Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{188929,
        author = {Prashanth Kumar M and Dr. Mohit Gangwar},
        title = {Feature-Optimized BERT with Aspect-Based Hybrid CNN for Interpretable Emotion Detection},
        journal = {International Journal of Innovative Research in Technology},
        year = {2025},
        volume = {12},
        number = {7},
        pages = {4211-4221},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=188929},
        abstract = {Background / Context: The sudden rise of digital communication and social media platforms generates huge volumes of emotion-rich text data, further presenting new challenges for accurate and interpretable emotion detection. Most of the classic transformer-based systems had been black-box models; therefore, their application in sensitive domains that require transparency in emotional reasoning was limited. It thus inspired the need for models that balance deep contextual understanding with interpretable feature-level insights. Problem/Gap: Most emotion detection systems currently lack an effective mechanism of aspect-level reasoning and are afflicted with redundant transformer embeddings, hence reducing efficiency and interpretability. The previous models also failed in jointly dealing with emotion categories and intensity while providing meaningful explanations. Aim/Objective: The work is motivated by developing a feature-optimized BERT combined with the aspect-based Hybrid CNN to improve the accuracy, efficiency, and interpretability of emotion detection. Methodology/Approach: For this purpose, the proposed framework preprocessed BERT embeddings with feature-selection mechanisms eliminating redundant dimensions while amplifying emotion-salient signals. It used a dependency-based analysis in aspect term extraction and combined the resultant aspect terms into the Hybrid CNN to capture localized emotional cues. Its experiment on the EmoBank dataset tested both emotion classification and VAD intensity prediction performance against a number of baselines. It thus compared performance with standard BERT, ABSA-LSTM, and hybrid deep learning architectures. Results / Findings: These resulted in some striking gains: absolute gain in macro-F1 by 6%–12%, a gain of 4%–9% in accuracy, and significant reductions in the prediction errors of Valence, Arousal, and Dominance. Feature optimization significantly enhanced the clarity of the signal-to-noise ratio and accelerated convergence, while aspect-aware CNN filters encouraged fine-grained emotional reasoning. Indeed, the results of interpretability verified sharper patterns of salience and stronger associations between aspects and emotions. Implications / Significance: hese results showed that emotion detection systems can be both highly accurate and interpretable in a transparent manner without sacrificing efficiency. The creation of practical value in intrinsically emotionally contextual domains involves mental health monitoring, customer experience analytics, and social media intelligence. This work contributes to an explainability-driven approach towards large-scale, next-generation NLP emotion systems.},
        keywords = {Emotion Detection; BERT; Feature Optimization; Hybrid CNN; Aspect-Based Learning; Interpretability; VAD Intensity; Explainable NLP.},
        month = {December},
        }