Intelligent Feedback Analytics- Sentiment Analysis and Defect Detection in Elevating Product Quality and Customer Experience

• Published in: Communications and Signal Processing, International Conference on pp. 954 - 959
• Main Authors: Padoor, Veda, P, Tharun Kumar, Pani, Y L Krishna, Bhalke, Vaibhavi, Kasturi, Nivedita
• Format: Conference Proceeding
• Language: English
• Published: IEEE 05.06.2025
• Subjects: Accuracy; BERT Multilingual Model; Defect Prioritization; Feature extraction; OpenAI Whisper; Organizations; Product design; Quality assessment; Sentiment analysis; Signal processing; spaCy; Systematics; Technological innovation; Text analysis; Zero Shot Classification
• ISSN: 2836-1873
• DOI: 10.1109/ICCSP64183.2025.11088646

The competitiveness of the market requires customer feedback to achieve product success and innovation. This study presents a system for automatic product feedback analytics which applies high-tech integration methods to convert customer feedback into usable insights. A system utilizes voice-to-text conversion models to convert audio feedback into text which achieves high accuracy rates throughout the process except when there are some exceptions. The system implements text analysis methods together with sentiment analysis and keyword extraction and theme identification processes for value insight extraction. Organizations should prioritize defect classification and prioritization strategies as their essential foundation because this analysis method uses sentiment and context to identify which defects require immediate attention. This integrated and adaptable method facilitates organizations to utilize customer data systematically for impactful results while making data-based choices regarding product enhancement. Based on its use of the BERT model the system achieves sentiment analysis accuracy at 94.7 percent thereby reaching 94.7 percent precision, recall, and F1 Score benchmark. The integrated system utilizes BERT for negativeness intensity while employing spaCy dependency parsing and zero-shot classification to extract features and classify defects which results in accurate operational outcomes with 80 percentage success rates and precision and recall scores of 95 percentage. Previous research indicates that the proposed framework successfully generates highly capable results for productive insights which can be quantified through existing metrics.