Various research papers on learning models in IR
Directly Optimizing Evaluation Measures in Learning to Rankhot! 02/05/2009 Hits: 791
Microsoft - In this paper, we conduct a study onthe approach of directly optimizing evaluation measures in learningto rank for Information Retrieval (IR). We focus on the methodsthat minimize loss functions upper bounding the basic loss functiondefined on the IR measures. We first provide a general frameworkfor the study and analyze the existing algorithms of SVMmap andAdaRank within the framework. The framework is based on upperbound analysis and two types of upper bounds are discussed. Moreover,we show that we can derive new algorithms on the basis of thisanalysis and create one example algorithm called PermuRank. Wehave also conducted comparisons between SVMmap, AdaRank, PermuRank,and conventional methods of Ranking SVM and Rank-Boost, using benchmark datasets. Experimental results show thatthe methods based on direct optimization of evaluation measurescan always outperform conventional methods of Ranking SVM andRankBoost. However, no significant dierence exists among theperformances of the direct optimization methods themselves.
Learning to Rank Relational Objectshot! 02/05/2009 Hits: 765
Microsoft - This paper addresses the issueand formulates it as a novel learning problem, referred toas, `learning to rank relational objects'. In the new learningtask, the ranking model is de¯ned as a function of not onlythe contents (features) of objects but also the relations be-tween objects. The paper further focuses on one setting ofthe learning problem in which the way of using relation in-formation is predetermined. It formalizes the learning taskas an optimization problem in the setting. The paper thenproposes a new method to perform the optimization task,particularly an implementation based on SVM.