From focused elements to snippets

Abstract

Information Retrieval is a field of computing which traditionally deals with searching a large collection of documents and retrieving documents based on their similarity to the query. INEX [10] provides a platform (e.g., document collection, queries and uniform evaluation metrics) for the development and evaluation of retrieval algorithms for XML documents. The focus of INEX is to reduce the granularity of search results from the entire document to the element level. In 2011, INEX introduced a new track, called the Snippet Retrieval Track. In 2012, INEX improved this track to make the task of assessment easier. Its goal is to determine how best to generate informative snippets for search results. Such snippets should provide sufficient information to allow the user to determine the relevance of each document without viewing the document itself. The Snippet Retrieval track uses the 50.7GB INEX Wikipedia collection of about 2.7 million articles. We use the Smart [15] experimental retrieval system, based on the Vector Space Model [16], for indexing and retrieval. This thesis describes the approaches taken by UMD to generate runs to participate in the INEX 2011 and 2012 Snippet Retrieval track. We use our method of dynamic element retrieval [7] to generate the element vectors of the XML document tree at run time, thus producing a rank-ordered list of elements from each highly correlated document. These elements are further processed using our methods to generate snippets. The methods used, experimental results, and conclusions are described herein.