Music　recommendation　systems　based　on　deep　learning　have　been　actively　explored　using　hybrid　approaches.　However,　most　of　the　models　proposed　by　previous　studies　adopt　coarse-grained　embedding　approaches　(e.g.,　CNNs)　to　characterize　audio　features.　Users’　fine-grained　preferences　for　music　content　have　not　been　effectively　explored　yet.　In　this　work,　we　propose　a　hybrid　music　recommendation　model　based　on　attention　mechanism,　which　integrates　user’s　historical　behaviour　records　and　audio　content　and　can　capture　the　user’s　fine-grained　preferences　for　music　content　due　to　the　introduction　of　attention　mechanism.　We　experimented　with　a　subset　of　the　last.fm-1b　dataset　(30,753　users,　10,000　songs,　1533,245　interactions).　The　experimental　results　show　that　our　method　outperforms　baselines　approaches.　©　2021,　Springer　Nature　Switzerland　AG.