app.py

import gradio as gr
import numpy as np
import torch
from transformers import BertTokenizer, AutoModel
import tensorflow as tf

model_name = "aubmindlab/bert-base-arabertv02"
tokenizer = BertTokenizer.from_pretrained(model_name)
bert_model = AutoModel.from_pretrained(model_name)
bert_model.eval()

model = tf.keras.models.load_model("rnn_Bi.h5")
print("✅ Model loaded successfully!")

def get_bert_embedding(text, max_length=100):
    inputs = tokenizer(text, return_tensors="pt", padding=True, truncation=True, max_length=max_length)
    with torch.no_grad():
        outputs = bert_model(**inputs)
    embedding = outputs.last_hidden_state[:, 0, :].numpy()
    embedding = embedding.reshape(1, 1, 768)
    return embedding

def predict_sentiment(text):
    embedding = get_bert_embedding(text)
    pred = model.predict(embedding)[0][0]
    label = "إيجابي" if pred > 0.5 else "سلبي"
    confidence = pred if pred > 0.5 else 1 - pred
    return label, confidence

css = """
body {
    font-family: 'Segoe UI', Tahoma, Geneva, Verdana, sans-serif;
    direction: rtl;
    text-align: right;
}
.gradio-container {
    max-width: 600px;
    margin: auto;
    background: #f9f9f9;
    padding: 20px;
    border-radius: 12px;
    box-shadow: 0 6px 15px rgb(0 0 0 / 0.1);
}
.gr-button {
    background-color: #007bff !important;
    color: white !important;
    font-weight: bold;
    border-radius: 8px !important;
    padding: 10px 25px !important;
    font-size: 16px !important;
}
.gr-textbox textarea {
    font-size: 18px !important;
    padding: 10px !important;
}
"""

with gr.Blocks(css=css) as iface:
    gr.Markdown("## تحليل المشاعر بالعربية 📝", elem_id="title")
    gr.Markdown("أدخل جملة لتحليل المشاعر: هل هي **إيجابية** أم **سلبية**؟", elem_id="description")
    text_input = gr.Textbox(lines=3, placeholder="اكتب جملتك هنا...", label="النص")
    predict_btn = gr.Button("تنبؤ")
    sentiment_output = gr.Label(label="النتيجة")
    confidence_score = gr.Textbox(label="نسبة الثقة")

    def on_predict(text):
        label, confidence = predict_sentiment(text)
        confidence_percent = f"{confidence*100:.1f}%"
        return label, confidence_percent

    predict_btn.click(fn=on_predict, inputs=text_input, outputs=[sentiment_output, confidence_score])

if __name__ == "__main__":
    iface.launch()