To　improve　the　effect　of　dynamic　recommendation,　the　time　characteristics　from　the　perspective　of　temporal　non-uniformness　and　consecutiveness　is　refined.　The　similarity　of　time　vectors　is　measured　using　grey　relational　analysis　(GRA)　and　incorporated　with　the　matrix　factorization　algorithm.　A　new　matrix　decomposition　algorithm　is　proposed,　which　can　alleviate　the　data　sparsity　caused　by　dividing　the　check-in　matrix　with　time　slots.　To　achieve　personalized　recommendation,　the　adaptive　kernel　density　estimation　is　leveraged　to　capture　the　personalized　spatial　preference,　and　thus　enhance　the　recommendation　quality.　On　this　basis,　a　novel　point-of-interest　(POI)　recommendation　algorithm　is　designed.　Experiment　results　show　the　proposed　algorithm　can　effectively　improve　the　precision　and　recall.　©　2022,　Editorial　Office　of　Systems　Engineering　and　Electronics.　All　right　reserved.