XQuery Processing with Relevance Ranking

Abstract

We are presenting a coherent framework for XQuery processing that incorporates IR-style approximate matching and allows the ordering of results by their relevance score. Our relevance ranking algorithm is based on both stem matching and term proximity. Our XQuery processor is stream-based, consisting of iterators connected into pipelines. In our framework, all values produced by XQuery expressions are assigned scores,and these scores propagate and are combined when piped through the iterators. The most important feature of our evaluation engine is the use of structural and content-based inverse indexes that deliver data in document order and facilitate the use of efficient merge joins to evaluate path expressions and search predicates. We present the rules for the translation from a large part of XQuery to iterator pipeline. Our modular approach of building pipelines to evaluate XQuery scales up to any query complexity because the pipes can be connected in the same way complex queries are formed from simpler ones.