Online Ranking Prediction in Non-stationary Environments

Recommender systems have to serve in online environments which can be highly non-stationary.1. Traditional recommender algorithmsmay periodically rebuild their models, but they cannot adjust to quick changes in trends caused by timely information. In our experiments, we observe that even a simple, but online trained recommender model can perform significantly better than its batch version. We investigate online learning based recommender algorithms that can efficiently handle non-stationary data sets. We evaluate our models over seven publicly available data sets. Our experiments are available as an open source project2.