Multi-purpose Recommender Platform using Perceiver IO

Abstract

Web services usually require many different types of recommender systems using large amount of user log and content data, in order to provide personalized content to their customers. Different recommenders may share the same customer-base or cross-use models/data. It is challenging to design different models for each recommendation task. In this work, we propose a general-purpose framework for various recommendation tasks based on Perceiver IO model. Perceiver lOis a general ma-chine learning architecture based on transformer-style attention modules, which helps eliminating feature engineering for various tasks. Different type of recommenders can be developed with minimal modifications and models can be transferred among dif- ferent tasks. Our experiments with a variety of recommendation scenarios confirm that our framework is able to handle those tasks while achieving state-of-the-art accuracy.