面向用户任务的查询推荐研究张晓娟

:2013-12-17:2014-01-27*2012“”(:2012104010201)*(武汉大学信息资源研究中心武汉430072)AOL,Session,,;;,G353.41,,,,,,Google,,,,Session,,,,“”,“”,,,,,2,:,,Session,:Huang[1]Session,,;Shi[2]Fonseca[3]Session;Jones[4]Session,,,,,24520144XIANDAITUSHUQINGBAOJISHU35,:Mei[5],,HittingTime;Boldi[6]query-flow,,;Song[7]:URL;Song[8],;[9]Session,SimRankSimRank;Cao[10],,,,;Cao[11](vlHMM),,,,,,,[12,13],Session,:Jansen[14]Session,()Session;Jones[15]Session;Lucchese[16]WikipediaWikiDictionary,,,,33.1:,,,:(λ,θ)r(q,q)=λθ(q,q)(1),r(q,q)qq(q,q)qq,(,)(q,q)qq,,1,Session[4](q,q)3.2:SessionSession:Lucchese[16]AOL,74%Session;Liao[17],30%Session;Jones[15]Yahoo,17%Session,,(1),qq,(Session,),123t1t1Sq,q,q...q,q,q...Session,q,,,1q2q,2q3q,t1qt1qq,,q(2)decay(i,j,S)SessionSqq|ji|decay(i,j,S)(2),jqS,iqS,[6],0.8(3)qqSession,T(q,q)Sessionkdecay(i,j,S)T(q,q),kn|ji|Sk1T(q,q)n(3)(2),[16],,,[16]Di-gramJaccard,:expressionjaclevenstein(uu)2u(q,q)e(4),jacuDi-gramJaccard,levensteinu[16]jaculevensteinu1,,1,jaclevenstein(uu)2eexpressionu(q,q),,(3),[3],,,Wikipedia(ClueWeb09CategoryB[18]Wikipedia),Wikipedia,[3],t:12WC(t)(c,c,......c),Wicti,tf-idf(5)q()tqC(q)C(t)(5),:semanicC(q).C(q)s(q,q)|C(q)||C(q)|(6)r(q,q)(qq),,,(3)(4)(6),123,14[7-9],,(),,,4.1G=(V,E),V,E:24520144XIANDAITUSHUQINGBAOJISHU37,:CCr(q,q)P(q|q)(1)r(q,q)(7),P(q|q),Cr(q,q)qq,Cr(q,q)qSessionqqCr(q,q),5.2,URL,URL,URL,,URLURLURL[19]iqfURL(idf),:jjj|Q|iqf(d)log|Q|logn(d)logn(d)(8)ijijjcfiqf(q,d)ciqf(d)(9),jn(d)jd,|Q|cfiqfiqf,ijciqjd[19]query-url:jijjiijdDcfiqf(q,d)P(d|q)cfiqf(q,d)(10),ijjDcfiqf(q,d)iqcfiqf(11)(),1,0:,;,p(d|d)1(11)4.2:,[20],1,1q2q3q1d2d3d,,,,,,2q1d3d,,KL,图1经随机游走后的查询与文档间关系55.1AOL[21],200631531,2,IDURLURL,:“15”[13]Session图2AOL数据集格式5.2200,3(7),90%,0.6,,0.6图3随机游走参数ε取值与随机游走步数之间关系,4:(1)(1)(q,q),[6](,)(“Baseline1”);(2)(1)(q,q),(“Baseline2”);(3)(1)(q,q),[6](“Baseline3”);(4)(1)(q,q),(“OurWork”),,,,,“”“”Cohenkappa()[22],0.68,,P@N(PrecisionatN)Recip_Rank,P@NN,Recip_Rank“OurWork”“Baseline3”,((3)(4)(6))123,200,100,;100,,,“Baseline3”P@3P@5P@10Recip_Rank0.470.540.670.68;“OurWork”P@3P@5P@10Recip_Rank0.500.600.720.731P@3P@5P@10Recip_RankBaseline10.400.520.600.62Baseline20.420.530.600.63Baseline30.470.540.670.68OurWork0.500.600.720.7314,“Baseline3”“Baseline1”,“OurWork”“Baseline2”,,,“Baseline2”“Baseline1”,“OurWork”“Baseline3”,,[6](,),,2,,[23],[24]2P@3P@5P@10Recip_Rank0.420.580.490.710.640.800.640.820.430.570.520.680.630.810.670.79,,,:,Session,,Session24520144XIANDAITUSHUQINGBAOJISHU39;,,,,Session,,6,Session,,,,,:;,;,;;;,[1]HuangCK,ChienLF,OyangYJ.RelevantTermSuggestioninInteractiveWebSearchBasedonContextualInformationinQuerySessionLogs[J].JournaloftheAmericanSocietyforInformationScienceandTechnology,2003,54(7):638-649.[2]ShiX,YangCC.MiningRelatedQueriesfromWebSearchEngineQueryLogsUsinganImprovedAssociationRuleMiningModel[J].JournaloftheAmericanSocietyforInformationScienceandTechnology,2007,58(12):1871-1883.[3]FonsecaBM,GolgherPB,deMouraES,etal.UsingAssociationRulestoDiscoverSearchEnginesRelatedQueries[C].In:Proceedingsofthe1stConferenceonLatinAmericanWebCongress,2003.[4]JonesR,ReyB,MadaniO,etal.GeneratingQuerySubstitutions[C].In:Proceedingsofthe15thInternationalConferenceonWorldWideWeb.2006:387-396.[5]MeiQ,ZhouD,ChurchK.QuerySuggestionUsingHittingTime[C].In:Proceedingsofthe17thACMConferenceonInformationandKnowledgeManagement.2008:469-478.[6]BoldiP,BonchiF,CastilloC,etal.TheQuery-flowGraph:ModelandApplications[C].In:Proceedingsofthe17thACMConferenceonInformationandKnowledgeManagement.2008:609-618.[7]SongY,HeL.OptimalRareQuerySuggestionwithImplicitUserFeedback[C].In:Proceedingsofthe19thInternationalConferenceonWorldWideWeb.2010:901-910.[8]SongY,ZhouD,HeL.QuerySuggestionbyConstructingTerm-TransitionGraphs[C].In:Proceedingsofthe5thACMInternationalConferenceonWebSearchandDataMining.2012:353-362.[9],,.SimRank[J].,2010,24(3):3-10.(LiYa’nan,XuSheng,WangBin.ChineseQueryRecommendationbyWeightedSimRank[J].JournalofChineseInformationProcessing,2010,24(3):3-10.)[10]CaoHH,JiangDX,PeiJ,etal.Context-awareQuerySuggestionbyMiningClick-throughandSessionData[C].In:Proceedingsofthe14thACMSIGKDDInternationalConferenceonKnowledgeDiscoveryandDataMining(KDD’08).2008:875-883.[11]CaoHH,JiangDX,PeiJ,etal.TowardsContext-awareSearchbyLearningaVeryLargeVariableLengthHiddenMarkovModelfromSearchLogs[C].In:Proceedingsofthe18thInternationalConferenceonWorldWideWeb(’09).2009:191-200.[12]CatledgeL,PitkowJCharacterizingBrowsingBehaviorsontheWorldWideWeb[J]ComputerNetworksandISDNSystem,1995,27(6):1065-1073[13]HeDQ,GokerA.DetectingSessionBoundariesfromWebUserLogs[C].In:Proceedingsofthe22ndAnnualColloquiumonInformation.2000.[14]JansenB,SpinkA,KathuriaV.DefineSearchingSessionsonWebSearchEngines[J].JournaloftheAmericanSocietyforInformationScienceandTechnology,2007,58(6):862-871.[15]JonesR,KlinknerKL.BeyondtheSessionTimeout:AutomaticHierarchicalSegmentationofSearchTopicsinQueryLogs[C].In:Proceedingsofthe17thACMConferenceonInformationandKnowledgeM