Clustering using table constraint satisfaction methods

The research focuses on developing cluster analysis methods, specifically clustering methods with partial teacher involvement, where background knowledge from the subject area is used when assigning objects to classes. The traditional approach to this problem involves modifying existing clustering methods, most of which are local search methods. The article proposes a systematic approach to searching for optimal partitions within the constraint programming paradigm. The originality of this research lies in solving the clustering problem as a constraint satisfaction problem, utilizing specialized table constraints, known as D-type smart tables, to model basic and additional conditions. Table constraint reduction rules are employed to organize logical inference procedures on D-type smart tables. The advantages of this approach are discussed, demonstrating how analyzing one of the optimal solutions can help identify objects on the boundary of clusters and those belonging to the same cluster for any optimal partition.