K.-H. Jung and J. Lee (Korea)
Learning to Rank, Information Retrieval, Data Mining
As the volume of database grows, retrieving only relevant information and ordering them according to the relevance becomes an important issue. This ranking problem of database has been studied extensively in the view of link structure analysis and some of successful methods have been adapted to widely used commercial search engines. Recently, ranking problem is started to be seen as a machine learning problem and various learning to rank methods have been proposed. Even though there are several probabilistic ranking models developed, they have difficulties in applying to information retrieval in the realistic setting. In this paper, we propose a new ranking method in the probabilistic framework based on the domain described multi-class classifier. Experimental results on both synthetic and real large-scale dataset show that the proposed method has comparable or better performance with useful properties for information retrieval.
Important Links:
Go Back
