Named　entity　recognition　is　an　important　and　basic　work　in　text　mining.　To　overcome　the　shortcomings　of　existing　multi-domain　named　entity　recognition　methods,　a　multi-domain　named　entity　recognition　method　based　on　the　part-of-speech　attention　mechanism,　called　BiLSTM-ATTENTION-CRF,　was　proposed　in　this　paper.　The　domain　dictionary　was　constructed　to　represent　multi-domain　semantic　information　and　the　BiLSTM　network　was　used　to　capture　the　grammatical　and　syntactic　features,　as　well　as　multi-domain　semantic　features　in　context　information.　A　part-of-speech　attention　mechanism　was　designed　to　obtain　the　contribution　weight　of　part-of-speech　for　entity　recognition.　Finally,　a　group　of　experiments　were　performed　on　the　multi-domain　dataset　to　compare　various　fusion　strategies　of　multi-level　entity　information.　The　experimental　results　show　that　BiLSTM-ATTENTION-CRF　has　a　high　precision　and　recall　rate,　and　can　effectively　recognizes　the　multi-domain　named　entities.