Recommender-system

Data : Anime Recommendation Database 2020 [ https://www.kaggle.com/hernan4444/anime-recommendation-database-2020 ]

Content Based Filtering

Content based filtering is a method that uses similariteis in feature to make decisions

[Process]

1. Create Item profile

• Item profile is a dataframe which has ID of a certain object as row, and features describing the object as columns

2. Embedding columns

• In order to compare the significance of IDs

3. Use cosine similarity to compare each items

Latent Factor Model & Matrix Factorization

-> Latent factor model uses Dimensionality Reduction on user-item matrix to find the latent factor

We use SVD ( Singular Vector Decomposition ) on User-item matrix to avoid sparsity.

-> Matrix factorization is a class of collaborative filtering algorithm. This improves SVD

Factorization machine

This is a supervised algorithm that can be used for classification, regression, and ranking tasks

Wide & Deep

This is a ranking recommendation algorithm presented from Google. Read the Paper in this LINK : https://arxiv.org/pdf/1606.07792.pdf