app.py

# ============================================================
# 🌍 Multi-Lingual Sentiment Analysis (English + Persian)
# With SHAP Interpretability
# ============================================================

import gradio as gr
import joblib
import numpy as np
import shap
import matplotlib.pyplot as plt
import os

# ------------------------------------------------------------
# 1️⃣ Load Pretrained Models and Vectorizers
# ------------------------------------------------------------
english_model = joblib.load("best_model.pkl")
english_vectorizer = joblib.load("tfidf_vectorizer.pkl")

persian_model = joblib.load("logistic_regression.pkl")
persian_vectorizer = joblib.load("tfidf_vectorizer_persian.pkl")

# Define class labels
english_labels = ["Negative", "Neutral", "Positive"]
persian_labels = ["منفی", "خنثی", "مثبت"]

# ------------------------------------------------------------
# 2️⃣ SHAP Visualization Function
# ------------------------------------------------------------
def get_shap_plot(model, vectorizer, text, class_index, class_name):
    X_input = vectorizer.transform([text])
    explainer = shap.Explainer(model, vectorizer.transform([" ".join(text.split()[:50])]))
    shap_values = explainer(X_input)
    shap_for_class = shap_values.values[0][:, class_index]
    feature_names = np.array(vectorizer.get_feature_names_out())

    top_idx = np.argsort(-np.abs(shap_for_class))[:10]
    top_words = feature_names[top_idx]
    top_impacts = shap_for_class[top_idx]

    plt.figure(figsize=(6, 3))
    colors = ["crimson" if v > 0 else "steelblue" for v in top_impacts]
    plt.barh(top_words, top_impacts, color=colors)
    plt.title(f"Top Words driving {class_name} prediction")
    plt.xlabel("SHAP Value (Impact)")
    plt.gca().invert_yaxis()
    plt.tight_layout()
    plt.savefig("shap_plot.png", bbox_inches='tight')
    plt.close()

    return top_words.tolist(), "shap_plot.png"

# ------------------------------------------------------------
# 3️⃣ Prediction + Interpretability Function
# ------------------------------------------------------------
def predict_sentiment(text, language):
    if not text.strip():
        return "Please enter a comment.", None

    if language == "English":
        model, vectorizer, labels = english_model, english_vectorizer, english_labels
    else:
        model, vectorizer, labels = persian_model, persian_vectorizer, persian_labels

    X_input = vectorizer.transform([text])
    probs = model.predict_proba(X_input)[0]
    pred_idx = np.argmax(probs)
    pred_class = labels[pred_idx]
    conf = probs[pred_idx]

    # SHAP interpretation
    top_words, shap_plot = get_shap_plot(model, vectorizer, text, pred_idx, pred_class)

    # Final output
    explanation = f"""
**Predicted Sentiment:** {pred_class}  
**Confidence:** {conf:.2f}  
**Top Influential Words:** {', '.join(top_words)}
"""
    return explanation, shap_plot

# ------------------------------------------------------------
# 4️⃣ Gradio Interface
# ------------------------------------------------------------
title = "🌐 Multi-Lingual Sentiment Analysis (English + Persian)"
description = """
Select a language, type a comment, and see both the sentiment prediction and SHAP interpretability.
"""

examples = [
    ["I love this product! Highly recommend.", "English"],
    ["Worst experience ever, totally disappointed.", "English"],
    ["The service was okay, nothing special.", "English"],
    ["این محصول فوق‌العاده است", "Persian"],
    ["تجربه‌ی بدی بود، ناراضی‌ام", "Persian"],
    ["کیفیتش متوسط بود", "Persian"]
]

demo = gr.Interface(
    fn=predict_sentiment,
    inputs=[
        gr.Textbox(lines=3, label="Enter comment"),
        gr.Radio(["English", "Persian"], label="Choose Dataset/Language", value="English")
    ],
    outputs=[
        gr.Markdown(label="Prediction & Explanation"),
        gr.Image(label="Top Word Contributions")
    ],
    title=title,
    description=description,
    examples=examples,
)

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