Video　quality　is　essential　for　many　consumer　electronic　devices.　However,　existing　quality　enhancement　methods　do　not　pay　enough　attention　to　the　video　characteristics.　The　success　of　artificial　intelligence　(AI)　can　help　improve　the　quality　of　compressed　video.　In　this　paper,　we　perform　an　analysis　of　compressed　video.　Based　on　this,　we　propose　utilizing　multi-frame　information　as　input　and　using　two　adjacent　high-quality　frames　to　enhance　the　low-quality　frames　in　between.　Therefore,　an　efficient　Multi-Frame　Quality　Recovery　(MFQR)　method　is　proposed.　In　MFQR,　we　first　develop　a　forward　feature　extraction　module　to　extract　information　from　multi-frame　input.　Then,　a　bidirectional　time　information　extraction　module,　consisting　of　a　Bi-directional　Long-Short　Time　Memory　(Bi-LSTM)　structure　and　channel　attention　mechanism,　is　developed　to　capture　bi-directional　temporal　information　and　enhance　spatiotemporal　information　expression　of　features.　Finally,　we　construct　a　residual　feature　enhancement　module　to　improve　model　performance.　Extensive　experiment　results　show　that　the　proposed　MFQR　method　achieves　an　average　increase　of　27%　in　PSNR　and　reduces　the　number　of　parameters　by　an　average　of　23%　than　the　representative　methods.