Learning to Rank
Microsoft (China) Co., Ltd.
Starting Date & Finishing Date:
Researcher in Charge:
Learning to rank aims to solve the problem that users can get the exact information with a query. There are two factors affecting the accuracy of the learned model: the preparation of training set and the design of the learning algorithm. It needs to find an interpretable model with a certain predictive accuracy. We suggest that Reduct theory is employed to form the sets of rules, and each reduct is served as ‘weak classifier’. Because these rules on reduct are interpretable, it is possible to analyze whether it is the preparation of the training set or the learning algorithm that results in the low prediction accuracy of the model. If we are lucky enough, some clues on how to improve the preparation of the training set might be found.