The impact of e-WOM sentiment on sales: an attribute level analysis of search and experience attributes

The proliferation of e-WOM has transformed how consumers evaluate products before purchase, yet predictive models largely treat online sentiment as a uniform signal, overlooking differences in sentiment intensity and product attribute type. Grounded in Innovation Diffusion Theory (IDT) and Information Economics Theory (IET), this study develops the Attribute-based Sentiment Classification Score (ABSCS) model, which distinguishes between sentiment related to search attributes, evaluable prior to purchase, and experience attributes, assessable only through use. The ABSCS model integrates BERTopic for attribute extraction and a BERT classifier for fine-grained, 5-point sentiment scoring, enabling attribute-level tracking of e-WOM signals over time. Using nearly 3,000 Amazon reviews and 10 million X (Twitter) posts for the PlayStation 5, the ABSCS model improves in-sample explanatory power to 70% compared to 43% in the traditional benchmark model. Critically, experience-attribute sentiment exerts a stronger and more persistent influence on sales than search-attribute sentiment, consistent with IET’s prediction that experiential uncertainty amplifies reliance on peer evaluations. These findings extend IDT by demonstrating that the informational role of e-WOM evolves dynamically across attribute types and adoption stages, and offer practical guidance for reputation management and attribute-targeted communication strategies.