In　recent　years,　with　the　continuous　increase　of　electronic　medical　record　data,　people　have　begun　to　pay　attention　to　medical　named　entity　recognition.　Medical　named　entity　recognition　can　transform　the　free　text　in　an　electronic　medical　record　from　information　to　data,　so　it　has　high　research　value　and　application　value.　However,　most　of　the　current　deep　learning　methods　use　character-level　segmentation　for　semantic　feature　extraction,　which　leads　to　the　loss　of　local　dependencies　between　characters.　In　order　to　solve　this　problem,　this　paper　proposes　a　multi-word　segmentation　method,　which　mainly　uses　＇jieba＇　to　segment　as　many　sentences　as　possible,　and　then　uses　the　BILSTM　model　to　train　the　words　after　segmentation.　The　purpose　of　this　method　is　to　enrich　the　semantic　features　that　can　be　used　in　the　model　learning　process,　so　as　to　further　improve　the　entity　recognition　ability　of　the　model.　Experimental　results　show　that　this　method　can　improve　the　performance　of　the　model.　©　2022,　The　Author(s),　under　exclusive　license　to　Springer　Nature　Singapore　Pte　Ltd.