A study on the network characteristics of two collaboration networks constructed from the ACM and DBLP digital libraries is presented. Different types of generic network models and several examples are reviewed and experimented on re-generating the collaboration networks. The results reveal that while these models can generate the power-law degree distribution sufficiently well, they are not able to capture the other two important dynamic metrics: average distance and clustering coefficient. While all current models result in small average distances, none shows the same tendency as the real networks do. Furthermore all models seem blind to generating large clustering coefficients. To remedy these shortcomings, we propose a new model with promising results. We get closer values for the dynamic measures while having the degree distribution still power-law by having link addition probabilities change over time, and link attachment happen within local neighborhood only or globally, as seen in the two collaboration networks.
All Science Journal Classification (ASJC) codes
- Social Sciences(all)
- Computer Science Applications
- Library and Information Sciences