ChinesesearchenginegiantBaiduhasdevelopedamachine-learningalgorithmthatpredictswhethercrowdsarelikelytoformatcertainlocationsintwohours,whichcouldbeusedtopreventstampedes.
ThealgorithmusesdatafromBaidu'smapapp,saidaresearchreportpublishedbyBaidu'sBigDataLabTuesday,studyingthenumberofmapqueriesandthenumberofusersinanarea.
Itsmapappaccountsforover70percentofChina'smappingservicesmarket,thecompany'suserstatisticsshow.ManyChineseusersoftensearchforanidealtravelrouteusingBaiduMaps.
ThelabbegantoconcentrateonhowtopredictcrowdsincertainareasfollowingtheShanghaistampedeonNewYear'sEveinDec2014,where36peoplewerekilledduringcelebrationsontheBundriverfront,accordingtoWuHaishan,co-authorofthereport.
Usingthemapdata,theyfoundawaytodeterminethenumbersinrealtimeandtriggerwarnings30minutestotwohoursaheadoftimeifunusuallylargecrowdsareexpectedtogather.Thisoccurswhenthenumberofqueriesforaspecificlocationcrossesasetthreshold.
Machinelearningisappliedsothatthealgorithmcanlearnfromsearchqueriesandbetterpredictfuturecrowds.
Wusaidthatthealgorithmcouldbeusedbylocalgovernments,authoritiesandevenvenueoperatorstomonitorcrowds.
ItmayeventuallyprovidethefunctionasapublicservicefeatureonBaiduMaps,accordingtoWu.Ifimplemented,thefeaturewouldallowuserstoseewhatthecrowdsarelikeinvariouslocationsusingtheapp.
(CHINADAILY)