The　collaborative　filtering　algorithm　(CF)　is　one　of　the　most　important　algorithms　in　the　recommendation　system.　Recently,　the　use　of　neural　word　embedding　methods　to　learn　the　latent　representation　of　words　has　become　a　mature　method　in　the　field　of　natural　language　processing　(NLP).　Inspired　by　the　SGNS　algorithm　in　the　NLP　field,　we　propose　f-item2vec　as　a　new　method　to　learn　the　latent　representation　in　item　vector　space,　and　based　on　this　vector　to　calculate　the　similarity　between　items.　In　addition,　according　to　the　forgetting　process,　we　propose　a　new　user　preference　model　to　accurately　identify　the　user＇s　short-term　preferences　and　long-term　preferences.　The　experimental　results　show　that　the　proposed　method　is　effective　and　superior　to　the　traditional　algorithm　in　predicting　the　score　and　generating　the　recommendation　list.