In　recent　years,　we　have　witnessed　the　development　of　location-based　services　which　benefit　users　and　businesses.　This　paper　aims　to　provide　a　unified　framework　for　location-aware　recommender　systems　with　the　consideration　of　social　influence,　categorical　influence　and　geographical　influence　for　users＇　preference.　In　the　framework,　we　model　the　three　types　of　information　as　functions　following　a　power-law　distribution,　respectively.　And　then　we　unify　different　information　in　a　framework　and　learn　the　exact　function　by　using　gradient　descent　methods.　The　experimental　results　on　real-world　data　sets　show　that　our　recommendations　are　more　effective　than　baseline　methods.