EURO 2024 Copenhagen
Abstract Submission

EURO-Online login

2059. Towards effective and efficient personalized ranking in news article recommender systems: balancing clicks and diversity

Contributed abstract in session MA-31: Recommender systems, stream Analytics.

Monday, 8:30-10:00
Room: 046 (building: 208)

Authors (first author is the speaker)

1. Lukas De Kerpel
Marketing, cluster Data Analytics, Ghent University
2. Dries Benoit
Ghent University

Abstract

As the consumption of news articles continues to shift towards online platforms, personalized news recommendation engines have become an indispensable asset for news providers seeking to enhance users’ reading experience.

However, as the shelf life of news items is limited, traditional collaborative filtering approaches fall short to provide accurate recommendations because they heavily rely on user signals, which may not be readily available for newly introduced news articles. Contextual multi-armed bandits (CMAB) emerge as a promising alternative by leveraging intrinsic characteristics and attributes of both news items and users.

While CMAB systems predominantly optimize for clicks, this research argues for a more comprehensive evaluation framework that also promotes the diversity of recommended items, as diversity-aware recommenders are able to respond to news readers’ needs on information variety and to expose users to counter-attitudinal behavior.

By conducting a benchmark study on a large-scale proprietary dataset obtained from an international news provider, we demonstrate the feasibility of creating an effective and efficient personalized news article recommendation system based on CMAB models while simultaneously addressing the multifaceted nature of news article recommendation quality. In particular, this study highlights the necessity of considering diversity alongside clicks to enhance user satisfaction and engagement.

Keywords

Status: accepted

Back to the list of papers