AutoCyclone: Automatic Mining of Cyclic Online Activities with Robust Tensor Factorization

Given a collection of seasonal time-series, how can we find regular (cyclic) patterns and outliers (i.e. rare events)? These two types of patterns are hidden and mixed in the time-varying activities. How can we robustly separate regular patterns and outliers, without requiring any prior information? We present CycloneM, a unifying model to capture both cyclic patterns and outliers, and CycloneFact, a novel algorithm which solves the above problem. We also present an automatic mining framework AutoCyclone, based on CycloneM and CycloneFact. Our method has the following properties; (a) effective: it captures important cyclic features such as trend and seasonality, and distinguishes regular patterns and rare events clearly; (b) robust and accurate: it detects the above features and patterns accurately against outliers; (c) fast: CycloneFact takes linear time in the data size and typically converges in a few iterations; (d) parameter free: our modeling framework frees the user from having to provide parameter values. Extensive experiments on 4 real datasets demonstrate the benefits of the proposed model and algorithm, in that the model can capture latent cyclic patterns, trends and rare events, and the algorithm outperforms the existing state-of-the-art approaches. CycloneFact was up to 5 times more accurate and 20 times faster than top competitors.