排序是信息检索中的一个重要的环节,当今已经提出百余种用于构建排序函数的特征,如何利用这些特征构建更有效的排序函数成为当今的一个热点问题,因此排序学习(learning to rank)作为信息检索与机器学习的交叉学科,越来越受到人们的重视.根据不同的原则,查询可以分为不同的类别.不同类别的查询,排序特征的重要性不同,在排序函数的构建过程中的权重也会不同.为所有的查询都采用统一的排序函数是不合理的.针对这一问题,首先对利用关键词匹配原则得到的查询特征进行分析,选择出适当的查询特征集合构建查询特征向量,然后基于查询特征向量之间的距离对查询进行聚类,并为每个聚类类别学习得到排序函数,最后为一个新来的查询选择最适合的排序函数对文档进行排序.实验结果显示,在经过查询特征选择的查询聚类基础上得到的排序函数,和在所有的查询类别上得到的排序函数,两者的性能具有可比性,甚至前者优于后者.
Ranking is an essential part of information retrieval. Nowadays there are hundreds of features for constructing ranking functions and it is a hot research topic that how to use these features to construct more efficient ranking functions. So learning to rank, an interdisciplinary field of information retrieval and machine learning, has attracted increasing attention. Queries could be classified into several types based on different criterions, and the importance of ranking features is divergent for different types of queries. It is unpractical to apply a general ranking function for all queries. In this paper, we analyse the query features based on keyword mathcing and constrcut quey feautre vectors through the selected query features. Then the queries are clustered into several clusters and ranking functions are learned for each cluster. Finally, the fittest ranking function is chosen for a new coming query and ranks the documents. The experimental results show that the ranking functions based on query clustering with selected query features are comparable with or even outperfom the one based on all queries.