Clustering is a fundamental task that has been utilized in many scientific fields, especially in machine learning and data mining. In clustering, dissimilarity measures play a key role in formulating clusters. For handling categorical values, the simple matching method is usually used for quantifying their dissimilarity. However, this method cannot capture the hidden semantic information that can be inferred from relationships among categories. In this paper, we propose a new clustering framework for categorical data that is capable of integrating not only the distributions of categories but also their mutual relationship information into the pattern proximity evaluation process of the clustering task. The effectiveness of the proposed clustering algorithm is proven by a comparative study conducted on existing clustering methods for categorical data.
展开▼
