Matching Similarity for Keyword-Based Clustering

Semantic clustering of objects such as documents, web sites and movies based on their keywords is a challenging problem. This requires a similarity measure between two sets of keywords. We present a new measure based on matching the words of two groups assuming that a similarity measure between two individual words is available. The proposed matching similarity measure avoids the problems of traditional measures including minimum, maximum and average similarities. We demonstrate that it provides better clustering than other measures in location-based service application.