关键词: 符号聚类, 区间数据, 相互距离, 相似性传播, K-均值

Abstract: Clustering for symbolic data is an important extension of conventional clustering, and interval representation for symbolic data is often used. The symmetrical measures in conventional clustering algorithms are sometimes not fit to interval data and the initialization is another severe problem that can affect the clustering algorithms. One metric called mutual distances for interval data was proposed; based on the metric, a new clustering method named affinity propagation clustering that could solve the problem initialization was used. Then, affinity propagation clustering for symbolic interval data based on mutual distance was given. Theoretical explanation and experiments indicate that the proposed algorithm outperforms K-means based on Euclidean distances for the interval symbolic data.

Key words: clustering of symbol, interval data, mutual distance, affinity propagation, K-means