Abstract: The construction and development of the Web of data are affected seriously by the loss of links amongsemantic datasets fromheterogeneousdata sources. The construction of co-reference between instances in semantic datasets will be helpful for enriching the links among datasets, which will improve the reasoning and query across datasets. The property weights and the similarities of property values play an important role in the co-reference resolution based on similarity analysis. This paper proposed a new model for obtaining the weights of properties from the statistical information of datasets, and proved its rationality theoretically. Furthermore, the advantages of this model compared with traditional method were analyzed. Based on the new method proposed by this paper, a system for building co-reference among semantic datasets was implemented, and the performance was verified with some open datasets.

Key words: Linked data, Co-reference, Property weight