A Hybrid Transformer and Lexicon-Based Ensemble Model for Sentiment Analysis of DeepSeek AI Tweets with Data Augmentation

This study investigated real-time public sentiment to inform the future development of AI chatbots.Specifically, it examined public attitudes toward the DeepSeek AI chatbot by analyzing Twitter data.A hybrid sentiment analysis framework was employed, combining lexicon-based models such as valence aware dictionary and sentiment reasoner (VADER) and TextBlob with a transformer-based model such as bidirectional encoder representations from transformers (BERT).This ensemble approach leverages the strengths of each method to enhance sentiment classification accuracy, particularly in the context of noisy, informal social media text.Recognizing the typical bias in social media datasets, which are often skewed toward positive sentiments, this study implemented data augmentation techniques, such as synonym replacement and contextual paraphrasing, to address class imbalance and ensure a more representative training set.By integrating multiple models and visual analytics, this study enhanced the interpretability of sentiment trends, empowering non-technical stakeholders to make informed decisions regarding AI deployment.Notably, the ensemble-based random forest (RF)+Word2Vec (W2V) model using hybrid annotations achieved a 15.66% improvement in accuracy over baseline transformer-based approaches.