The　traditional　collaborative　filtering　algorithm　only　uses　the　item-user　scoring　matrix,　but　does　not　consider　the　semantics　of　the　item.　The　recommendation　effect　is　often　not　ideal.　The　connotation　knowledge　of　the　movie　and　the　interest　preference　of　the　user　are　added　to　the　knowledge　graph,　wherein　the　connotation　knowledge　is　represented　by　the　relationship　between　the　movie　entity　and　its　characteristic　entity,　and　the　user＇s　interest　preference　is　to　use　the　like　relationship　and　the　like　relationship　between　the　movie　entity　and　the　user　entity.　Said.　After　adding　the　intension　knowledge　and　the　user＇s　interest　preferences　in　the　knowledge　graph,　the　complex　and　diverse　relationship　between　the　entities　is　represented　in　the　form　of　a　triple.　The　knowledge　graph　is　used　to　represent　the　learning　method,　the　movie　knowledge　graph　is　embedded　in　a　low-dimensional　semantic　space,　the　movie　entity　is　vectorized,　and　then　the　semantic　similarity　between　the　movie　entities　is　calculated,　similar　to　the　item-based　collaborative　filtering.　Sexual　combination,　the　film＇s　own　connotation　information　and　user　preferences　in　the　knowledge　graph　are　integrated　into　the　collaborative　filtering　for　recommendation,　which　can　not　only　make　up　for　the　problem　of　semantic　sparsity,　but　also　can　solve　the　lack　of　user　subjectivity　of　the　knowledge　graph　because　the　knowledge　graph　contains　the　user＇s　subjective　interest　preferences.　In　the　knowledge　graph,　both　the　objectivity　of　the　connotation　knowledge　of　the　film　itself　and　the　subjectivity　of　the　user＇s　interest　are　considered.　Compared　with　the　connotation　knowledge　of　the　film　itself　in　the　knowledge　graph,　user　preferences　and　object-based　items　are　added　to　the　knowledge　graph.　The　synergistic　filtering　fusion　enhances　the　recommended　effect.　©　2019　IEEE.