In　the　current　memory　network　model,　the　words　of　the　context　are　independent　of　each　other,　and　the　influence　of　word　order　information　on　microblog　sentiment　is　not　taken　into　account.　Therefore,　a　perspective　level　microblog　sentiment　classification　method　based　on　convolutional　memory　network　is　proposed.　In　the　method,　memory　network　can　effectively　model　the　semantic　relation　between　the　query　and　the　text.　Consequently,　the　view　and　the　text　are　abstracted　via　this　property.　Furthermore,　the　word　order　in　context　is　extended　by　convolutional　operation.　Then,　the　result　is　utilized　to　capture　the　attention　signals　of　different　terms　in　context　for　the　weighted　representation　of　text.　Experimental　results　on　three　public　datasets　indicate　that　the　proposed　method　achieves　higher　accuracies　and　Macro-F1　values　compared　with　other　methods.　©　2018,　Science　Press.　All　right　reserved.