In　order　to　accurately　obtain　potential　features　and　improve　the　recommendation　performance　of　the　collaborative　filtering　algorithm,　this　paper　puts　forward　a　collaborative　filtering　recommendation　algorithm　based　on　deep　neural　network　fusion　(CF-DNNF).　CF-DNNF　makes　the　best　of　the　implicit　attributes　of　data,　where　the　text　attributes　and　the　other　attributes　are　extracted　from　the　data　through　the　long　short-term　memory　(LSTM)　network　and　the　deep　neural　network,　respectively,　so　as　to　obtain　the　feature　matrix　that　contains　the　user　and　item　attribute　information.　Deep　belief　network　(DBN)　uses　the　feature　matrix　and　outputs　the　probability.　Besides,　this　paper　initially　discusses　an　interpretable　collaborative　filtering　recommendation　algorithm　based　on　deep　neural　network　fusion　(ICF-DNNF).　The　paper　compares　the　CF-DNNF　algorithm　with　probabilistic　matrix　factorisation　(PMF),　SVD,　and　restricted　Boltzmann-based　collaborative　filtering　(RBM-CF)　algorithms　on　the　MovieLens　dataset　and　the　Amazon　product　dataset.　Results　indicate　that　the　root　mean　square　error　(RMSE)　of　CF-DNNF　is　improved　by　2.015%,　and　the　mean　absolute　error　(MAE)　is　improved　by　2.222%.