semi supervis featur select graph classif xiangnan kong philip yu problem graph classif attract great interest decad current research graph classif assum exist larg amount label train graph applic label graph data expens difficult obtain copiou amount unlabel graph data paper studi problem semi supervis featur select graph classif propos solut call gssc effici search optim subgraph featur label unlabel graph exist featur select method vector space assum featur set perform semi supervis featur select graph data progress subgraph featur mine process deriv featur evalu criterion name gsemi estim us subgraph featur base label unlabel graph propos branch bound algorithm effici search optim subgraph featur judici prune subgraph search space empir studi real world task demonstr semi supervis featur select approach effect boost graph classif perform semi supervis featur select effici prune subgraph search space label unlabel graph 