license: cc-by-4.0
AFRD: Arabic Fake Reviews Detection dataset
Description
Arabic Fake Reviews Detection (AFRD) is the first gold-standard dataset comprised of three domains, namely, hotel, restaurant, and product domains. Each domain has a set of attributes, the reviewer’s age, the reviewer’s gender, the service name, the review’s text, the rating, the text’s polarity, and the review’s class. The overall balanced dataset is consisted of 1728 reviews, 310 reviews for the hotel domain, 714 reviews for the restaurant domain, and 704 reviews for the product domain, the two classes in each domain are balanced. However, there are unbalanced version with 1958 reviews. The following table demonstrate the number of reviews in each class for the balanced dataset:
| Domain | Fake class | Truthful class | Total |
|---|---|---|---|
| Hotel | 155 | 155 | 310 |
| Restaurant | 357 | 357 | 714 |
| Product | 352 | 352 | 704 |
| Multi-domain | 864 | 864 | 1728 |
Moreover, the review sentiment is balanced in each class. Following figure shows how the negative and positive reviews are balanced:
For more information refer to the paper:
Multiscale cascaded domain-based approach for Arabic fake reviews detection in e-commerce platforms
Citation
Please cite the following paper if you used the dataset:
Qandos, N., Hamad, G., Alharbi, M., Alturki, S., Alharbi, W., & Albelaihi, A. A. (2024). Multiscale cascaded domain-based approach for Arabic fake reviews detection in e-commerce platforms. Journal of King Saud University-Computer and Information Sciences, 101926.