Community Analysis of Fashion Coordination Using a Distance of Categorical Data Sets

While various clustering methods have been proposed for categorical data, few studies exist on the clustering of the categorical variables that are represented as item sets (groups), e.g. fashion coordination data. In order to fill this gap, this study focuses on the patterns of similarities that are found between coordinated fashion items, so as to define a new set of score value rows for calculating similarity indexes from the patterns. These similarity indexes are then inverted into distances, based on which the clustering method proposed in this study (hereinafter ―the Proposed Method‖) is achieved. Then, it was do two evaluated experiment using about 150,000 coordination data obtained from the wear's web site. Firstly, a silhouette analysis confirms that the Proposed Method offers better cluster partitioning in comparison to previous studies. And secondly, an efficacy evaluation reveals seasonal patterns in fashion coordination, as well as trends for brands and colors of individual fashion items.