Anime Recommendation

This is an academic project based on Recommendation Systems for recommending Animes based on user interactions.

Dataset

• Source: Kaggle
• Two CSV files:
  • anime.csv
  • ratings.csv
• The anime.csv file contains information about each anime title like:
  • title
  • genre
  • type
  • episodes
  • rating (group rating/score)
• The rating.csv file contains user ratings for each anime title, with each row containing:
  • a user ID
  • an anime ID
  • a rating (user rating)

Models

The following recommendation models are explored in this project:

• Retrieval Model - A model that retrieves a list of top K anime titles for a given user based on implicit interactions.
• Ranking Model - A model that predicts the rank of an anime for a given user based on their explicit interactions like ratings.
• Multitask Model - A model that combines the ranking and retrieval models.
• Deep & Cross Network Model (DCN) - A model that uses deep learning techniques to learn the interactions between users and anime titles.

Requirements

Apart from the Python programming language and other standard packages like numpy, pandas, and others, the following key packages are required for the project:

• Tensorflow
• Tensorflow Recommenders

To install:

# tensorflow
pip install tensorflow

# tensorflow-recommenders
pip install tensorflow-recommenders

Notebooks

The code for the project is divided into multiple Jupyter notebooks. Each notebook contains code for a specific type of model along with the required data preprocessing.

• retrieval_and_ranking_models.ipynb - This notebook contains the code for building basic retrieval and ranking models.
• ranking_model_using_metadata.ipynb - This notebook contains the code for updating the ranking model to use metadata.
• multitask_recommender_model.ipynb - This notebook contains the code for building the multitask recommender model.
• dcn_model.ipynb - This notebook contains the code for building deep & cross network model.
• main.ipynb - This notebook contains code for all the models and compares their performance.

Members

1. Ashutosh Ojha
2. Harisha Korapati
3. Zeba Wahab