RMIT University at TREC 2008: Enterprise Track
Abstract:
The Royal Melbourne Institute of Technology RMIT in Australia participated in the 2008 Enterprise Track document search task. Their experiments investigated the use of local outdegree, and whether this can improve the ranking quality of a search result list. Unlike global outdegree, which counts the number of out-links of a page that point to any other pages in a collection, local outdegree only counts the out-links that point to pages contained in a search result list. Intuitively, restricting the outdegree to the result set of a query transforms this source of evidence from something general into a topically focused source of information, and may help to reduce the problem of topic shift. For their experiments, they used the Zettair search engine to index and search the CSIRO collection used for the 2008 Enterprise Track. This collection is a crawl of the public-facing web of the Australian Commonwealth Scientific and Industrial Research Organization CSIRO in 2007 Bailey et al., 2007. Document weights were calculated using the Okapi BM25 similarity function Sparck Jones et al., 2000, with query words being terms from the query fields of the track topics. During indexing and search, words were stemmed and stopped.