What’s in a Link? From Document Importance to Topical Relevance

Abstract

Web information retrieval is best known for its use of the Web’s link structure as a source of evidence. Global link evidence is by nature query-independent, and is therefore no direct indicator of the topical relevance of a document for a given search request. As a result, link information is usually considered to be useful to identify the ‘importance’ of documents. Local link evidence, in contrast, is query-dependent and could in principle be related to the topical relevance. We analyse the link evidence in Wikipedia using a large set of ad hoc retrieval topics and relevance judgements to investigate the relation between link evidence and topical relevance.