Boosting Arabic Fake Reviews Detection by Integrating Textual and Metadata Features: A Transformer-Based Model
Bibliographic record
Abstract
Fake reviews present a significant threat to e-businesses and content providers, lowering consumer trust and damaging brand reputation. As such, the detection and prevention of fake reviews is essential for maintaining the integrity and success of e-businesses. On the other hand, the Arabic language presents unique challenges due to its complex linguistic structure and the wide variety of dialects spoken across different regions. However, the availability of Arabic datasets for fake review detection remains limited, where the available ones either suffer from small sample sizes or are translated from English to Modern Standard Arabic, failing to capture the natural, colloquial language typically used in reviews. Moreover, most Arabic fake reviews research has focused mainly on the textual content of the reviews and has not considered the metadata. Therefore, there has been no comprehensive research investigating the benefits and effects of integrating metadata features with textual content for classifying Arabic fake reviews. To this end, this paper is two-fold. Firstly, a balanced Egyptian Arabic dataset has been created, translated from the YelpZip English dataset using transformers, which includes the metadata. Secondly, a comprehensive and comparative study is conducted to investigate the effects of augmenting the textual content with the metadata features. Baseline experiments with textual content only and fine-tuned pre-trained Arabic BERT models achieved an F1-score of around 69%. Combining pre-trained Arabic language model embeddings with handcrafted metadata features significantly boosts performance, with the best-performing system achieving an F1-score of 77% without the user and product IDs as features and as high as 87% with the inclusion of user and product IDs.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.003 |
| Meta-epidemiology (narrow) | 0.002 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.002 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.002 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.002 | 0.003 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".