Privacy　leakage　resulting　from　information　sharing　in　online　social　networks　(OSNs)　is　a　serious　concern　for　individuals.　One　of　the　culprits　behind　the　problem　is　that　existing　privacy　policies　developed　for　OSNs　are　not　fine-grained　or　flexible　enough,　resulting　in　privacy　settings　that　could　hardly　meet　the　privacy　requirements　of　individuals.　Neither　would　such　privacy　settings　allow　individuals　to　control　where　the　information　could　go.　In　addition,　there　are　hardly　any　effective　mechanisms　for　measuring　potential　threats　to　privacy　during　information　propagation.　To　alleviate　the　situation,　in　this　paper,　we　propose　a　novel　privacy-preserving　information　sharing　scheme　for　OSNs　in　which　information　flow　can　be　controlled　according　to　the　privacy　requirements　of　the　information　owner　and　the　context　of　the　information　flow.　Specifically,　we　first　formally　define　the　privacy-dependent　condition　(PDC)　for　information　sharing　in　OSNs　and　then　design　a　PDC-based　privacy-preserving　information　sharing　scheme　(PDC-InfoSharing)　to　protect　the　privacy　of　individuals　according　to　the　heterogeneous　privacy　requirements　of　individuals　as　well　as　the　potential　threats　that　they　may　face.　Furthermore,　to　balance　information　sharing　and　privacy　protection,　the　techniques　of　reinforcement　learning　is　utilized　to　help　individuals　reach　a　trade-off.　PDC-InfoSharing　would　allow　the　privacy　policies　for　specific　information　audience　to　be　derived　based　on　PDC　to　achieve　dynamical　control　of　the　flow　of　information.　Theoretical　analysis　proves　that　the　proposed　scheme　can　assist　individuals　in　adopting　fine-grained　privacy　policies　and　experiment　shows　that　it　can　adapt　to　different　situations　to　help　individuals　achieve　the　trade-off　between　information　sharing　and　privacy　protection.