Nowadays,　people　live　in　an　era　of　information　explosion.　People　have　to　spend　a　lot　of　time　and　energy　to　filter　their　own　resources　every　day.　How　to　intelligently　recommend　resources　to　users　is　a　big　challenge.　At　the　same　time,　the　sparseness　of　user-to-item　rating　data　is　one　of　the　major　factors　that　deteriorate　the　quality　of　the　recommendation　system.　In　order　to　deal　with　the　problem,　several　recommended　techniques　have　been　proposed,　which　additionally　consider　the　auxiliary　information　to　improve　the　accuracy　of　the　rating　prediction,　such　as　textual　data　for　comments,　summaries,　or　summaries.　However,　due　to　the　inherent　limitations　of　the　traditional　word　pocket　model,　the　contextual　information　of　the　document　will　be　difficult　to　be　effectively　utilized.　Therefore,　this　paper　presents　a　novel　context-aware　recommendation　Long　Short　Term　matrix　decomposition　(LstmMF)　that　integrates　the　Long　Short　Term　neural　network　(LSTM)　into　the　probability　matrix　factor　(PMF)　to　capture　the　context　information　of　the　review　document　and　further　improve　the　accuracy　of　the　rating　prediction.　The　extensive　evaluation　of　the　movielens　dataset　shows　that　LstmMF　is　significantly　superior　to　the　most　advanced　recommendation　model　and　has　a　83.6%　accuracy　even　if　the　rating　data　is　very　sparse.