Sentiment analysis is a core task in natural language processing， involving the assessment of emotions or sentiment expressed within texts. In the current research on sentiment analysis， most models rely on bidirectional encoder representations from transformers （BERT） as a feature extractor， focusing mainly on relatively simple binary or ternary tasks. To address fine - grained sentiment classification， the paper introduces a new hybrid dual-channel gated recurrent unit and convolutional neural network （GRU - CNN） sentiment analysis model（GGC）. This model uses generative pre-trained transformer （GPT） as a feature extractor， capturing the deeper meanings in the text more precisely. Based on this， the text features extracted are fed into multi - channel GRU and CNN， capturing both global and local features respectively. The model also incorporates an attention mechanism， which dynamically fuses these two types of features. This mechanism allows the model to allocate different weights to different parts according to their importance， thus capturing key emotional information in the text more accurately. Experimental results show that this method achieves excellent performance in sentiment analysis tasks.