A framework for unbiased explainable pairwise ranking for recommendation

Khalil Damak,Sami Khenissi,Olfa Nasraoui

doi:10.1016/j.simpa.2021.100208

Abstract

<h2>Abstract</h2> Recent research in recommender systems has demonstrated the advantages of pairwise ranking in recommendation. In this work, we focus on the state-of-the-art pairwise ranking loss function, Bayesian Personalized Ranking (BPR), and aim to address two of its limitations, namely: (1) the lack of explainability and (2) exposure bias. We propose a recommendation framework that encompasses various loss functions that are based on BPR and which aim to mitigate the aforementioned limitations. Our open-source framework includes code to train and tune state-of-the-art pairwise ranking recommender systems on benchmark datasets and evaluate them based on the three criteria of ranking accuracy, explainability, and popularity debiasing.

Highlights

Bayesian Personalized Ranking (BPR) is a pairwise ranking approach [1] that has recently received significant praise in the recommender systems community because of its capacity to rank implicit feedback data with high accuracy [2]
The latter assumption engenders exposure bias, which is a notorious issue in recommendation from implicit feedback, and that is usually characterized by a bias against less popular items having a lower propensity of being observed [3]
In our previous work [6], we proposed novel loss functions for pairwise ranking recommendation, which aim to improve the explainability of BPR and mitigate exposure bias

Summary

Introduction

Bayesian Personalized Ranking (BPR) is a pairwise ranking approach [1] that has recently received significant praise in the recommender systems community because of its capacity to rank implicit feedback data with high accuracy [2]. Aiming to rank relevant items higher than irrelevant items, pairwise ranking recommender systems often assume that all non-interacted items are irrelevant The latter assumption engenders exposure bias, which is a notorious issue in recommendation from implicit feedback, and that is usually characterized by a bias against less popular items having a lower propensity of being observed [3]. Most state-of-the-art recommender systems, including BPR, are black boxes that do not justify why or how an item was recommended to a user This might engender unfairness issues if inappropriate content gets recommended to a user. In our previous work [6], we proposed novel loss functions for pairwise ranking recommendation, which aim to improve the explainability of BPR and mitigate exposure bias. Our framework aims to facilitate incorporating explainability and exposure debiasing into pairwise ranking models for recommendation. We will delve in more detail about the different characteristics and functionalities of our framework

Loss functions

Models

Training the models

Tuning the models

Evaluation criterion