The　colloquialism　and　conciseness　of　microblog　text　bring　additional　challenges　to　emotion　classification.　This　paper　proposes　a　new　emotion　classification　model　based　on　hybrid　learning.　In　the　first　stage,　the　improved　dictionary　classification　method　is　used　to　calculate　emotion　score　in　the　whole　data　set,　and　the　data　with　high　or　low　scores　are　directly　marked;　in　the　second　stage,　the　rest　of　the　method　is　based　on　emotion　dictionary　and　Bi-GRU　fusion　model　to　calculate　emotion　score,　and　the　two　stage　hybrid　frame　makes　the　method　effectively　applied　in　microblog　emotion　classification.　The　microblog　experiment　with　emoticons　in　COAE2014　(Chinese　opinion　analysis　and　evaluation　2014)　dataset　shows　that　the　single　model　is　not　ideal　for　many　kinds　of　complex　contexts,　and　it　is　difficult　and　low　accuracy.　The　multi　model　fusion　method　can　effectively　improve　the　error　preference　of　the　single　model　and　improve　the　classification　effect.　©　2020　IEEE.