LTRRS: A Learning to Rank Based Algorithm for Resource Selection in Distributed Information Retrieval

摘要

Resource selection is a key task in distributed information retrieval. There are many factors that affect the performance of resource selection. Learning to rank methods can effectively combine features and are widely used for document ranking in web search. But few of them are explored for resource selection. In this paper, we propose a resource selection algorithm based on learning to rank called LTRRS. By analyzing the factors affecting the eflfec-tiveness of resource selection, we extract multi-scale features including term matching features, topical relevance features and central sample index (CSI) based features. By training LambdaMART learning to rank model, we directly optimize NDCG metric of resource ranking list in LTRRS. Experiments on the Sogou-QCL dataset show that LTRRS algorithm can significantly outperform the baseline methods in NDCG and precision metrics.