Gated Soft-Hard Attention Enabled Deep Learning Framework for Sentiment Analysis in Social Media Data
Jay Narayan Thakre, Neelu Nihalani · 2025
Sentiment analysis examines the emotions of users from the interpretations posted on social media such corresponding to various polarities. The research focused on text sentiment analysis was inclined to numerous complications such as the inability to handle large amounts of data, higher computational demands, feature losses, complexity in capturing semantic tones, as well as non-consideration of subtle nuances, which highly impacts the detection accuracy. Hence, this research proposes a Gated Soft-hard attention-enabled deep recurrent neural network (GSH-DRNN) framework that addresses the complexities in sentiment detection with higher accuracy. The research designs a Gated Soft-hard attention (GSHA) mechanism that facilitates the model to focus on crucial portions of the sentence by providing more emphasis on the overall phrases. Further, the GSH-DRNN model effectively captures the ambiguity and identifies the polarities via the feature extraction phase. Moreover, focusing on the contextual nuances improves the efficiency of detection while disregarding the texts with lower weights. Furthermore, the efficiency of GSH-DRNN is compared with the prevailing sentiment detection approaches that reveal superior outcomes with a higher precision of$\mathbf{9 6. 8 4 \%, ~} \mathbf{9 6. 4 5 \% ~ F 1}$score, 96.06 % recall, and 96.45 % accuracy.