The Open University at TREC 2006 Enterprise Track Expert Search Task

Abstract

The Multimedia and Information Systems group at the
\nKnowledge Media Institute of the Open University participated
\nin the Expert Search and Document Search tasks
\nof the Enterprise Track in TREC 2007. In both the document
\nand expert search tasks, we have studied the effect of
\nanchor texts in addition to document contents, document
\nauthority, url length, query expansion, and relevance
\nfeedback in improving search effectiveness. In the expert
\nsearch task, we have continued using a two-stage language
\nmodel consisting of a document relevance and cooccurrence
\nmodels. The document relevance model is
\nequivalent to our approach in the document search task.
\nWe have used our innovative multiple-window-based cooccurrence
\napproach. The assumption is that there are
\nmultiple levels of associations between an expert and
\nhis/her expertise. Our experimental results show that the
\nintroduction of additional features in addition to document
\ncontents has improved the retrieval effectiveness.