SmartRecs

A Hybrid Movie Recommendation System using Machine Learning

Overview

SmartRecs is a personalized media recommendation engine built using a hybrid approach that combines content-based filtering and collaborative filtering. It leverages large-scale IMDb datasets and machine learning techniques such as TF-IDF, Cosine Similarity, and Singular Value Decomposition (SVD) to suggest movies tailored to user preferences.

Features

• Hybrid Recommendation: Combines genre similarity and collaborative ratings
• Fuzzy matching for flexible title inputs
• Genre-safe filtering (e.g., Horror is excluded unless requested)
• Large-scale compatibility using Kaggle Notebooks
• Easily extendable architecture for future work

Dataset

This project uses official IMDb datasets:

• title.basics.tsv (~1.01 GB)
• title.ratings.tsv (~27 MB)

Note: The dataset files are not included in this repository because of their large size (over 1 GB), which exceeds GitHub's file size limits and is impractical for version control.
Instead, we provide direct download links to the IMDb dataset source so users can fetch the latest and official data themselves.

Download Links:

• https://datasets.imdbws.com/title.basics.tsv.gz
• https://datasets.imdbws.com/title.ratings.tsv.gz

After downloading, unzip the .gz files before using them in the notebook.

Methodology

Content-Based Filtering

• Extracted genre vectors using TF-IDF
• Computed cosine similarity between genre vectors

Collaborative Filtering

• Used IMDb average rating as implicit feedback
• Applied SVD from the Surprise library

Hybrid Score Calculation

final_score = 0.6 * svd_score + 0.4 * genre_similarity

Recommendations are sorted by this hybrid score and filtered for genre relevance.

Development Environment

This project was developed using Kaggle Notebooks, which offer up to 30 GB of RAM — necessary for handling large matrices and high-dimensional TF-IDF vectors. Local systems and free Colab instances did not have sufficient memory for end-to-end execution.

How to Run

1. Download the IMDb dataset from the links above.
2. Upload the unzipped .tsv files to your working directory or Kaggle Notebook.
3. Open the smartrecs.ipynb notebook.
4. Run the notebook and follow the prompt:
   • Input a movie title
   • Select the correct match
   • View top-N hybrid recommendations

Example Results

Input: Inception

Returns top Action/Sci-Fi recommendations like Rogue One, Avengers: Endgame

Input: Cars 2

Returns animated family-friendly films like Cars 3, Finding Dory

Documentation

Only the final PDF versions of the project documents are included in this repository under the docs/ folder:

• Report.pdf
• Presentation Slides.pdf

Drafts, .docx, and .pptx files are excluded to keep the repository clean and lightweight.

Future Work

• Add filtering by actor/director/runtime
• UI deployment using Streamlit or Flask
• User-based personalization via login
• Model enhancement with embeddings or transformers

License

This project is licensed under the MIT License.
You are free to use, modify, and distribute this project with attribution.

Author

Wineel Wilson Dasari
MS in Computer Science – Santa Clara University
Project under CSEN 240: Machine Learning

References

• IMDb Datasets: https://datasets.imdbws.com/
• Surprise Library: https://surpriselib.com/
• Scikit-Learn, Pandas, NumPy
• Kaggle Notebook: https://www.kaggle.com/code/wineelscu/smartrecs