app.py

import pandas as pd
import numpy as np
from sklearn.metrics.pairwise import cosine_similarity
from sklearn.model_selection import train_test_split
from sklearn.feature_extraction.text import TfidfVectorizer
import requests
import gradio as gr
import os

ratings = pd.read_csv("ratings.csv")
movies = pd.read_csv("movies.csv")

OMDB_API_KEY = os.environ.get("omdbapikey")

movie_lookup = movies.set_index("movieId")["title"].to_dict()
reverse_movie_lookup = {v.lower(): k for k, v in movie_lookup.items()}

train_df, test_df = train_test_split(ratings, test_size=0.2, random_state=42)
train_matrix = train_df.pivot_table(index='userId', columns='movieId', values='rating')
train_matrix_filled = train_matrix.fillna(0)

user_similarity = cosine_similarity(train_matrix_filled)
user_similarity_df = pd.DataFrame(user_similarity, index=train_matrix_filled.index, columns=train_matrix_filled.index)

item_rating_matrix = train_matrix_filled.T
item_similarity = cosine_similarity(item_rating_matrix)
item_similarity_df = pd.DataFrame(item_similarity, index=item_rating_matrix.index, columns=item_rating_matrix.index)

data = pd.merge(ratings, movies, on='movieId')
data['genres'] = data['genres'].fillna('')
vectorizer = TfidfVectorizer(token_pattern=r'[a-zA-Z0-9\-]+')
tfidf_matrix = vectorizer.fit_transform(data['genres'].values)
movie_ids = data['movieId'].values
unique_movie_ids, indices = np.unique(movie_ids, return_index=True)
movie_id_to_index = {mid: idx for idx, mid in enumerate(unique_movie_ids)}
movie_genre_matrix = tfidf_matrix[indices]

def get_movie_poster(title):
    if not OMDB_API_KEY:
        return ''
    try:
        response = requests.get(f"http://www.omdbapi.com/?t={title}&apikey={OMDB_API_KEY}")
        data = response.json()
        return data.get('Poster', '')
    except:
        return ''

def user_cf_recommend(user_id):
    try:
        user_id = int(user_id)
        if user_id not in user_similarity_df.index:
            return "User ID not found."
        
        similar_users = user_similarity_df[user_id].drop(user_id)
        top_similar_users = similar_users.sort_values(ascending=False).head(10)
        
        scores = {}
        sim_sums = {}
        
        for other_user, similarity in top_similar_users.items():
            other_ratings = train_matrix.loc[other_user].dropna()
            for movie_id, rating in other_ratings.items():
                if movie_id not in train_matrix.loc[user_id] or pd.isna(train_matrix.loc[user_id, movie_id]):
                    scores[movie_id] = scores.get(movie_id, 0) + similarity * rating
                    sim_sums[movie_id] = sim_sums.get(movie_id, 0) + abs(similarity)
        
        ranked_movies = sorted([(movie_id, score / sim_sums[movie_id]) for movie_id, score in scores.items() if sim_sums[movie_id] > 0],
                              key=lambda x: x[1], reverse=True)[:5]
        
        result = []
        for movie_id, score in ranked_movies:
            title = movie_lookup.get(movie_id, 'Unknown')
            poster = get_movie_poster(title)
            result.append((title, round(score, 2), poster))
        
        return result
    except:
        return "Invalid input."

def item_cf_recommend(movie_title):
    movie_title = movie_title.lower().strip()
    if movie_title not in reverse_movie_lookup:
        return "Movie not found."
    
    target_movie_id = reverse_movie_lookup[movie_title]
    
    if target_movie_id not in item_similarity_df:
        return "No similarity data available."
    
    similar_scores = item_similarity_df[target_movie_id].drop(target_movie_id)
    top_similar_ids = similar_scores.sort_values(ascending=False).head(5).index
    
    result = []
    for mid in top_similar_ids:
        title = movie_lookup.get(mid, 'Unknown')
        poster = get_movie_poster(title)
        result.append((title, poster))
    
    return result

def cb_recommend(movie_title):
    movie_title = movie_title.strip().lower()
    movies['title_lower'] = movies['title'].str.lower()
    
    if movie_title not in movies['title_lower'].values:
        return "Movie not found."
    
    input_index = movies[movies['title_lower'] == movie_title].index[0]
    movie_id = movies.loc[input_index, 'movieId']
    
    if movie_id not in movie_id_to_index:
        return "No genre data available."
    
    input_vec = movie_genre_matrix[movie_id_to_index[movie_id]]
    sims = cosine_similarity(input_vec, movie_genre_matrix).flatten()
    sim_indices = sims.argsort()[::-1]
    
    seen = set()
    result = []
    for i in sim_indices:
        rec_movie_id = unique_movie_ids[i]
        title = movies[movies['movieId'] == rec_movie_id]['title'].values[0]
        if title.lower() != movie_title and title not in seen:
            poster = get_movie_poster(title)
            result.append((title, poster))
            seen.add(title)
        if len(result) == 5:
            break
    
    return result

def format_recommendations(recommendations):
    if isinstance(recommendations, str):
        return recommendations
    
    formatted = []
    for item in recommendations:
        if len(item) == 3:
            title, score, poster = item
            if poster:
                formatted.append(f"<div style='display:flex;margin-bottom:10px;'><img src='{poster}' style='width:80px;height:120px;object-fit:cover;margin-right:10px;'><div><b>{title}</b><br>Predicted rating: {score}</div></div>")
            else:
                formatted.append(f"<div><b>{title}</b><br>Predicted rating: {score}</div>")
        else:
            title, poster = item
            if poster:
                formatted.append(f"<div style='display:flex;margin-bottom:10px;'><img src='{poster}' style='width:80px;height:120px;object-fit:cover;margin-right:10px;'><div><b>{title}</b></div></div>")
            else:
                formatted.append(f"<div><b>{title}</b></div>")
    
    return "<br>".join(formatted)

def respond(message, history):
    message = message.lower().strip()
    
    if message.startswith("recommend for user"):
        try:
            user_id = int(message.split()[-1])
            recs = user_cf_recommend(user_id)
            return format_recommendations(recs)
        except:
            return "Please provide a valid user ID after 'recommend for user'"
    
    elif message.startswith("similar to"):
        movie_title = message[10:].strip()
        recs = item_cf_recommend(movie_title)
        return format_recommendations(recs)
    
    elif message.startswith("recommend like"):
        movie_title = message[14:].strip()
        recs = cb_recommend(movie_title)
        return format_recommendations(recs)
    
    else:
        return "Available commands:\n1. 'recommend for user [ID]'\n2. 'similar to [Movie Title]'\n3. 'recommend like [Movie Title]'"

demo = gr.ChatInterface(
    respond,
    title="Movie Recommendation Chatbot",
    description="Ask for recommendations using these commands:\n1. 'recommend for user [ID]'\n2. 'similar to [Movie Title]'\n3. 'recommend like [Movie Title]'",
    examples=[
        ["recommend for user 42"],
        ["similar to Toy Story"],
        ["recommend like The Dark Knight"]
    ]
)

demo.launch()