Page 171 - The-5th-MCAIT2021-eProceeding

Explainable Recommender – Implementation Approaches

a
b
a
a*
Neeraj Tiwary , Shahrul Azman Mohd Noah , Fariza Fauzi , Steffen Staab
a Faculty of Information Science and Technology, Universiti Kebangsaan Malaysia, Bangi, Malaysia
b WAIS, University of Southampton, UK & Analytic Computing, Universität Stuttgart, DE
* Email: p105920@siswa.ukm.edu.my, Neeraj.tiwary@gmail.com

Abstract

During the last few decades, with the rise of YouTube, Amazon, Netflix, and many other web services, recommender
systems are playing a more important role in our lives and are well-researched. Organizations want to recommend the
contents/products based on individual needs. Personalization is the one aspect where every organization is paying more
attention. Personalized services enable users to find related/relevant information quickly, be it a shopping item, movie, news,
travel ideas, or restaurants that best suits their tastes from the countless choices. Thus, the demand for personalized
recommenders has become pervasive in all domains. Besides providing the right recommendations, explainability for those
recommendations is also very much required. The ability to explain how/why these recommendations are suggested will
bridge the confidence and trust of the users in the system. Knowledge graph, reinforcement learning, and language models
are some recent state-of-the-art technological advancements. Recommendation systems can benefit from these
advancements by effectively applying them to enhance the efficacy of the explainable recommendations. The objective of
this paper is to discuss the various implementation methods for developing an explainable recommendation engine. The
contribution of this paper is to explain the concept of explainable recommendation as well as provide different
implementation mechanisms and niche advancements in the explainable recommendation area.

Keywords: Explainable AI; Knowledge Graph; Reinforcement Learning; Language Models

1. Introduction

Recommender systems (RS) are algorithms aimed at suggesting relevant items (i.e. movies, books, products,
etc.) to users based on his/her preferences towards those items (Cosley et al., 2003). They are critical in many
industries like entertainment, e-commerce, media, and advertisements and pivotal in driving huge revenue for
organizations. Some of the initial approaches for developing recommender systems were using content-based
or collaborative filtering-based algorithms. Content-based recommendation is in which a user is recommended
items that are similar to those that the user liked in the past, whereas collaborative recommendation is where a
user is recommended items that other users with similar tastes liked in the past (Ricci et al., 2011). One of the
main issues that these recommender systems has is the explainability of the recommendations. Though the
concept of explainable recommendation was formally introduced by Zhang et al. 2014, this concept is not new,
and many other researchers (Schafer, 1999; Herlocker, 2000) coined this idea earlier. Schafer et al., 1999
explained the recommendations reasoning through the customer past relationships with the e-commerce system.
Herlocker et al., 2000 conducted a study on collaborative filtering algorithms in the MovieLens dataset and
analyzed the explainability based on user surveys.
Recently, a series of AI regulations have entered into force, such as the EU General Data Protection
Regulation (GDPR) and the California Consumer Privacy Act of 2018, which emphasize the “right to
explanation” of algorithmic decisions. Overall, the explainability of AI systems is always an imperative topic.
This paper is organized to explain past and recent implementation approaches for explainable recommendation
systems and niche advancements in this area.

E- Proceedings of The 5th International Multi-Conference on Artificial Intelligence Technology (MCAIT 2021) [158]
Artificial Intelligence in the 4th Industrial Revolution