A corpus linguistics enhanced hybrid deep learning for semantic-driven emotion detection on textual data

Corpus linguistics offers a structured approach to assess huge set of texts (corpora) to uncover language patterns, structural relationships, and semantic interpretations. Within text-based emotion recognition (TER), corpus linguistics facilitates the identification of linguistic features that show underlying emotional expressions. However, the existing machine learning-based models often struggle to learn subtle semantic relationships, contextual relationships, and differences in emotional expression across textual content. Recent advancements in deep learning have demonstrated excellent capabilities for contextual and relational representation learning, yet accurate and robust multi-class emotional classification remains challenging. Therefore, this study proposes a Hybrid Deep Learning Framework for Semantic-driven Emotion Detection (HDLF-SED) on Textual data, aiming to improve learning representation and the accurate separation of multi-class emotional categories. To achieve this, the model uses RoBERTa-Large as a feature embedding technique to learn subtle linguistic patterns and contextual relationships to produce discriminative feature representations for further analysis. The model employs hybrid learning by integrating a temporal convolutional network with a multi-head attention mechanism to jointly capture sequential and local contextual patterns and selectively focus on emotion-relevant features to perform accurate multi-class emotion classification. Furthermore, the RAdam optimization algorithm is used to facilitate more stable and effective optimization during model training and improve classification accuracy. Experimental evaluation is carried out using two benchmark datasets across various performance measures; these results highlight that the proposed HDLF-SED model demonstrates the potential of combining contextual embedding with hybrid learning for more accurate and reliable motion detection and performs multi-class classification.