Movie Recommendation System

A prototype web application that offers personalized movie recommendations using Flask, MongoDB, and two complementary recommendation techniques:

1. Collaborative Filtering
   Predicts what rating a user would give to movies they haven’t rated yet (based on their past ratings and other users’ behavior), then presents the top‑predicted movies.

2. Content‑Based Filtering
   Finds movies similar to those the user has already watched, using TF‑IDF features and cosine similarity.

---

Features

• Username‑only access — no passwords; 10 test users are preloaded in MongoDB.
• Movie dashboard
• Top‑20 globally highest‑rated movies
• Interactive selection/search via datalist
• Movie poster thumbnails via placeholder API
• Rate movies and record watch history.
• Dual recommendation modes
• Collaborative Filtering: predicts unseen‑movie ratings and shows the top‑10 predicted favorites
• Content‑Based Filtering: suggests top‑10 similar titles based on TF‑IDF and cosine similarity
• User activity logging (login, rating, recommendation requests, logout) with timestamps.
• MongoDB
• users collection for user profiles & history
• user_logs collection for action logs

---

Project Structure

movie-recommender/
│
├── app.py
|
├── test.py
|
|── requirements.txt
├── requirements-linux.txt
|
├── a-b testing.ipynb
|
├── C_filtering.ipynb
├── CB_filtering.ipynb
│
├── runtime.txt
|
├── final_report.md
│
│
├── mongo_export/
│   ├── users.json                # Preloaded user profiles
│   └── log_sample.json           # Sample user action logs
|
└── templates/
    ├── index.html                # Username entry
    └── recommend.html            # Movie dashboard & recommendations

---

Setup Development Environment

1. Clone the repository

git clone https://github.com/mohitkumhar/movie-recommendation.git
cd movie-recommendation

2. Install Python dependencies

pip install -r requirements.txt

3. Start MongoDB

Run Docker container of mongoDB

# mongoDB Container on port 27017
docker run --name movieDB -v D:\movieDB:/data/db -p 27017:27017 -d mongo:latest 

# import the users.json into mongodb users collection
docker exec -i movieDB sh -c "mongoimport -c users -d user_db --jsonArray" < mongo_export\users.json

# import the log_sample.json into mongodb users_log collection
docker exec -i movieDB sh -c "mongoimport -c user_logs -d user_db --jsonArray" < mongo_export\log_sample.json

Make sure to import preloaded MongoDB data and Ensure that MongoDB is running locally on mongodb://localhost:27017/

4. Download Dataset

Make Sure to Download dataset, It is necessary to generate all .pkl files.

Download from-

MovieLens Dataset: https://www.kaggle.com/datasets/grouplens/movielens-20m-dataset

OR

Run Following Commands

curl -L -o movielens-20m-dataset.zip https://www.kaggle.com/api/v1/datasets/download/grouplens/movielens-20m-dataset

tar -xf movielens-20m-dataset.zip -C ./dataset

This dataset should be extracted in dataset/ folder on the root dir

5. Generate all important pickle files and models

Before running the Flask app, you must have all .pkl files.

Run both .ipynb files:

C_filtering.ipynb       # for Collaborative filtering
CB_filtering.ipynb      # for Content-Based filtering

Make sure all dataset/*.csv exists - it is required for Jupyter Notebook (.ipynb) to create .pkl files.

Note: To run in Jupyter Notebook, you should install jupyter notebook by pip install jupyter or can also be run in VS Code

6. Run the app

python app.py

Open your browser at http://localhost:5000.

---

---

How It Works

Collaborative Filtering

1. Load user–movie rating matrix (C_rating.pkl) and pre‑trained Surprise model (C_filtering_model.pkl).
2. For the current user, identify movies they haven’t rated.
3. Predict a rating for each unseen movie.
4. Sort predictions descending and return the top‑N movie titles.

Content‑Based Filtering

1. Load TF‑IDF vectorizer (CB_tfidfVectorizer.pkl) and cosine similarity matrix (CB_cosine_sim_matrix.pkl).
2. Compute an average similarity vector over the user’s watched‑movie indices.
3. Rank all movies by similarity (excluding already watched) and return top‑N titles.

---

Sample Data Files

• mongo_export/users.json

  [
    { "userId": 1, "username": "mohit", "moviesHistory": [] },
    { "userId": 2, "username": "alice", "moviesHistory": [] },
    …
  ]

• mongo_export/log_sample.json

  {
    "user": { "userId": 1, "username": "mohit" },
    "action": "request recommendation",
    "details": { "type": "Collaborative Filtering" },
    "timestamp": "2025-07-01T14:22:31.123Z"
  }

---

Future Work

• Secure Authentication: add password hashing & signup workflow
• User Profiles & Preferences: genres, watchlists, favorites
• UI Enhancements: responsive design, movie trailers via TMDB API
• Analytics Dashboard: visualize user engagement & model performance
• Cloud Deployment: MongoDB Atlas, Docker
• Social Features: allow users to follow friends and share recommendations

---

Author

Mohit Kumhar

• GitHub: @mohitkumhar
• LinkedIn: @mohitkumhar
• Email: mohitmolela@gmail.com