CMAP: A flexible and efficient framework for constraint-based mining of activity patterns

Human and machine activities have been recorded in many applications. Recurrent patterns discovered from such activities can provide invaluable insights and enable effective actions in these applications. This paper introduces a flexible framework for discovering activity patterns from multi-relational data. The data comes from multiple heterogeneous sources. The patterns describe the recurring relationship among different types of records in the dataset. The complexity of data and generality of patterns pose great challenges on developing efficient mining algorithms. The major contributions of this paper include the formalization of different types of constraints, and a generic mining algorithm which exploits constraints to improve the mining efficiency, as demonstrated by our experiments.