A Rich Ranking Model Based on the Matthew Effect Optimization
Most existing approaches of learning to rank treat the effectiveness of each query equally which results in a relatively lower ratio of queries with high effectiveness (i.e. rich queries) in the produced ranking model. Such ranking models need to be further optimized to increase the number of rich queries. In this paper, queries with different effectiveness are distinguished, and the queries with higher effectiveness are given higher weights. We modify the gradient in the LambdaMART algorithm based on a new perspective of Matthew effect to highlight the optimization of the rich queries and to produce the rich ranking model, and we present a consistency theorem for the modified optimization objective. Based on the effectiveness evaluation criteria for information retrieval, we introduce the Gini coefficient, mean-variance and quantity statistics to measure the performances of the ranking models. Experimental results show that the ranking models produced by the gradient-modified LambdaMART algorithm based on Matthew effect exhibit a stronger Matthew effect compared to the original LambdaMART algorithm.
KeywordsLearning to rank Ranking model Matthew effect LambdaMART algorithm Gradient
- 1.Page, L., Brin, S., Motwani, R., Winograd, T.: The pagerank citation ranking: bringing order to the web. Technical report, Stanford Digital Library Technologies Project (1999)Google Scholar
- 4.Burges, C.J.: From ranknet to lambdarank to lambdaMART: an overview. Microsoft Research Technical report MSR-TR-2010-82 (2010)Google Scholar
- 7.Xu, J., Zeng, W., Lan, Y., Guo, J., Cheng, X.: Modeling the parameter interactions in ranking SVM with low-rank approximation. IEEE Trans. Knowl. Data Eng. (2018, in Press)Google Scholar
- 9.Chapelle, O., Metlzer, D., Zhang, Y., Grinspan, P.: Expected reciprocal rank for graded relevance. In: Proceedings of the 18th ACM Conference on Information and Knowledge Management, pp. 621–630. ACM (2009)Google Scholar