Proposing a Useful Algorithm in order to Find Almost-Balanced Sub graphs of Social Network using Spectral Balance Index

As social communications become widespread, social networks are expanding day by day, and the number of members is increasing. In this regard, one of the most important issues on social networks is the prediction of the link or the friend s suggestion, which is usually done using similarities among users. In the meantime, clustering methods are very popular, but because of the high convergence velocity dimensions, clustering methods are usually low. In this research, using spectral clustering and diminishing dimensions, reducing the amount of information, reduces clustering time and reduces computational complexity and memory. In this regard, the spectroscopic clustering method, using a balanced index, determines the number of optimal clusters, and then performs clustering on the normal values of the normalized Laplace matrix. First, the clusters are divided into two parts and computed for each cluster of the harmonic distribution index. Each cluster whose index value for it is greater than 1 will be redistributed to two other clusters, and this will continue until the cluster has an index of less than 1. Finally, the similarity between the users within the cluster and between the clusters is calculated and the most similar people are introduced together. The best results for the Epinions, Google+ and Twitter data sets are 95.95, 86.44 and 95.45, respectively. The computational results of the proposed method and comparison with previous valid methods showed the superiority of the proposed approach.

Keywords:

Improved Algorithm Accuracy , Balanced Subcategories , Social Networks , Balanced Index.