Searching Distributed Collections With Inference Networks：（分布式搜索集合推理网络）.pdfVIP

Searching Distributed Collections With Inference Networks James P Callan Zhihong Lu W Bruce Croft Computer Science Department University of Massachusetts Amherst MA USA fcallanzlucroftgcsumassedu Abstract ronment therefore a decision must b e made ab out which of them to search A retrieval system should provide tech The use of information retrieval systems in networked en niques that can make this decision automatically b ecause vironments raises a new set of issues that have received lit users may b e unable or unwilling to make selections by ex tle attention These issues include ranking do cument col haustively examining long lists of the available collections lections for relevance to a query selecting the b est set of Having selected the collections to search the retrieval sys collections from a ranked list and merging the do cument tem must also provide techniques for eectively merging the rankings that are returned from a set of collections This individu al ranked lists of do cuments that are pro duced pap er describ es metho ds of addressing each issue in the in This pap er describ es how these issues can b e addressed ference network mo del discusses their implementation in in a retrieval system based on the inference net a probabilis the INQUERY system and presents exp erimental results tic mo del of information retrieval In the next section we demonstrating their eectiveness describ e related work on collection selection and merging of