I. Introduction
In edge classification of networks, labels are associated with a subset of the edges and it is desired to classify the unlabeled edges. This problem has been studied recently in the context of binary prediction in signed networks [2], and in the generic multi-class setting [1]. The problem of edge classification of social networks with content arises in a number of different scenarios. In counter-terrorism applications, it may be desirable to monitor the communications between entities in order to detect suspicious communications. In such cases, both the content and the structure of the communication can play a role in the classification process. In many social networking applications, numerous forms of communications between entities (such as wall posts in Facebook) are public in nature. The content and local structure of such posts can be used in order to classify its relevance to a particular product.