Presenting Semi-Structured Text Retrieval Results

DEFINITION Presenting semi-structured text retrieval results refers to the fact that, in semi-structured text retrieval, results are not independent and a judgment on their relevance needs to take their presentation into account. For example, HTML/XML/SGML documents contain a range of nested sub-trees that are fully contained in their ancestor elements. As a result, semi-structured text retrieval should make explicit the assumptions on how the retrieval results are to be presented. Four of the main assumptions to be addressed are the following. First, the unit of retrieval assumption: is there a designated retrieval unit (such as the document or root node of the semi-structured document) or can every sub-tree be retrieved in principle? Second, the overlap assumption: may retrieval results contain text or content already part of other retrieval results (such as a full article and one of its individual paragraphs)? Third, the context assumption: can results from the same semi-structured document be interleaved with results from other semi-structured documents? Fourth, the display assumption: is a retrieval result (say a document sub-tree corresponding to a paragraph) presented as an autonomous unit of text, or as an entry-point within a semi-structured document?