A Self-Training Listwise Method for Learning to Rank with Partially Labeled Data

摘要：

Recently there has been a general direction in ranking algorithms that combine labeled and unlabeled data in document retrieval. Previous work for learning to rank has focused on cases where only the pairwise approach is available for essential ranking algorithms. This paper addresses the semisupervised ranking problems where the listwise approach is used to construct ranking models. The method is an iterative self-training algorithm that in each iteration a ranking function is built by learning from the current set of labeled queries. The newly learned ranking function is produced, then it is used to teaching unlabeled query. The likelihood loss is employed to evaluate the similarity of two permutations for a given query. The experimental results show the effectiveness of the method proposed in this paper.

关键词： learning to rank semi-supervised listwise selftraining document retrieval

作者: Hai-jiang He

作者单位: Department of Computer Science and Technology Changsha University Changsha, China

会议类型: 国际会议

会议名称: 2010 International Conference on Signal and Information Processing(2010年IEEE信号与信息处理国际会议 ICSIP2010)

会议地点: 长沙

会议语种:英文

页码: 300-303

在线出版日期: 2010-12-14（万方平台首次上网日期，不代表论文的发表时间）