Categorizing customer perceptions of sustainable product packaging: Analyzing online reviews using large language models

초록

Online reviews provide rich insights into customer experiences, yet their unstructured nature limits conventional text-mining to surface-level analyses. This study leverages a large language model (Meta AI-based LLaMA) to analyze customer feedback on sustainable cosmetic packaging and examine satisfaction formation mechanisms through the Kano model, using LLaMA generated construct and sentiment scores from Amazon reviews for hypothesis testing. The results reveal a hierarchical pattern: one-dimensional attributes exhibit the strongest positive association with satisfaction, followed by must-be and attractive attributes, whereas indifferent and reverse attributes show comparatively weak relationships. Interaction analyses demonstrate configurational patterns: baseline compliance strengthening performance and innovation cues, while performance signals buffer trade-off perceptions. Sentiment analysis shows stronger negative reactions to reverse attributes and stronger positive responses to one-dimensional and attractive attributes. Moderation analyses indicate variations across product types. The study advances AI-enabled marketing analytics by deriving scalable sustainability insights from customer reviews about product packaging.

키워드