ETH Zürich at INEX: Flexible Information Retrieval from XML with PowerDB-XML

When searching for relevant information in XML documents, users want to exploit the document structure when posing their queries. Therefore, queries over XML documents dynamically restrict the context of interest to arbitrary combinations of XML element types. State-of-the-art information retrieval (IR) however derives statistics such as document frequencies for the collection as a whole. With contexts of interest defined dynamically by user queries, this may lead to inconsistent rankings with XML documents that have heterogeneous content from different domains. To guarantee consistent retrieval, our XML engine PowerDBXML derives the appropriate IR statistics that consistently reflect the scope of interest defined by the user query onthe-fly, i.e., at query runtime. To compute the dynamic IR statistics efficiently, our implementation relies on underlying basic indexes and statistics data. This paper reports on our experiences from participating in INEX, the INitiative for the Evaluation of XML retrieval.