Multi-relational Clustering Based on Relational Distance

When clustering the tuples in the target table which is in a relational database, the prior task is to exactly and effectively calculate the relational distance between tuples. A lot of methods are used today, such as the relational distance measuring based on RIBL2. However, all these methods fail to consider the differences of similarity between the objects in both non-target table and target table, which stopped them from getting a high clustering accuracy. Using canonical correlation analysis in this paper and setting a weight for each table in the relational database, the weight indicated its role in the calculation of the distance among target tables. In addition, when calculating the distance between the two clusters to find the center of each cluster, turn the calculation of the distance between clusters into a distance between center points. Experiments show that this method ensures clustering efficiency and improves clustering accuracy.