Back to projects

k-NN Anime Recommender

Model Developer•View Repository

Overview

k-NN Anime Recommender is a final requirement for our machine learning course. We had to compare two different models/approaches for research and we went with Collaborative Filtering vs. Content-Based Filtering on recommenders.

My Contribution

I mainly worked on the Collaborative Filtering model, which recommends related shows based on what other MAL users rated. If they rated show A and B high and you wanted recommendations for show A, the model will recommend show B to you.

Technologies Used

  • Python
  • Sci-kit learn
  • Pandas
  • NumPy
  • MatPlotLib
  • Seaborn

Challenges

As this was my first go at training machine learning models, the challenges were mainly about getting my grasp on sci-kit and how to train models in general.

Key Takeaway

Based on our findings testing both Collaborative Filtering and Content-Based Filtering, I learned that it is better to actually use both at once. Content-Based Filtering recommends shows based on semantics and performed better quantitatively. Collaborative Filtering performed worse quantitatively but qualitatively I would argue it recommended better shows.

Future Improvements

Potential enhancements include:

  • Using or making a more recent dataset.
  • Building a hybrid model that combines both Collaborative and Content-Based Filtering.
  • Factoring in other user behaviors like watch time and dropped shows, not just 1-10 ratings.
  • Exploring deep learning or other advanced machine learning algorithms to improve predictions.

Links