With the rise of Internet medicine, the amount of drug review data has exploded. The effective analysis of this data is crucial for drug research and development, clinical decision-making, and patient medication guidance. This study focuses on solving the problem of the accuracy of drug review classification. Given the limitations of traditional text classification methods in handling complex semantics and long-text dependencies, an innovative model architecture integrating BERT, Convolutional Neural Network (CNN), multi-head attention mechanism, and GBDT is proposed. Through the preprocessing and in-depth analysis of a large-scale real drug review dataset, this model can accurately classify reviews and demonstrates significantly better performance than existing methods on multiple evaluation metrics. It provides an efficient and reliable solution for the intelligent processing of drug review data and is expected to promote the further development of the field of pharmacoinformatics.