DP-ROBERTa: Dual-Perspective Emoji-Aware BERT Integrated with Semantic Scoring and Swarm-Scaled Sentient LSTM-gCNN for Twitter Sentiment

Social media sentiment analysis is of critical importance in understanding public opinion, brand sentiments and societal reactions in real time; however, sentiment detection from Twitter accurately is challenging because of sarcasm, emojiinduced polarity changes, adversarial manipulation and semantic ambiguity. Existing transformer-based models mainly focus on textual representations and poorly model the process of affective emoji-text interaction, and therefore make unstable predictions under noisy conditions. This work is an execution of DPROBERTa, a novel Dual-Perspective Emoji-Aware RoBERTa model with integrated Semantic Content Scoring and SwarmScaled Sentient LSTM-gCNN for robust Twitter sentiment classification. The methodology is a joint linguistic context and emoji based affective signals using dual attention learning, prediction confidence calibration using Semantic Content Score, and swarm intelligence hyper parameter optimization with adversarial emoji perturbation defense. Extensive experiments on benchmark datasets of large corpora of tweets are shown, and it is found that DP-ROBERTa achieves the highest accuracy of 92.4%, Macro-F1 of 91.8%, and AUC of 96.2%, which consistently outperforms 5 state-of-the-art transformer models. The results validate the significant improvement in robustness, generalization and reliability in sentiments obtained using the proposed framework making it applicable for real-world applications of social media intelligence.