Location　recommendation　is　an　important　research　content　of　recommendation　system,　but　it　often　faces　the　problems　of　sparse　data　and　low　degree　of　personalization.　The　top-k　recommendation　is　selected　as　the　research　objective　to　model　users＇　rating　behavior　of　explicit　feedback　behavior.　A　personalized　location　recommendation　algorithm　LRA-CNN　based　on　convolutional　neural　network　(CNN)　is　designed　and　implemented.　The　LRA-CNN　combines　various　features　between　locations　and　study　their　joint　influence　between　users.　More　concretely,　co-appearing　and　geography　effects　in　locations　are　used　to　alleviate　check-in　data　sparse　matter　in　location　recommendation,　and　converted　into　the　feature　vector　representation　of　users　and　locations　by　feature　embedding.　Besides,　the　embedding　users　and　locations　are　fed　into　CNN　for　learning　high-order　interactions　among　various　features　adaptively.　Experimental　results　show　that　compared　with　several　traditional　methods,　the　proposed　algorithm　can　effectively　improve　the　accuracy　of　location　recommendation.