Abstract:One text sentiment classifier constructed based on the mechanism of dynamic classification of sample space has been proposed to improve the accuracy of Chinese text sentiment recognition by starting from the perspective of integrated learning. This algorithm makes full use of the identification information within training sample space, makes adaptive classification for sample space by introducing kernel smoothing method, forms several multi-granularity subspaces with differences, and then constructs base classifier in each subspace respectively and finally integrates the output of all base classifiers to produce the final prediction results. Experimental results on Chinese data set have shown that this algorithm is superior to Bagging, Adaboost and other algorithms in precision ratio and recall ratio etc and is also with good application prospect in the sentiment recognition of large-scale sample set.