Customers share their opinions about a product through online reviews. Companies incorporate customer opinions into product design to increase customer satisfaction and market success. Many studies have attempted to analyze customer opinions on specific product features. However, these studies do not attempt to understand the actual intent of customers when mentioning a product feature in the context of a review. To overcome this limitation, this study develops a contextual meaning-based approach that considers the contextual meanings of product features in reviews. This approach enables a deeper understanding of the intent behind the target features of a product. First, a large language model-based word embedding model and clustering algorithms are introduced to divide product features into sub-features based on contextual meanings. Second, a new method is developed for creating a contextual word map to interpret the clustering results. Third, a sentiment analysis is performed to evaluate customer satisfaction for each sub-feature using BERT. A case study of a television product was conducted to demonstrate the applicability of the developed approach. The results showed that, unlike in other studies, several sub-features were identified from four targeted TV features based on their contextual meanings. Furthermore, customer satisfaction was evaluated for each sub-feature. This study is the first attempt at providing a fine-grained online product review analysis based on unsupervised learning to clarify meanings according to the context. The developed approach can be useful for determining detailed directions for improvement in the product design process and is expected to offer a new perspective.INDEX TERMS Bidirectional encoder representations from transformers (BERT), contextual meaning, customer opinion, online product review, polysemy word, product design, product feature, sentiment analysis, sub-feature identification, television (TV).