网页学习排序算法分析

《网页学习排序算法分析》由会员上传分享，免费在线阅读，更多相关内容在工程资料-天天文库。

1、ABSTRACTHigh-endadvancedinformationtechnologybroughtusintoavastdigitalage.Theinfluxofalargeamountofdatamakessearchenginebecomemoreandmoreimportant.Howtoquicklylocateneededinformationfromhugeamountsofdataisverycritical.Searchenginescontainmorethanonecomponentand

2、thewebpageschedulingpartwhichdeterminesthesearchenginesequencingresultsanddirectlyaffectstheperformanceofsearchenginesanduserexperienceisthecoreofsearchenginedesignproblem.Ininformationretrieval,therearemanywebpagesortingalgorithms,whichcanberoughlyclassifiedas

3、thepoint-wiseapproach,thepair-wiseapproachandthelist-wiseapproach.Researchershavemadegreatcontributionstousingavarietyofalgorithmsinthesethreekindsofmethodsandtheweblearningresearchisstillintheclimaxstage.Inviewoftheweblearningschedulingproblem,Firstly,wemodele

4、dtheweblearningschedulingmodelsbasedonSupportVectorMachinetheoryinpoint-wiseandpair-wiseapproachrespectively.BasedontheideaofcrossvalidationtochoosetheparametersoftheSVMmodelsandtheselectionofkernelfunctionisanalyzed・Normalizationandfractalfeaturedates*visualiz

5、ationanalysisgotdoneinthepreprocessingpart.Inpair-wisemodel,thisarticlegotthetrainingsampleintherandomsequencematchingmethod.Secondly,basedontheuseofheuristicmethodwehaveestablishedthesortinggeneticalgorithmtooptimizetheBPneuralnetworklearningmodel.Themodelused

6、theoptimizationabilityofgeneticalgorithmtogetgoodinitialweightsofBPnetworkandthresholdinordertoimprovetheperformanceofBPnetwork.Theprincipalcomponentanalysiswasusedtocompressthetrainingdataandtoguaranteehighdimensiondatafidelityaswellastomakeappropriatedegreeof

7、BPnetworkstructure.Thirdly,wedesignedaboostingpagesortingmodelwhichaimedatstudyingstrongcollator'spromotionperformanceonthebasisofweaklearningability.ExperimentsspecialforthethreemodelsonOHSUMEDdatasetwereconducted,resultanalysisandalgorithmcomparisonsdonenext.

8、Theexperimentalresultsshowthatpair-wiseapproachmodelperformedslightlybetterthanpoint-wiseapproachmodelandtheuseofgeneticalgorithmtooptimizeBPnetwork'sweightsandthresholdcani