A Two Step Ranking Solution for Twitter User Engagement

In this paper, we describe our solution to the RecSys2014 challenge and results on the test set. We briefly describe some of the challenges, then describe the methodology which starts with feature extraction and construction using the provided tweet data, in combination with IMDB as an external source. Feature construction also involved computing similarity values in a latent factor space to deal with the sparsity and lack of semantics of text-based and other nominal features. We also describe our machine learning models which consist of several stages, including a classifier, followed by a Learning to Rank (LTR) model, with a repairing mechanism to further correct minority class (non-zero engagement) predictions that are close to the boundary. Finally, we draw conclusions in the form of lessons learned and future work toward improving our results.