multi label learn exploit label depend min ling zhang kun zhang multi label learn train set label task predict proper label set unseen due tremend exponenti number label set task learn multi label exampl challeng kei success multi label learn effect exploit correl label facilit learn process paper propos bayesian network structur effici encod condit depend label featur set featur set common parent label make practic give approxim effici procedur find network structur network multi label learn decompos seri singl label classif problem classifi construct label incorpor parent label addit featur label set unseen exampl predict recurs label order network extens experi broad rang data set valid effect approach establish method 