In this paper, an intensive study on online social networks is studied. Through the presented mothod, the relationships between entities can be analyzed to be positive or negative. The positive relationship indicates trust or friendship and the negative relationship represents opposition or antagonism. We investigate some basic characteristics of signed networks and make certain extensions to particular features. A modified version of the PageRank algorithm is proposed, which is applicable to signed networks. Based on the creative features, an edge sign predictor using supervised machine learning algorithms is also established. The experimental results show that our model can significantly improve the prediction accuracy and decrease the false positive rate. 