A Comparative Study of Deep Learning and Transformer Models for Twitter Sentiment Analysis

In this work, we investigate sentiment classification on Twitter using the Sentiment140 dataset and compare traditional deep learning approaches, including CNN, BiLSTM, and GRU, with a lightweight transfer learning model, DistilBERT. The results indicate that the CNN, BiLSTM, and GRU models achieve accuracy levels ranging from 71% to 79%, whereas the fine-tuned DistilBERT model attains an accuracy of 86% with an F1-score of 0.86. These findings suggest that, while conventional neural models tend to focus on more local patterns, lightweight transformer-based architectures are better able to capture[HEDGING-END] broader linguistic representations, while still maintaining relatively efficient computational performance. Overall, this study offers a systematic and practical comparison of sentiment classification models, supported by a clearly defined methodology, empirical results, and corresponding tables and figures. The analysis may provide useful guidance for researchers and practitioners interested in sentiment analysis applications for social media data. Smart citations: https://scite.ai/reports/10.61467/2007.1558.2026.v17i1.1241 Dimensions.Open Alex.