An Efficient Flexible Semantic Distance Function

Quantifying similarity between two objects plays an important role in clustering and classification, etc. The quality of the similarity scores can be improved by considering the semantic information related with the features of objects. In this paper, we propose a semantic distance function, X-Dist, which not only utilize the semantic information to measure the difference between two objects and a solution of the transportation problem in linear optimization, but also is a metric which can make searching efficiently. The experimental results show that this distance function can be as well as the previously proposed similarity measures in nearest neighbor searching, discriminative power and computing speed.