Research on the Construction of a Fine-Grained Sentiment Analysis Algorithm for Text Content Based on Pre-Trained Models

To address the limitations of traditional sentiment analysis in modeling fine-grained semantics, a pre-trained model with multi-task joint learning is constructed to unify sentiment unit recognition and polarity classification. The proposed architecture is optimized and evaluated on SemEval and LAPTOP-REST datasets. Experimental results demonstrate superior performance in F1-score and accuracy, with improved semantic understanding and generalization, offering a high-precision solution for sentiment analysis.