ContextAggregator: A heuristic-based approach for automated feature construction and selection

Our research goal is to work towards a personal contextaware assistance and retrieval of relevant resources to computer users during a certain work task. This paper presents a general-purpose, algorithmic approach for automated context aggregation by heuristic-based feature construction. Our implementation of the context reasoning layer combines lower-level context features to new aggregated higher-level context features. Our approach allows --in contrast to most other approaches --an automated feature combination to achieve a high prediction accuracy of the user's work task.