From: Text-rating review discrepancy (TRRD): an integrative review and implications for research
Paper | Used Model | Domain | (Manually labelling) Linking between Experts’ ratings and available rating scores were checked? | (Automatic labelling) Linking between lexicon-based review ratings and available rating scores were checked? | Degree of consistency between text reviews and rating scores |
---|---|---|---|---|---|
Zhu et al. [68] | Regression Model | Tourism and hospitality (Airbnb) | No | Yes Tobit Models | Coefficients = 0.3072 for positive and –4.2846 for negative Degree: Weak for positive |
Li [29] | Statistical Model | Airlines | No | Yes “Semantria” Kendall’s tau and the Spearman’s rho Correlation | Kendall’s tau = 0.207 and the Spearman’s rho = 0.268 Degree: Weak |
Geetha et al. [18] | Naive Bayes Classifier | Hotels | No | Yes Linear Regression Model | A linear relationship between customer sentiment and rating R square has a value of 0.21. So, 21% of the variation in the customer ratings is explained by customer sentiment polarity Degree: Weak |
Tsang And Prendergas, 2009 [60] | Statistical Study | Movies | No | Yes Analysis of covariance (ANCOVAs) | A significant interaction between text and rating valences on trustworthiness |
Ganu et al. [16] | Support Vector Machine classifiers (sentiment classification) Regression (numeric scores) | Restaurants | Yes | Yes Private lexicon done by researchers Percentage Analysis | 56% of the reviews were annotated as positive and 18% as negative. The associated numeric ratings provided by users pointed that 73% of reviews having positive rating Degree: Average |