With　the　rapid　development　of　the　Internet,　social　networks　have　become　an　important　platform　for　people　to　express　their　opinions.　Through　the　sentiment　analysis　of　the　text,　the　user＇s　emotional　tendency　can　be　captured　in　time.　Multi-label　sentiment　classification　is　one　of　the　difficult　issues.　Therefore,　the　paper　proposes　a　multi-label　sentiment　classification　model　for　short　texts　of　Weibo　based　on　multiscale　CNN.　Firstly,　aiming　at　the　insufficiency　of　multi-label　emotional　corpus,　we　give　a　method　to　construct　a　multi-label　emotional　corpus　based　on　emotional　seed　words,　which　uses　synonym　forest　and　TF-IDF　weights　to　automatically　label　the　corpus.　Then　a　multi-label　sentiment　analysis　model　based　on　multi-scale　CNN　is　proposed,　which　uses　different　convolution　scales　to　extract　features　of　text.　Finally　we　found　that　the　accuracy　of　the　method　has　been　significantly　improved.　©　2020　IEEE.