Email Engagement Segmentation Using Bipartite Graph Co-clustering

In the industry of email marketing, it is important to send content relevant to the recipient. If the recipients are uninterested they may ignore the email or worse report it as spam. Such actions compromise the ability of the senders to deliver emails to the inboxes of other recipients and permanently harm their relationship with the uninterested recipients. Targeting highly engaged recipients with appropriate content helps senders maximize the value of their email marketing efforts. Traditional research on email recipient engagement has focused on identifying topic groups through text mining and sentiment analysis. In this paper, we investigate recipient engagement via co-clustering methods on graphs of sender-recipient engagement built from email opens and clicks on URL links. We study a real engagement graph from marketing emails sent over SendGrid in May of 2015. We find that the sender-recipient engagement graph has a self-similar or recurring dominant co-cluster structure and the minor co-clusters are paired as engagement contrasts. Additionally, we find that these co-clusters persist across subsequent months implying that this approach may be useful for predicting recipient engagement behavior within clusters.