Parallel News-Article Traffic Forecasting with ADMM

Predicting the traffic of an article, as measured by page views, is of great importance to content providers. Articles with increased traffic can improve advertising revenue and expand a provider’s user base. We propose a broadly applicable methodology incorporating meta-data and joint forecasting across articles, that involves solving a large optimization problem through the Alternating Directions Method of Multipliers (ADMM). We implement our solution using Spark, and evaluate it over a large corpus of articles and forecasting models. Our results demonstrate that our featurebased forecasting is both scalable as well as highly accurate, significantly improving forecasting predictions compared to traditional forecasting models.