Infinite Mixture Model of Markov Chains
This work addresses prediction and information extraction tasks for time series data with multiple patterns, such as user behavior analysis, but appears incremental as it extends existing hierarchical hidden Markov model approaches.
The authors tackled the problem of modeling categorical-valued time series with multiple underlying patterns (e.g., user behavior traces) by proposing a Bayesian nonparametric mixture model with an efficient inference scheme. The model excels at segmentation and prediction performance, successfully identifying generating patterns and enabling effective prediction of future observations.
We propose a Bayesian nonparametric mixture model for prediction- and information extraction tasks with an efficient inference scheme. It models categorical-valued time series that exhibit dynamics from multiple underlying patterns (e.g. user behavior traces). We simplify the idea of capturing these patterns by hierarchical hidden Markov models (HHMMs) - and extend the existing approaches by the additional representation of structural information. Our empirical results are based on both synthetic- and real world data. They indicate that the results are easily interpretable, and that the model excels at segmentation and prediction performance: it successfully identifies the generating patterns and can be used for effective prediction of future observations.