The Comparison of Top Leaders Algorithm and Other Algorithms

In this project I explored the Top Leaders algorithm [1], and compared it with several other community discovery algorithms. Community discovery is an important and interesting research field in the analytics of social network. By detecting communities in a social network, companies can adopt different marketing strategies and recommend different products for people in different communities, or provide personalized service for them. Thus, companies can make more profit, so community discovery is an interesting and practical research subject. The Top Leaders algorithm is inspired by K-means algorithm. According to the knowledge we learned in the class, the essence of K-means clustering is a hard EM for Gaussian mixture model. So, in essence Top Leader algorithm is a kind of unsupervised learning algorithm. Through this project, I learned Top Leaders algorithm is an effective method for community discovery, but its performance is usually inferior to that of spectral clustering algorithm, Girvan and Newman’s divisive algorithm and Newman’s greedy optimization of modularity algorithm. Moreover, the initial selection of leaders has an important influence on the final result. These will be shown and discussed in Section 4 and Section 5.