From: Text-rating review discrepancy (TRRD): an integrative review and implications for research
Paper | Datasets | Incorporating both users’ numeric ratings and users’ text reviews? |
---|---|---|
Yu et al. [67] | Amazon Review dataset (Arts, Jewelery, Watches, Cell Phones and Accessories, etc.) | Yes They mapped between aspects sentiments in review texts and rating scores to better rating predication |
Ling et al. [31] | Amazon Review dataset (Arts, Jewelery, Watches, Cell Phones and Accessories, etc.) | Yes They applied topic modelling techniques on the review text and aligned the topics with rating dimensions to improve prediction accuracy. They were able to improve the accuracy over existing strong baseline methods, that use only rating for recommendations especially under the cold start problem when the data is extremely sparse |
McAuley and Leskovec [35] | Amazon Review dataset (e.g. Books, Movies) + pub data from ratebeer.com + restaurant data from citysearch.com,  + Yelp dataset | Yes They proposed a model that works by aligning hidden factors in product ratings with hidden topics in product reviews. The proposed model allows to accurately fit user and product parameters with only a few reviews, which existing models cannot achieve using only a few ratings |