Research On Learning To Rank Based On LambdaXGB

With the increasing choices,Search engines and Recommendation systems more and more dependent on the sort.However,little factor is only considered by the traditional sorting algorithm.With the exponential growth of processed data,multiple factors need to be combined for sorting,combining with different weights.So there is Learning to Rank.Learning to Rank is a sort of supervised learning method,which can get a rank model according to the training data,and use this rank model to sort the data.The Pairwise algorithm is a kind of Learning to Rank algorithm.Pairwise algorithm is transformed into binary classification problem in ranking of documents.For the documents of the same query,the training samples of binary classifier training are obtained for any two different labels.All the document pairs are sorted to get a partial order relation of all the documents,and the final rank is achieved.The Pairwise approach includes RankNet,LambdaRank,LambdaMART,Ranking SVM,IR SVM,RankBoost.Because of low efficiency current situation,based on analysis and research of the LambdaMART algorithms and the XGBoost algorithm,this paper presents LambdaXGB L1 algorithm,LambdaXGB L2 algorithm and LambdaXGB algorithm.XGBoost is based on the GBDT model,which also called CART.So paper combine the LambdaMART and XGBoost,and add L1-regularization term and L2-regularization term into the loss function of the LambdaMART.Three algorithms are proposed,including LambdaXGB L1 algorithm,LambdaXGB L2 algorithm and LambdaXGB algorithm.Through the MQ2008 dataset,this paper reveals the NDCG evaluation result compared with RankNet and LambdaMART,and verifies the effectiveness of these algorithms.The results demonstrate that our approach gives state-of-the-art results on a rank of data set.