微博网络话题及用户影响力预测的研究

《微博网络话题及用户影响力预测的研究》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、摘要型与神经网络相比，结构上具有相当的灵活性，实验证明，在准确度和稳定性上具有一定优势，且具有较强的自适应性。（2）基于用户行为的社交影响力预测模型。该模型结合微吧用户行为建立复杂网络，通过训练算法的设计实现用户影响力的预测。对于用户影响力，本文主要根据用户行为定义影响力度量指标，并将其作为标准对用户进行聚类，得到具有不同影响力模式的用户群。应用到模型中，则通过预测用户类别来实现影响力的预测。该方法不仅能够提高预测的精度而且对于影响力的判断更具说服力。（3）在用户的分析中，主要应用统计学的方法对各类用户进行影响力，行为特征以及角色特征分析。该分析方法在个性化推荐，商业营

2、销等领域具有一定的应用价值关键词：复杂网络，话题流行度预测，自适应差分进化算法，聚类算法，影响力II万方数据上海交通大学硕士学位论文RESEARCHONTOPICANDUSERINFLUENCEPREDICTIONINMICRONETWORKABSTRACTWiththeinteractionbroughtbyweb2.0technology,socialnetworkhasbeenoneofthemostimportantsocialmedia.Micro-blogisoneofthemostpopularplatformsforinformationdissemin

3、ationandsharingbasedonsocialnetwork.ThecharacteristicpatternofMicro-bloghastriggeredlargeamountofusers’interest,whichacceleratesthefrequencyofthetopicreleaseandexpandthescopeofthetopictransfer.Thatisagreatchallengeforacademicresearchandapplication.Micro-barisapostbarbelongingtoMicro-blog

4、,whichisofstrongsubjectrelevant.SoMicro-barhasbeenstudiedasatopicinthispaper.Thetopicpopulationanduserinfluenceweremainlystudiedinthispaper.Themainjobs,conclusionsandinnovativepointsarelistedbelow:(1)Thetopicpopulationpredictionmodelwasbuiltbasedoncomplexnetwork.Thetopiccharactersrelated

5、withtopicpopulationwereabstractedbasedoncorrelationanalysisandqualitativeanalysis.Thenthesecharacterswereassignedtothemodelasinputvariables.Networkweightsweretrainedtopredictthetopicpopulation.Inthetrainingalgorithm,theself-adaptiverulefordifferentcrossoverstrategieswasdesignedtoensureth

6、equalityofthemutation.Themethodcannotonlystrengthenthemonitoringofthemutationquality,butcovertheshortageonconvergencerateandindividualdisturbanceforonecrossoverstrategy.III万方数据摘要Themainalgorithmparametersweresetthroughcomplexitytheoreticalderivation,whichisofcertainsignificanceinpractica

7、lapplicationandtheoreticalresearch.Incontrastwithneuralnetwork,thestructurewasmoreflexible.Andsimulationresultsshowthatthemodelismoreexcellentinaccuracyandstabilitywithbetteradaptability.(2)Socialinfluencepredictionmodelwasproposedbasedonuserbehavior,whichcombinedwithcomp