异构网络学习排序模型及应用  

Learning to rank in heterogeneous network

作　　者：杨子[1] 唐杰[1] 李涓子[1] 

机构地区：[1]清华大学计算机科学与技术系,北京100084

出　　处：《中国科技论文在线》2011年第4期273-279,共7页

基　　金：高等学校博士学科点专项科研基金资助项目(20070003093);国家高技术研究发展计划(863计划)资助项目(2009AA01Z138)

摘　　要：针对网络排序问题中的基础,研究了话题层次的随机游走问题,提出了3步的方法解决该任务。重点从文档集中发现话题,在计算文档话题层次的排序得分等方面给出了详细的模型定义、求解过程和理论依据。提出了一个通用的异构网络排序模型,该模型一方面为源域和目标域之间的关联进行建模,同时在2个域中学习排序模型,并通过高效EM式的算法求解。通过实验验证了所提出的2种方法的性能。提出了特定专家搜索应用的概念,利用异构网络排序算法给出了该问题的解决方案。Massive heterogeneous resources have been currently available online,which brings richer information,along with more challenges.This paper concentrates on the issues of ＂learning to rank＂ in heterogeneous networks.First of all,the paper describes the investigation on topic-level random walk.We propose a three-step approach,especially focus on topic modeling of documents and the query and calculating a topic-level ranking score.Besides,we propose a general framework for heterogeneous cross-domain ranking,which simultaneously models the correlation between the source domain and the target domain,as well as learns the ranking models.We also develop an efficient EM-style solution,and discuss the generalized bound.The experiments show our proposed methods outperform the baseline methods.Finally,the thesis introduces the concept of specific expert search,which can be solved by the heterogeneous cross-domain ranking algorithm.

关 键 词：话题模型 随机游走 异构网络 专家搜索 伯乐搜索 

分 类 号：O223[理学—运筹学与控制论]