Toward Learning Similarity Measures for Uncertain Features

• Activity Number: 570
• Type: Contributed
• Date/Time: Thursday, August 6, 2009 : 8:30 AM to 10:20 AM
• Sponsor: Business and Economic Statistics Section
• Author(s): Ming Xie*+ and Bin Zhang and Li Xia and Jin Yan Shao and Wenjun Yin and Jin Dong
• Companies: IBM China Research Laboratory and IBM China Research Laboratory and IBM China Research Laboratory and IBM China Research Laboratory and IBM and IBM China Research Laboratory
• Keywords: similarity measure ; spatial data ; uncertain feature

Abstract

Similarity measure plays a key role in many AI areas, and similarity measure of spatial data is specially challenging. Firstly, given geographical features in a spatial cluster do not always exist, i.e., geographic data has uncertain features. Secondly, the distance metrics of geographical features should achieve some predefined goals. This paper proposes a flexible similarity measure which considers the existence of feature as a random variable. Then a coefficient is assigned for each variable for the calculation of similar measure. To determine these coefficients, we first propose a maximum likelihood clustering approach and prove some properties. And then a classification process is employed based on the likelihood. Through these approaches we get the more suitable similarity measure between spatial clusters. Finally the experiments demonstrate the advantage of this approach.