Modeling of corpus linguistics with semantic-aware deep neural decision forest for enhanced emotion detection from textual data

Emotion detection from textual data plays an important role in understanding the emotional content expressed through written language. Moreover, the natural language processing (NLP) provides useful techniques for analyzing textual content and identifying meaningful linguistic patterns associated with different emotions. It helps to recognize different emotional states from the words and context present in a text. However, traditional machine learning approaches are often relied on manually designed features and may not capture the underlying semantic relationships effectively. Deep learning models can learn representations automatically, but distinguishing between closely related emotions remains challenging. To address these limitations, this paper introduces a Deep Neural Decision Forest-based Framework for Semantic-Driven Emotion Detection (DNDF-SED) using textual data. The proposed model first preprocesses the text through cleaning, contraction expansion, lemmatization, negation handling, and subword tokenization to obtain clean and consistent text for subsequent word representation process. Then, DeBERTa-v3 is utilized to learn contextual and semantic representations from the processed text. These representations are given to a Deep Neural Decision Forest classifier, which integrates deep feature learning with decision-tree-based classification to distinguish different emotion categories. Moreover, NAdam is applied to optimize the model parameters and enhance the learning process. Comprehensive experimental outcomes demonstrate the effectiveness of the proposed DNDF-SED framework, achieving greater accuracy of 95.70% on Emotions dataset for NLP and 98.71% on Emotion Dataset Raw compared with existing methodologies. Additionally, the ablation study confirms the contribution of every component by showing improved classification performance with the proposed methodology. These results demonstrate the effectiveness of integrating contextual language representations with deep decision forest classification for semantic-driven emotion detection.