United We Stand, Divided We Fall: Leveraging Ensembles of Recommenders to Compete with Budget Constrained Resources

Abstract

In this paper we provide an overview of the approach we used as team Surricchi1 for the ACM RecSys Challenge 20221. The competition, sponsored and organized by Dressipi, involves a typical session-based recommendation task in the fashion industry domain. Our proposed method2 leverages an ensemble of multiple recommenders selected to capture diverse facets of the input data. Such a modular approach allowed our team to achieve competitive results with a score of 0.1994 Mean Reciprocal Rank at 100 (∼ 7.6% less than the first qualified team). We obtained this result by leveraging only publicly and freely available computational resources 3 and our own laptops. Part of the merit also lies in the size of this year’s dataset (∼ 5 million data points), which democratized the challenge to a larger public and allowed us to join the challenge as independent researchers.