtual networks and analyzed the relationship between those networks' structure features and six typical link prediction algorithms by the experiment's data in this paper. We found that if the network's assortativity coefficient is positive and the clustering coefficient is greater than the threshold which is about 0.1, the algorithms based on local information would get higher prediction results, otherwise the based global information would be better. And the clustering coefficient and efficiency are proportional to the accuracy of algorithms based local information and are reverse proportional to the algorithms based global information. These conclusions provide the quantitative basis for selecting the right algorithms in link prediction application.