On declarative modeling of structured pattern mining
Tias Guns, Sergey Paramonov, Benjamin Négrevergne · Lirias (KU Leuven) · 2016
Since the seminal work on frequent itemset mining, there has been considerable effort on mining more structured patterns such as sequences or graphs. Simultaneously, the field of constraint programming has been linked to the field of pattern mining resulting in a more general and declarative constraint-based itemset mining framework. A number of recent papers have logically proposed to extend the declarative approach to structured pattern mining problems. Because the formalism and the solving mechanisms are vastly different in specialised algorithm and declarative approaches, assessing the benefits and the drawbacks of each approach can be difficult. In this paper, we introduce a framework that formally defines the core components of itemset, sequence and graph mining tasks, and we use it to compare existing specialised algorithms to their declarative counterpart. This analysis allows us to draw clear connections between the two approaches and provide insights on how to overcome current limitations in declarative structured mining.