Toaddressthesecondproblem,wedevelopaProbabilisticTopic-Connection(PTC)modeltorepresenttheconnectionbetweenlatentsemantictopicintextdescriptionandlatentpatternsfromimagefeaturespace.WecomparetheperformanceofourmodelwiththecurrentlypopularCorrespondenceLDA(Corr-LDA)modelunderthesameautomati