论文部分内容阅读
目前的研究表明，Boosting算法在各种任务中都能提供良好的预测性能。而在学习排序中，基于Boosting的模型，例如Rankboost和LambdaMART，在对公共数据集的评估中表现优秀。本文通过研究随机森林算法和LambdaMART，将随机森林算法作为基础模型，学习一个排序函数，将函数的输出作为LambdaMART的初始函数，最终生成排序模型。在公共数据集上基于评价指标ERR和NDCG对排序模型进行验证，结果表明本排序模型均要优于原始算法。
The current research shows that Boosting algorithm can provide good predictive performance in a variety of tasks. In learning ranking, Boosting-based models, such as Rankboost and LambdaMART, performed well in the evaluation of public data sets. In this paper, by studying random forest algorithm and LambdaMART, taking stochastic forest algorithm as the basic model, we learn a sorting function, and take the output of the function as the initial function of LambdaMART to generate the sorting model finally. The validation of the ranking model based on the evaluation indexes ERR and NDCG on the public data set shows that the ranking model is better than the original one.