Optimizing sentiment analysis in product descriptions: effects on customer purchase intentions

Abstract This study investigates the role of product descriptions in reducing information asymmetry on e-commerce platforms, particularly for high-priced items. Utilizing a text analytic approach, we employ a novelty method called “SATORE,” where we used the Latent Dirichlet Allocation (LDA) model to extract topics and determine the optimal number using the Renyi Entropy technique. Sentiment scores were calculated based on the topic information for each product and integrated into a logistic regression model to assess their influence on buyers’ purchase intentions. The results indicate that the sentiment score calculated by SATORE significantly affects purchase intention and helps reduce information asymmetry. The effects of sentiment scores vary with product prices, with higher-priced items having a more substantial impact. In addition, our method proves to be more robust than word-frequency-based sentiment scores, which lose significance when prices exceed ¥100,000. These findings benefit platform owners by helping them maintain improved buyer-seller relationships.