Semi-supervised　learning　has　received　many　attention　during　the　few　past　years　in　the　machine　learning　community.　But　they　all　treat　the　two　errors　equally.　Since　in　many　important　applications,　however,　some　kinds　of　errors　are　more　important　than　others.　In　this　paper,　we　use　cost-sensitive　2v　-　SVM　to　solve　the　semi-supervised　binary　classification.　We　convert　the　problem　of　Semi-Supervised　binary　classification　into　a　feasible　Semi-definite　programming.