As　an　important　part　of　mobile　phone　auto-generated　animation　technology,　message　sentiment　classifications　is　to　analysis　and　classify　the　emotional　tendency　of　the　message,　provide　emotional　information　for　subsequent　animation　plot　planning　and　animation　scene.　Sentiment　classification　is　divided　into　two　steps:　subjective　and　objective　classification　of　the　message　and　sentiment　tendency　classification　of　subjective　message.　Due　to　single　classifiers　used　directly　for　these　two　perform　low　accuracy　in　Chinese　message　sentiment　classification.　The　aim　of　this　paper　is　to　use　a　heterogeneous　ensemble　of　classifiers　to　improve　the　accuracy,　which　combines　some　different　classifiers　by　ensemble　techniques　instead　of　focusing　on　ensemble　techniques　within　a　classifier.　This　ensemble　method　is　applied　to　a　Chinese　message　corpus　from　manual　collection.　The　results　show　that　the　proposed　heterogeneous　ensemble　approach　yields　higher　accuracy　when　compared　with　using　only　a　single　classifier,　better　meets　the　needs　of　mobile　phone　auto-generated　animation　technology.　©　2015　Taylor　&　Francis　Group,　London.