Automating semantic matching of attributes for the purpose of information integration is challenging, and the dynamics of the Web further exacerbate this problem. Believing that many facets of metadata can contribute to a resolution, we present a framework for multifaceted exploitation of metadata in which we gather information about potential matches from various facets of metadata and combine this information to generate and place confidence values on potential attribute matches. To make the framework apply in the highly dynamic Web environment, we base our process on machine learning when sufficient applicable data is available and base it otherwise on empirically observed rules. Experiments we have conducted are encouraging, showing that when the combination of facets converges as expected, the results are highly reliable.
semantic attribute matching; information integration; exploitation of metadata