Initial deploy

README.md

@@ -1,14 +1,56 @@
 ---
-title: Arabic Complaints Classifier
-emoji: 🚀
-colorFrom: indigo
-colorTo: green
+title: Arabic Restaurant Complaints Classifier
+emoji: 🍽️
+colorFrom: green
+colorTo: blue
 sdk: gradio
-sdk_version: 6.14.0
-python_version: '3.13'
+sdk_version: 4.44.0
+python_version: "3.11"
 app_file: app.py
-pinned: false
+pinned: true
 license: mit
+short_description: Classify Arabic restaurant complaints into 8 categories
+tags:
+  - arabic
+  - nlp
+  - classification
+  - bert
+  - saudi
+  - dialectal-arabic
+models:
+  - CAMeL-Lab/bert-base-arabic-camelbert-mix
+  - UBC-NLP/MARBERT
+  - aubmindlab/bert-base-arabertv02
+language:
+  - ar
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# Arabic Restaurant Complaints Classifier
+
+Classify Arabic restaurant complaints into 8 actionable categories. Saudi-Gulf dialect specialization.
+
+## Performance
+
+| Metric | Value |
+|---|---:|
+| Test accuracy | 95.05% |
+| Test weighted F1 | 95.08% |
+| Test macro F1 | 92.03% |
+| Min class F1 (عامة) | 84.84% |
+
+## Categories
+
+| Arabic | English |
+|---|---|
+| التوصيل | Delivery |
+| السعر والقيمة | Price / value |
+| النظافة | Cleanliness |
+| جودة الطعام | Food quality |
+| خدمة الموظفين | Staff service |
+| دقة الطلب | Order accuracy |
+| عامة | General |
+| وقت الانتظار | Wait time |
+
+## Source
+
+https://github.com/FerasMad/NLP-complaints-system

app.py

@@ -0,0 +1,109 @@
+"""HuggingFace Spaces entry point — Arabic Restaurant Complaints Classifier.
+
+Loads the single best CAMeLBERT-mix model from HuggingFace Hub and serves
+a Gradio UI. Set HF_REPO_ID env var in the Space settings to point at
+your model repo.
+"""
+import os
+import re
+
+import gradio as gr
+import torch
+from transformers import AutoModelForSequenceClassification, AutoTokenizer
+
+HF_REPO_ID = os.environ.get(
+    "HF_REPO_ID", "FerasMad/arabic-complaints-classifier"
+)
+MAX_LENGTH = 192
+MIN_ARABIC_RATIO = 0.30
+
+CATEGORIES = [
+    "التوصيل",
+    "السعر والقيمة",
+    "النظافة",
+    "جودة الطعام",
+    "خدمة الموظفين",
+    "دقة الطلب",
+    "عامة",
+    "وقت الانتظار",
+]
+ID2LABEL = dict(enumerate(CATEGORIES))
+
+TASHKEEL = re.compile(r"[ً-ٟ]")
+NON_ARABIC = re.compile(r"[^؀-ۿa-zA-Z0-9٠-٩\s]")
+WHITESPACE = re.compile(r"\s+")
+ARABIC_CHAR = re.compile(r"[؀-ۿ]")
+
+
+def clean(text: str) -> str:
+    if not text:
+        return ""
+    t = TASHKEEL.sub("", text)
+    t = t.translate(str.maketrans({"أ": "ا", "إ": "ا", "آ": "ا", "ٱ": "ا", "ى": "ي", "ة": "ه"}))
+    t = NON_ARABIC.sub(" ", t)
+    return WHITESPACE.sub(" ", t).strip().lower()
+
+
+def is_arabic_enough(text: str) -> bool:
+    if not text or len(text) < 3:
+        return False
+    return len(ARABIC_CHAR.findall(text)) / max(len(text), 1) >= MIN_ARABIC_RATIO
+
+
+print(f"Loading {HF_REPO_ID} ...")
+device = "cuda" if torch.cuda.is_available() else "cpu"
+tokenizer = AutoTokenizer.from_pretrained(HF_REPO_ID)
+model = AutoModelForSequenceClassification.from_pretrained(HF_REPO_ID).to(device).eval()
+print(f"Model loaded on {device}.")
+
+
+@torch.no_grad()
+def predict(text: str):
+    if not text or len(text.strip()) < 3:
+        return {"النص قصير جدا — please type a longer Arabic complaint": 1.0}
+    if not is_arabic_enough(text):
+        return {"النص ليس باللغه العربيه — please use Arabic input": 1.0}
+
+    enc = tokenizer(clean(text), return_tensors="pt", truncation=True, max_length=MAX_LENGTH).to(device)
+    probs = torch.softmax(model(**enc).logits[0], dim=-1).cpu().numpy()
+    top_idx = probs.argsort()[::-1][:3]
+    return {ID2LABEL[int(i)]: float(probs[i]) for i in top_idx}
+
+
+EXAMPLES = [
+    "وصل الطلب بارد جدا والمندوب تاخر اكثر من ساعتين",
+    "الاسعار مبالغ فيها لا تناسب الجوده المقدمه ابدا",
+    "النظافه سيئه الطاولات متسخه والارض غير نظيفه",
+    "طلبت برجر بدون بصل لكنهم وضعوه رغم تنبيهي",
+    "انتظرت ساعه كامله في المطعم قبل ان ياتي طلبي",
+    "الموظف اسلوبه سيء جدا وغير محترم",
+    "الاكل بايخ ومالح والطبخ مو متقن",
+    "تجربه سيئه عموما لن اعود لهذا المكان",
+]
+
+
+demo = gr.Interface(
+    fn=predict,
+    inputs=gr.Textbox(
+        lines=4,
+        label="اكتب الشكوى",
+        placeholder="مثال: الاكل بايخ ومالح والطبخ مو متقن",
+        rtl=True,
+    ),
+    outputs=gr.Label(num_top_classes=3, label="التصنيف"),
+    examples=EXAMPLES,
+    title="تصنيف شكاوى المطاعم العربية — Arabic Restaurant Complaints Classifier",
+    description=(
+        "نموذج CAMeLBERT-mix مدرب على ٨ فئات من الشكاوى. لهجة سعودية / خليجية. "
+        "Fine-tuned CAMeLBERT-mix · 8 categories · 95% test accuracy · Saudi-Gulf dialect."
+    ),
+    article=(
+        "**Source:** [github.com/FerasMad/NLP-complaints-system]"
+        "(https://github.com/FerasMad/NLP-complaints-system)"
+    ),
+    allow_flagging="never",
+)
+
+
+if __name__ == "__main__":
+    demo.launch(server_name="0.0.0.0", server_port=7860)

requirements.txt

@@ -0,0 +1,6 @@
+torch>=2.0
+transformers>=4.40,<5
+huggingface_hub>=0.24,<1.0
+sentencepiece>=0.2
+gradio==4.44.0
+numpy>=1.24