Sparse Bayesian Learning for Ranking

In this paper, we propose a sparse Bayesian kernel approach to learn ranking function. In sparse Bayesian framework, a relevance determination prior over weights is used to automatic relevance determination. The inference techniques based on Laplace approximation is derived for model selection. By this approach accurate prediction models can be derived, which typically utilize dramatically fewer basis functions than the comparable SVM-based approaches while offering a number of additional advantages. This algorithm is implemented and analysis on synthesis data. The compared with two state-of-the-art algorithms is done on document retrieval data. Experimental results show that the right ranking function can be learned and the generalization performance of this approach competitive with SVM-based method and Gaussian process based method.