With　the　wide　application　of　deep　learning　in　the　field　of　recommendation,　item　recommendation　has　gradually　become　a　new　hotspot.　The　existing　item　recommendation　methods　ignore　the　context　relationship　between　multiple　attributes　and　fail　to　make　full　use　of　the　potential　structural　interaction　relationship.　This　paper　proposes　a　new　multi-attribute　context-aware　item　recommendation　method　(MCIR)　based　on　deep　learning　to　recommend　a　series　of　suitable　items　for　characters　in　the　match.　The　attribute　attention　encoder　(AAE)　module　proposed　in　this　method　mainly　decouples　the　local　and　global　features　of　multiple　attributes,　and　the　global　multiple　aggregation　(GMA)　module　integrates　the　obtained　context　features　in　a　graph　structure.　Extensive　experiments　on　the　kaggle　public　game　dataset　show　that　our　method　significantly　outperforms　previous　methods　in　terms　of　Precision,　F1　and　MAP　compared　to　other　existing　methods.　©　2022　IEEE.