app.py

from sklearn.metrics.pairwise import cosine_similarity
import numpy as np
import gradio as gr
import pandas as pd
from sentence_transformers import SentenceTransformer

# טען את המודל
model = SentenceTransformer("all-MiniLM-L6-v2")

# טען את הדאטהסט מהלינק
url = "https://huggingface.co/datasets/Pablinho/movies-dataset/resolve/main/9000plus.csv"
print("Loading dataset...")
dataset = pd.read_csv(url)

# ודא שהעמודות קיימות
assert "Title" in dataset.columns
assert "Overview" in dataset.columns

# נקה שורות עם Overview חסר או לא מחרוזת
dataset = dataset.dropna(subset=["Overview"])
dataset = dataset[dataset["Overview"].apply(lambda x: isinstance(x, str))]

# הגבל ל־500 סרטים
MAX_MOVIES = 500
dataset = dataset.head(MAX_MOVIES)

print(f"Encoding {len(dataset)} movie descriptions...")
dataset["embeddings"] = dataset["Overview"].apply(lambda x: model.encode(x).tolist())
print("Done encoding!")

def recommend_similar_movies(input_text, top_n=5):
    input_embedding = model.encode([input_text])
    similarities = cosine_similarity(input_embedding, np.vstack(dataset['embeddings'].to_numpy()))[0]
    top_indices = similarities.argsort()[::-1][:top_n]
    results = dataset.iloc[top_indices][['Title', 'Overview']]
    return "\n\n".join(f"🎬 **{row['Title']}**\n{row['Overview']}" for _, row in results.iterrows())

demo = gr.Interface(
    fn=recommend_similar_movies,
    inputs=gr.Textbox(lines=2, placeholder="Describe a movie..."),
    outputs="text",
    title="Movie Recommender",
    description="Get movie recommendations based on your description. Powered by sentence-transformers and cosine similarity."
)

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