Cost-Sensitive listwise ranking approach

Authors:
Min Lu;MaoQiang Xie;Yang Wang;Jie Liu;YaLou Huang
Affiliations:
College of Information Technology Science, Nankai University, Tianjin, China;College of Software, Nankai University, Tianjin, China;College of Information Technology Science, Nankai University, Tianjin, China;College of Information Technology Science, Nankai University, Tianjin, China;College of Information Technology Science, Nankai University, Tianjin, China
Venue:
PAKDD'10 Proceedings of the 14th Pacific-Asia conference on Advances in Knowledge Discovery and Data Mining - Volume Part I
Year:
2010

Citing 8
Cited 0

Cumulated gain-based evaluation of IR techniques

ACM Transactions on Information Systems (TOIS)
Optimizing search engines using clickthrough data

Proceedings of the eighth ACM SIGKDD international conference on Knowledge discovery and data mining
Learning to rank: from pairwise approach to listwise approach

Proceedings of the 24th international conference on Machine learning
AdaRank: a boosting algorithm for information retrieval

SIGIR '07 Proceedings of the 30th annual international ACM SIGIR conference on Research and development in information retrieval
Query-level loss functions for information retrieval

Information Processing and Management: an International Journal
Listwise approach to learning to rank: theory and algorithm

Proceedings of the 25th international conference on Machine learning
Generalization analysis of listwise learning-to-rank algorithms

ICML '09 Proceedings of the 26th Annual International Conference on Machine Learning
Cost-Sensitive learning of SVM for ranking

ECML'06 Proceedings of the 17th European conference on Machine Learning

Quantified Score

Hi-index	0.00

Visualization

Abstract

This paper addresses listwise approaches in learning to rank for Information Retrieval(IR). The listwise losses are built on the probability of ranking a document highest among the documents set. The probability treats all the documents equally. However, the documents with higher ranks should be emphasized in IR where the ranking order on the top of the ranked list is crucial. In this paper, we establish a framework for cost-sensitive listwise approaches. The framework redefines the probability by imposing weights for the documents. The framework reduces the task of weighting the documents to the task of weighting the document pairs. The weights of the document pairs are computed based on Normalized Discounted Cumulative Gain(NDCG). It is proven that the losses of cost-sensitive listwise approaches are the upper bound of the NDCG loss. As an example, we propose a cost-sensitive ListMLE method. Empirical results shows the advantage of the proposed method.