Top N Recommendations

This section of the Wiki is dedicated to models for predicting the next cards to select in a player's deck.

Global introduction

Top N recommendations models applied to the e-commerce field are able to predict the next most relevant item for an user depending of the items already present in his basket. To do so, it provides the top N recommendations in which there should be the next most relevant items for the user.

This field of study is very active since behind these models there is an important business stake for e-commerce companies.

Applied to Magic the Gathering

A player is using a deck from 60 to 100 cards to play Magic The Gathering.

The main idea is to consider a player's deck as a shopping basket. From this, top N recommendations models can be applied to predict the next card that may be relevant for the player based on the already selected cards in the deck.

To help the player to choose his next card, the model provides N cards that have obtained the best scores from the model. Each time the player adds a card to his deck, the model will provide N new recommendations that are relevant for the player for choosing the next card.

This model can be used to get recommendations around a set of few cards to know how with which cards they may be played with.

Difference from part 1

The first part of this project was to recommend a card based on the text or the selection of other players. It does not take into account previous cards selected by a player.

In other words, it recommends products without taking into account what a visitor has already selected in his shopping cart.