Research on a Model of Context-driven Similarities Computation between Data Objects

Abstract

A practical problem that has been more and more acknowledged is how to measure similarities of data objects. With high interconnectivity to heterogeneous information sources, the primary issue in many applications such as retrieval, interoperability, etc, is focused on determining which data is relevant. Many methods for measuring have been investigated. We propose a complementary approach, based on context information to enhance "understanding" of object-to-object associations, which measures similarities of data objects that are an abstraction from real world. In our method, confidence level on similarities can be greatly improved by "observing" data objects from multi-perspectives. It is also demonstrated by our experiments that this approach is effective as compared to other computation approaches.