A production-grade NLP system that dissects Arabic customer reviews — identifying what aspects are mentioned and how the customer feels about each one. Built on MARBERTv2 with semi-supervised learning.
Type or paste any Arabic customer review and watch the model break it down into aspects and sentiments in real-time.
Try an example:
Enter an Arabic review and click Analyze to see the results.
How our system processes Arabic reviews from raw text to structured sentiment predictions.
Normalize Alef/Taa forms, strip diacritics, collapse repeated chars, remove non-Arabic noise. Filter out reviews with <50% Arabic characters.
Pre-trained on 1B+ dialectal Arabic tweets. 12-layer transformer producing 768-dim [CLS] representations. Fine-tuned end-to-end.
Independent linear layers (768→4) per aspect. Softmax forces mutual exclusivity: each aspect gets exactly one of [absent, positive, negative, neutral].
Inverse-frequency class weights handle imbalance. Semi-supervised pseudo-labeling adds 1,418 high-confidence samples from unlabeled data.
Evaluation results on the hidden test set — measured using Micro F1-score.
1,731 Arabic-filtered samples. Early stopping at epoch 7.
Val F1: 78.19%2,972 unlabeled samples → 1,418 high-confidence pseudo-labels (≥90% confidence).
47.7% yield rate3,584 combined samples (train + val + pseudo). 10 epochs, loss: 1.0 → 0.04.
Test F1: 82.8%Distribution of aspects and sentiments across all 500 test predictions.
The 9 aspect categories our model detects in every review.
الطعام
Quality, taste, freshness, portion size
الخدمة
Staff attitude, speed, professionalism
السعر
Value for money, pricing fairness
النظافة
Hygiene, tidiness of the venue
التوصيل
Speed, packaging, accuracy
الأجواء
Atmosphere, decor, noise level
التطبيق
App usability, bugs, interface
عام
Overall impression, recommendation
لا يوجد
No specific aspect mentioned