Deep　neural　networks,　especially　graph　neural　networks,　have　made　great　progress　in　aspect-based　sentiment　analysis.　Knowledge　graphs　can　provide　rich　auxiliary　information　for　aspect-based　sentiment　analysis.　However,　existing　models　cannot　effectively　learn　aspect-specific　sentiment　features　from　the　review　text　and　external　knowledge.　They　cannot　accurately　select　knowledge　entities　that　are　highly　relevant　to　the　aspect.　They　also　ignore　the　semantic　interaction　between　the　review　text　and　external　knowledge.　To　address　these　issues,　we　propose　a　knowledge-enhanced　interactive　graph　convolutional　network　(KE-IGCN).　First,　we　introduce　a　subgraph　construction　strategy　to　construct　a　syntax-guided　knowledge　subgraph,　which　can　guide　KE-IGCN　in　selecting　highly　relevant　knowledge　entities.　Second,　we　propose　a　knowledge　interaction　mechanism　to　exploit　the　semantic　interaction　between　external　knowledge　and　the　review　text.　We　then　use　multilayer　graph　convolutional　networks　to　learn　aspect-specific　sentiment　features　from　the　review　text　and　external　knowledge　jointly　and　interactively.　We　also　use　a　multilevel　feature　fusion　mechanism　to　aggregate　aspect-specific　sentiment　features　from　semantic　and　syntactic　information　of　the　review　and　external　knowledge.　Experimental　results　on　four　public　datasets　demonstrate　that　KE-IGCN　outperforms　other　state-of-the-art　baseline　models.