Abstract: Aiming at the problems of long template iterative update cycle and reduced generalization ability in the previous learning method, A dual-channel sentiment analysis model was proposed based on the improved based on the improved learning method. In the learning method, the serialized templates were firstly introduced into the attention mechanism structure, and meanwhile, the templates were iteratively updated as the input word vectors were updated in the multi-layer attention mechanism. Then the semantic information was extracted by the ALBERT (A Lite BERT) model in another channel, and the extracted semantic features were proposed in an integrated way to improve the generalization ability of the overall model. The model was tested on the SemEval2014 Laptop and Restaurants datasets, the ACL’s (Association for Computational Linguistics) Twitter dataset, and the SST-2 dataset created by Stanford University. The classification accuracy reached 80.88%, 91.78%, 76.78%, 95.53% and compared with the baseline model Bert Large, there are 0.99%, 1.13%, 3.39%, and 2.84% improvements respectively. Compared with the model of the method before improvement, it has 2.88%, 3.60%, and 2.06% improvements on the Restaurants dataset, Twitter dataset, and SST-2 dataset respectively, and reaches the convergence state earlier than the original method.