Most　existing　team　discovery　methods　are　based　on　collaboration　networks　using　papers　or　patents　data.　They　usually　have　low　efficiency　because　they　have　to　create　the　whole　network　containing　all　researchers.　In　addition,　these　methods　can＇t　immediately　output　research　topics　for　each　discovered　team.　A　novel　team　discovery　method　is　presented　to　solve　these　problems.　The　method　extracts　institutional　names　from　papers　and　patents　to　build　the　institution　base,　and　extracts　authors　and　inventors　to　build　the　researcher　base　after　name　disambiguation.　Then,　the　method　exploits　Author　Topic　model　to　mine　distributions　of　topics　and　researchers　in　papers　and　patents　and　builds　research　topic　base.　The　component　analysis　technique　is　used　to　discover　teams　under　each　research　topic　by　analyzing　its　collaboration　network.　Experiments　show　the　proposed　method　can　identify　teams　without　establishing　a　whole　network　by　integrating　papers　and　patent　data.　Meanwhile,　the　method　can　provide　research　topics　for　found　teams.　©　2019　Association　for　Computing　Machinery.