In　this　paper,　we　propose　a　mining　method　to　discover　high-level　semantic　knowledge　about　human　social　interactions　in　small　group　discussion,　such　as　frequent　interaction　patterns,　the　role　of　an　individual　(e.g.,　the　＇centrality＇　or　＇power＇),　subgroup　interactions　(e.g.,　two　persons　often　interact　with　each　other),　and　hot　sessions.　A　smart　meeting　system　is　developed　for　capturing　and　recognizing　social　interactions.　Interaction　network　in　a　discussion　session　is　represented　as　a　graph.　Interaction　graph　mining　algorithms　are　designed　to　analyze　the　structure　of　the　networks　and　extract　social　interaction　patterns.　Preliminary　results　show　that　we　can　extract　several　interesting　patterns　that　are　useful　for　interpretation　of　human　behavior　in　small　group　discussion.　©　Springer-Verlag　Berlin　Heidelberg　2011.