In order to solve the problem of insufficient single word features in English text sentiment analysis, a dual-channel convolutional neural network model (DC-CNN) that integrates word channel and phrase channel is constructed to improve the accuracy of sentiment binary classification by mining local and overall semantic features. This paper preprocesses the SST-2 (Stanford Sentiment Treebank) dataset, including cleaning and word segmentation. Pre-trained word vectors are used to construct word channels, and n-gram is used to extract phrases to construct phrase channels. Multi-scale convolution kernels are used to extract local semantics, and maximum pooling and self-attention are combined to enhance key information. Finally, the features of the two channels are fused and input into the fully connected layer, and Softmax is used to achieve binary classification. Experimental results show that the model achieves an accuracy of 0.988 on the SST-2 dataset, which is better than the existing mainstream models and verifies the effectiveness of multi-granularity feature fusion. The DC-CNN model has excellent generalization and robustness, providing a new method for English text sentiment analysis.