A blended sentiment analysis, large language model, and multi-criteria decision-making (BSLMM) approach to sustainable product innovation

Owing to the growing popularity of electronic marketplaces, individual online retailers must evaluate a large volume of customer text reviews to gather feedback about consumer preferences and support sustainable product evaluation. Product evaluation involves multiple product attributes simultaneously, making it a multi-criteria decision-making problem. While traditional sentiment analysis captures isolated customer insights, this approach is subject to three key limitations. These are reliance on context-blind extraction methods for attribute identification, subjective weighting schemes that cannot handle correlated criteria, and insufficient treatment of linguistic uncertainty arising from inter-reviewer variability in sentiment intensity. To address this, this study introduces BSLMM (BERT-based Sentiment analysis with Language Models and Multi-Criteria Decision Making), a systematic framework that integrates BERT for contextual attribute extraction and aspect-level sentiment scoring, Principal Component Analysis (PCA) for transforming correlated attributes into independent criteria with objectively derived weights, and fuzzy COPRAS for ranking product alternatives. This study contributes a data analytics framework that blends sentiment analysis, pre-trained language models and MCDM for product benchmarking and development. Furthermore, it offers practitioners a low-cost data-driven approach to competitive product development towards sustainable innovation.