How to choose a recommender

Input Data

What is the input?

RePlay models differ by types of data they can process:

Collaborative use only user-item interaction logs.
Content-based use only user or item features.
Hybrid can use both intercations and features.

Are interactions explicit?

Our information can be either explicit e.g. ratings, or implicit e.g. view, number of play counts.
Some models transform any type of data to implicit (unary ratings).

Will there be new users?

Some models need to be completely retrained to give predictions for new users (not present in train data, but have interactions history) while others don’t.

Will there be new items?

The same goes for new items.

Algorithm	Data	Interactions	New Users	New Items
Popular By Users	Collaborative	implicit feedback	-	-
Alternating Least Squares	Collaborative	implicit feedback	-	-
Wilson Recommender	Collaborative	binary ratings	+	-
UCB	Collaborative	binary ratings	+	+
KL-UCB	Collaborative	binary ratings	+	+
LinUCB	Collaborative	binary ratings	+	-
Thompson Sampling	Collaborative	binary ratings	+	+
Conservative Q-Learning (Experimental)	Collaborative	binary ratings	+	-
DDPG (Experimental)	Collaborative	binary ratings	+	-
Popular Recommender	Collaborative	converted to unary ratings	+	-
Random Recommender	Collaborative	converted to unary ratings	+	+
K-Nearest Neighbours	Collaborative	converted to unary ratings	+	-
BERT4Rec	Collaborative	converted to unary ratings	-	-
SASRec	Collaborative	converted to unary ratings	-	-
Mult-VAE (Experimental)	Collaborative	converted to unary ratings	+	-
Word2Vec Recommender	Collaborative	converted to unary ratings	+	-
Association Rules	Collaborative	converted to unary ratings	+	-
Neural Matrix Factorization (Experimental)	Collaborative	converted to unary ratings	-	-
SLIM	Collaborative	unary ratings, explicit feedback	+	-
ADMM SLIM (Experimental)	Collaborative	unary ratings, explicit feedback	+	-
ULinUCB (Experimental)	Hybrid	binary ratings	-	+
Neural Thompson Sampling (Experimental)	Hybrid	binary ratings	+	-
Category Popular Recommender	Hybrid	converted to unary ratings	+	-
Cluster Recommender	Hybrid	converted to unary ratings	+	-
LightFM Wrap (Experimental)	Hybrid	depends on loss	+	+
Implicit Wrap (Experimental)	Collaborative	depends on model	-	-
Two Stages Scenario (Experimental)	Hybrid	converted to unary ratings for second level	`*`	`*`

* - depends on base models.

Типы взаимодействий:

binary ratings - модель использует как положительные, так и отрицательные взаимодействия.
unary ratings - модель учитывает только факт положительного взаимодействия. Все взаимодействия считаются позитивными.
implicit feedback - неявные оценки.
unary ratings, explicit feedback - модель способна работать как с унарными оценками (факт взаимодействия), так и с явными оценками (например, рейтингами).
depends on loss - зависит от функции потерь.

Model requirements

Should recommendations be personalized?

Should cold users get recommendations? (without any interactions).

Should model be able to recommend unseen items?

Algorithm	Personalized	Cold Users	Cold Items	Recommends Unseen Items
Popular By Users	+	-	-	-
Alternating Least Squares	+	-	-	+
Wilson Recommender	-	+	-	+
UCB	-	+	+	+
KL-UCB	-	+	+	+
LinUCB	+	+	-	+
Thompson Sampling	-	+	+	+
Conservative Q-Learning (Experimental)	+	+	-	+
DDPG (Experimental)	+	+	-	+
Popular Recommender	-	+	-	+
Random Recommender	-	+	+	+
K-Nearest Neighbours	+	+	-	+
BERT4Rec	+	-	-	+
SASRec	+	-	-	+
Mult-VAE (Experimental)	+	-	-	+
Word2Vec Recommender	+	-	-	+
Association Rules	+	-	-	+
Neural Matrix Factorization (Experimental)	+	-	-	+
SLIM	+	-	-	+
ADMM SLIM (Experimental)	+	-	-	+
ULinUCB (Experimental)	+	-	+	+
Neural Thompson Sampling (Experimental)	+	+	-	+
Category Popular Recommender	-	+	-	+
Cluster Recommender	+	+	-	+
LightFM Wrap (Experimental)	+	+	+	+
Implicit Wrap (Experimental)	+	-	-	+
Two Stages Scenario (Experimental)	+	`*`	`*`	`*`