Experiments in the Probabilistic Retrieval of Full Text Documents.
William S. Cooper, Aitao Chen, Fredric C. Gey · 1994
he experiments described here constitute a confi tinuation of a research program whose object is to nd probabilistically sound, yet simple and powerful, . T ways of combining search clues in full-text retrieval he methodology investigated for ad hoc retrieval is t that of logistic regression, in which the retrieval rule akes the form of a regression equation fitted to e r learning data. Most of the variables used in th egression take the form of means rather than the s l more customary sums, and it is argued that this i ogically preferable. Radical manual reformulations t r of the topics were tried out and found to boos etrieval effectiveness. For routing retrieval, an - d approach based on the Assumption of Linked Depen ence, involving the extraction of relevance- - g associated stems from feedback documents, is investi ated. One characteristic of this approach is that f only a very minimal use is made of the original topic ormulation. Introduction The problem of full text retrieval...