Prediction and clustering of user relationship in social network

Prediction and clustering are two fundamental and important problems in the analysis of social network. Most existing work are based on unsigned network which regards all the relationships as nonnegative proximity. However, employing these work directly on signed network is not obvious. In recent years, some approaches have been proposed for prediction and clustering of signed network. Based on those work, this paper presented a uniform framework to address both unsigned and signed network and predict relationships through mining in the latent feature space. In this paper, we first review the classical link prediction model, then propose our enhanced model which introduced embeddedness as parameters of objective loss function and the experiments yield higher accuracy. We then further employ the spectral analysis for clustering the signed network after link prediction is done. A serial of experiments which covered both functionality and performance was designed to test out model and the results show that out model improved both accuracy and performance.