LLMs Excel in Implicit Sentiment Analysis for Product Desirability

Understanding implicit sentiment from qualitative product feedback is crucial for product development and marketing, yet it often proves challenging to measure effectively. This research proposes a novel framework that leverages large language models (LLMs) to provide a scalable and interpretable solution for quantifying product desirability from such nuanced user experiences. The framework was evaluated using two Product Desirability Toolkit (PDT) datasets, which contained 106 respondent term groupings with expert human annotations. The study assessed the LLMs' ability to perform zero-shot continuous numerical sentiment scoring and categorical sentiment classification without relying on explicit review scores. The results were highly promising: LLMs generated numerical sentiment scores that closely matched expert labels, achieving Pearson correlations up to 0.97, and classification accuracy reached up to 94%. Notably, the LLMs demonstrated robustness across various data formats and consistently expressed high confidence in their assessments. In contrast, traditional lexicon-based and transformer baselines failed to produce statistically significant results. Among the models tested, GPT-4o-mini stood out by delivering performance comparable to much larger models at a remarkable 94% lower cost, making it ideal for scalable deployment. The framework also incorporates model confidence ratings and human-readable rationale explanations (xAI), significantly enhancing interpretability, transparency, and trust for practical product satisfaction assessment.