Learnable Weighting of Intra-Attribute Distances for Categorical Data Clustering with Nominal and Ordinal Attributes

This paper addresses a common issue in categorical-data clustering: nominal and ordinal attributes are often handled identically, causing ordinal order information to be lost. It proposes a unified intra-attribute distance metric that preserves ordering among ordinal values while modeling dependencies across nominal and ordinal attributes. The associated clustering algorithm jointly learns distance weights and object partitions instead of optimizing them in separate stages, aiming to avoid suboptimal solutions. The abstract reports improved performance over existing methods, but provides no algorithm name, benchmark details, metrics, or numerical gains.