Search Result Refinement via Machine Learning from Labeled-Unlabeled Data for Meta-search
İbrahim Burak Özyurt, Greg G. Brown · 2007
For a user, retrieving relevant information from search engines involves encoding her intent, at best partially, in search keywords. A small amount of user feedback, can be beneficial in refining the results returned by the search engines and aiding exploratory search for scientific literature and data. In this paper, three new variants to EM method for semi-supervised document classification by K. Nigam et al. (2000) is introduced for biomedical literature meta-search result refinement. Multi-mixture per class EM variant with agglomerative information bottleneck clustering by N. Slonim and N. Tishby (1999) using Davies-Bouldin cluster validity index by D. Davies and D. Bouldin (1979), has shown retrieval performance rivaling the state of the art transductive support vector machines (TSVM) by T. Joachims (1999) with more than one order of magnitude improvement in execution time