The　rapid　growth　of　Location-Based　Social　Networks　(LBSNs)　has　led　to　the　generation　of　large　amounts　of　users＇　check-in　data,　which　has　driven　the　development　of　many　location-based　recommendation　services.　Point-of-Interest　(POI)　recommendation　is　one　such　service　that　helps　users　find　places　they　are　interested　in　based　on　the　current　time　and　location.　Unlike　traditional　recommendation　tasks,　users＇　check-in　data　contains　rich　heterogeneous　data　such　as　time,　geographical　information　and　social　relationship　information;　thus　it　is　challenging　to　capture　the　complex　contextual　relationships　between　these　heterogeneous　information　for　POI　recommendation.　To　solve　this　problem,　we　propose　a　Metapath　Aggregated　Heterogeneous　Graph　Embeddings　method(MAHGE).　Specially,　it　firstly　proposes　a　novel　method　to　construct　the　heterogeneous　LBSN　graph　which　innovatively　models　time　as　the　relationship　on　the　edges　of　the　graph　in　order　to　capture　the　complex　dependency　between　user　and　time.　Then,　it　proposes　to　profile　the　target　node　based　on　meta-paths　because　meta-path　reflects　the　characteristics　of　target　node　from　a　multi-dimensional　perspective.　Moreover,　it　introduces　a　graph　embedding　method　based　on　meta-path　aggregation　to　learn　the　vector　representation　of　the　target　node　with　attention　mechanism.　Finally,　extensive　experiments　on　two　real-word　datasets　are　conducted,　and　the　results　show　the　effectiveness　of　this　method.