A layered architecture for probabilistic complex pattern recognition to detect user preferences

The recognition of complex patterns is nowadays one of the most challenging tasks in machine learning, and it promises to be of great benefit for many applications, e.g. by allowing advanced human computer interaction to access the user’s situative context. This work examines a layered architecture that operates on different temporal granularities to infer complex patterns of user preferences. Classical hidden Markov models (HMM), conditioned HMM (CHMM) and fuzzy CHMM (FCHMM) are compared to find the best configuration in the lower architecture layers. In the uppermost layer, a Markov logic network (MLN) is applied to infer the user preference in a probabilistic rule-based manner. For each layer a comprehensive evaluation is given. We provide empirical evidence showing that the layered architecture using FCHMM and MLN is well-suited to recognize patterns on different layers.