Content-oriented XML retrieval with HyRex

The eXtensible Markup Language (XML)1 is the emerging standard for representing knowledge in almost arbitrary applications. At least almost every kind of knowledge can be represented in XML. The major purpose of XML markup is the explicit representation of the logical structure of a document. From an information retrieval (IR) point of view, users should benefit from the structural information inherent in XML documents. The XML Information Retrieval Query Language (XIRQL) [Fuhr & Großjohann 01, Fuhr & Großjohann 02] has been developed to serve this purpose. XIRQL extends the XPath [Clark & DeRose 99] part of the (proposed standard) query language XQuery [Chamberlin et al. 01] by features important in IR style applications. For instance, IR research has shown that document term weighting as well as query term weighting are crucial concepts for effective information retrieval. XIRQL allows for term weighting with regard to the components of the documents’ logical structure. This is used for implementing the retrieval paradigm suggested by the FERMI multimedia model for IR [Chiaramella et al. 96]: Instead of treating documents as atomic units, we aim at retrieving those document components (elements) which answer a given information need in the most specific way. This strategy is used to process the content-only (CO) topics provided within the INitiative for the Evaluation of XML retrieval (INEX)2, where no structural conditions are used within the queries. Given the logical structure inherent to XML documents, users want to pose queries not only on content but also on the structure of the documents. The INEX content-and-structure (CAS) topics reflect that. As an extension of XPath, the XIRQL query language is capable of processing these queries. The Hyper-media Retrieval Engine for XML (HyREX)3 [Abolhassani et al. 02] provides an implementation of the XIRQL query language. In the following we describe its implementation with regard to processing the INEX CO and CAS topics. In Section 2 we show how ranking of most specific document components is done in HyREX, thus serving for processing the content-only topics. Section 3 details the algorithms used to produce such a ranking of document components while Section 4 displays the evaluation results of our approach. Section 5 shows how XIRQL concepts are used in order to process the CAS topics. In addition we give a brief overview on the concepts of data types and vague predicates which can lead to high precision searches, in combination with structural retrieval. A conclusion and an outlook on further research is given in Section 6.