<![CDATA[Sentiment analysis has become an important research area in natural language processing (NLP), particularly for analyzing user opinions on social media platforms such as Sina Weibo. Transformer-based models have significantly improved sentiment classification performance, including ChineseBERT, which integrates glyph and pinyin information for Chinese language understanding. However, the performance of transformer models is highly influenced by hyperparameter configurations. Therefore, this study aims to compare the effectiveness of two hyperparameter optimization methods, namely Genetic Algorithm and Optuna, on ChineseBERT for Weibo sentiment classification using the WeiboSenti100k dataset. The experiments were conducted by fine-tuning the ChineseBERT model using baseline hyperparameters and optimized hyperparameters generated by both optimization methods. The optimized parameters include learning rate, dropout rate, batch size, and number of epochs. Model performance was evaluated using accuracy, precision, recall, and F1-score, with a primary focus on classification accuracy. The experimental results show that the baseline ChineseBERT model achieved an accuracy of 97.6%. Meanwhile, ChineseBERT optimized using Genetic Algorithm achieved an accuracy of 97.2%, while ChineseBERT optimized using Optuna achieved the highest accuracy of 98.2%. These results indicate that Optuna provides more effective hyperparameter optimization performance compared to Genetic Algorithm for ChineseBERT-based Weibo sentiment classification. Overall, this study demonstrates that appropriate hyperparameter optimization can improve the performance of transformer-based models for Chinese sentiment analysis tasks.]]>
