The helpfulness of online reviews depends on their textual portion. Using the information provided by the seller as a baseline, this study applies latent semantic analysis (LSA) to assess what parts of that textual portion contribute to helpfulness by separating the text into three categories of high entropy words: (1) unique (i.e. does not appear in previous reviews) corroboration entropy, (2) recommendation entropy, and (3) unique opinion entropy. Unique corroboration entropy is calculated based on the number of words in this review that describe the product on the seller's site, confirming the seller's claims, which did not appear in previous reviews. Recommendation entropy is based on the number of words that are associated with explicit recommendations. Unique opinion, referred as “regular opinions” in the literature, entropy is based on the number of all the other words in the review that provide positive or negative evaluations of products as well as other additional informational elements that did not appear in previous reviews. The results show that both recommendation and unique opinion entropies (only marginally) increase review helpfulness evaluations, while greater unique corroboration entropy is insignificant.
All Science Journal Classification (ASJC) codes
- Management Information Systems
- Information Systems
- Information Systems and Management
- Latent semantic analysis
- Online consumer reviews
- Review helpfulness