Most　link　prediction　algorithms　are　designed　for　static　knowledge　graphs　(KGs).　Notably,　KGs　usually　evolve　with　time.　Currently,　researchers　pay　more　attention　to　temporal　knowledge　graph　completion　(TKGC).　However,　existing　methods　are　insufficient　to　utilize　the　complex　structure　between　entities　and　their　neighbor　nodes,　and　modeling　of　entities　is　not　comprehensive　enough.　Therefore,　we　propose　DEGAT　(Dynamic　Embedding　Graph　Attention　Networks),　an　attention-based　TKGC　method.　Specifically,　we　use　a　generalized　graph　attention　network　as　an　encoder　to　aggregate　the　features　of　neighbor　nodes　and　relations.　Thus,　the　model　can　learn　the　features　of　entities　from　their　neighbors　without　complicated　matrix　operations.　In　decoder,　we　introduce　a　diachronic　embedding　function　for　entities.　We　build　a　novel　model　DE-ConvKB　for　TKGC　by　equipping　static　models　with　a　dynamic　entity　embedding　function　that　provides　the　features　of　entities　at　any　point　time.　The　results　show　significant　improvements　over　the　state-of-the-art　methods　on　three　public　datasets.　©　2022,　The　Author(s),　under　exclusive　license　to　Springer　Nature　Switzerland　AG.