Term-specific eigenvector-centrality in multi-relation networks

Abstract

Fuzzy matching and ranking are two information retrieval techniques widely used in web search. Their application to structured data, however, remains an open problem. This article investigates how eigenvector-centrality can be used for approximate matching in multi-relation graphs, that is, graphs where connections of many different types may exist. Based on an extension of the PageRank matrix, eigenvectors representing the distribution of a term after propagating term weights between related data items are computed. The result is an index which takes the document structure into account and can be used with standard document retrieval techniques. As the scheme takes the shape of an index transformation, all necessary calculations are performed during index time.