多查询相关的排序支持向量机融合算法.pdf

ISSN 1000- 1239PCN 11- 1777PTP Journal of Computer Research and De elopment 48( 4) : 558- 566, 2011 1 1, 2 2 1 1 1 王 扬 黄亚楼 谢茂强 刘 杰 卢 敏 廖 振 1 ( 300071) 2 ( 300071) ( wangyang022@ mail. nankai. edu. cn) A Multiple Query Dependent Ranking SVM Aggregation Algori thm 1 1, 2 2 1 1 1 Wang Yang , Huang Yalou , Xie M aoqiang , Liu Jie , Lu M in , and Liao Zhen 1 ( Colleg eof Inf ormation Technical Science, N ankai Univ ersity , T ianj in 30007 1) 2 ( Colleg eof Sof tw are, Nankai Univ ersity , Tianj in 30007 1) Abstract Super ised ranking appr oaches ha e been becoming more and mor e important in the fields of infor mation retrie al and machine learning. In r anking for document retrie al, quer ies often ary greatly fr om one to another. Only the documents retr ie ed from the same query are to be ranked against each other. How e er, in most of the ex isting approaches, losses from different queries ar e defined as the same. T he significant di ersities existing among queries are taken into consideration, and a r ank aggregation fr amew or k for multiple dependent queries is proposed. T his fr amew or k contains tw o steps, training of base rankers and query-le el aggr egation. T raining of base ranker sets up a number of query-dependent base rankers based on each query and its rele ant documents, and then turns the output of base rankers into featur e ectors. Query-le el aggr egation uses a super ised approach to learn query-dependent w eights w hen these base r anker s are combined. As a case study , an SVM based model is employ