A Similarity Graph Matching Approach for Instance Disambiguation

摘要

Instance matching acts as a significant part of information integration in semantic web research. While ontology matching focuses on the schema level of data, instance matching deals with massive instances objects. Ambiguation is a common problem which may lead to error matching when different instances share the same names or descriptions. To cope with this problem structural approach is used by many matching systems for disambiguation. However, existing structural approach has a hidden problem named 'error propagation' which would affect the precision of matching result. In this paper, we investigate instance matching techniques and propose a new instance matching framework. It is based on a novel structural matching algorithm which calculates similarity separately on sub graphs. The structural information is fully taken advantage of to realize disambiguation and several indexing strategies are used to cut down the computing overhead. We have conducted experiments on instance matching benchmark and results show that our proposed matching approach is comparable to state-of-art systems. And experiment on real dataset has proved the validity of our approach in instance disambiguation.