Generalized Ensemble Model for Document Ranking in Information Retrieval

摘要

A generalized ensemble model (gEnM) for document ranking is proposed in thispaper. The gEnM linearly combines basis document retrieval models and tries toretrieve relevant documents at high positions. In order to obtain the optimallinear combination of multiple document retrieval models or rankers, anoptimization program is formulated by directly maximizing the mean averageprecision. Both supervised and unsupervised learning algorithms are presentedto solve this program. For the supervised scheme, two approaches are consideredbased on the data setting, namely batch and online setting. In the batchsetting, we propose a revised Newton's algorithm, gEnM.BAT, by approximatingthe derivative and Hessian matrix. In the online setting, we advocate astochastic gradient descent (SGD) based algorithm---gEnM.ON. As for theunsupervised scheme, an unsupervised ensemble model (UnsEnM) by iterativelyco-learning from each constituent ranker is presented. Experimental study onbenchmark data sets verifies the effectiveness of the proposed algorithms.Therefore, with appropriate algorithms, the gEnM is a viable option in diversepractical information retrieval applications.