skip to main content
introduction
Free access

Special Issue on Responsible Recommender Systems Part 1

Published: 28 August 2024 Publication History
Recommender systems are information filtering systems that suggest items tailored to individual users or user groups. They represent a powerful machine learning tool to support various human decision-making activities in e-commerce, social networks, entertainment, transportation, healthcare, and cybersecurity. Existing recommender systems typically focus on accuracy and personalization but increasingly call for an effective means of ensuring the systems work responsibly. Without appropriate responsible techniques, recommender systems could have undesired effects on users, communities, and society. For example, a recommendation algorithm trained on imbalanced data might be biased toward catering to the preferences of a majority group of users while overlooking minority groups; a system without countermeasures to misinformation may amplify the spread of misinformation, and a recommendation that lacks appropriate disclosure of the decision-making process or intuitive explanations may not be easily trusted by users.
Responsible recommender systems require innovative ways of assessing recommendation contexts and processing, communicating and presenting the recommendations. While responsible recommender systems have attracted attention in recent years, the corresponding challenges and threats in technical, social/societal, and ethical contexts are yet to be fully addressed. Furthermore, while existing approaches to the issue mostly focus on a single aspect, such as fairness, explainability, or social trust, a systematic/holistic approach to responsible recommender systems is still in demand. To address the challenges and seize the opportunities posed by responsible recommender systems, existing methodologies, models, techniques, and applications must be reassessed, adapted, or transformed significantly. Besides, emerging techniques for analyzing all types of side information (e.g., multimodal attributes, social information, external knowledge) should be fully explored to advance or expand the scope of state-of-the-art recommendation research toward delivering more responsible systems.
This special issue comprises 15 papers that present recent advances and novel contributions in the emerging yet promising field of responsible recommender systems. They represent some of the most recent progress in advancing responsible recommender systems research in four directions below.
Bias, Debiasing, and Miscalibration. Ahn and Lin propose a unified framework for examining miscalibration, bias, and stereotype in recommender systems. They reveal differences between algorithms and disparate impacts on groups and particular individuals in those key measures. Besides identifying the interactions between user characteristics (typicality and diversity), system-induced effects, and miscalibration, they find oversampling underrepresented groups and individuals effective for mitigating system-induced effects. Coppolillo et al. aim to mitigate popularity bias and propose a quality measure that rewards debiasing techniques that successfully push a recommender system to suggest niche items without sacrificing global recommendation accuracy. The proposed strategy is demonstrated to be effective in highlighting the debiasing techniques with the biggest improvements in the exposure of low-popularity items, exhibiting an advantage over several competitors. Duran et al. propose a deontological approach, in contrast to a consequentialist perspective, to overcoming diverse undesired effects in recommender systems due to biases. Besides introducing two distinct principles to avoid overconfidence in predictions and accurately model users’ interests, the approach circumvents the need to define a multi-objective system as many complex recommenders require.
Privacy and Security. Liu et al. focus on privacy-preserving point-of-interest (POI) recommendations for geological traveling. The study leverages original user preferences to create POIs that reflect users’ preferences but do not reveal their private information related to location check-ins. Li et al. adopt a decentralized, federated recommendation model to protect users’ privacy. They propose to share model parameters with related neighboring users in the process of model training using a privacy-aware structured client-level graph that contains randomly sampled fake entries. Using their designed collaborative training mechanisms, the proposed approach achieves high accuracy and low communication cost in their experiments. Ali et al. further address additional challenges beyond privacy protection for practical recommender systems, such as single-point failure and data/model tempering, by proposing a recommendation framework based on blockchain-empowered asynchronous federated learning. Gao et al. also propose a blockchain-based responsible recommender system. Yet, they aim to enhance the security, efficiency, and quality of service process creation and recommendation. While blockchain establishes a trusted service provision environment, the skip-gram model and recurrent neural networks are employed to generate semantic vectors and produce recommendations.
Fairness and Recommender Systems for Societal Good. Li et al. aim to quantify and reduce envy and inferiority for a fair recommendation of limited resources, especially job opportunities. They define inferiority and envy to be used in combination with utility as objectives to post-process the scores delivered by standard recommenders in a multi-objective optimization problem. Li et al. and Ma et al. focus on the application of recommender systems for building fashion and healthy lifestyles. Specifically, the former proposes a contrastive multimodal cross-attention network for body shape-aware fashion recommendation; the latter proposes a meal recommendation model that emphasizes the necessity for health-oriented and responsible meal recommendation systems to boost healthiness exposure by using nutritional standards as knowledge. Elahi et al. use knowledge graphs (KG) to represent recommendation scenarios to overcome the data sparsity and cold start problems. The proposed end-to-end recommendation framework encodes both relational and contextual information of entities, and their results highlight the importance of attention mechanisms and high-order connectivity in the responsible recommendation system for health informatics.
Explainability, Robustness, and Trust. Wang et al. provide a systematic overview and discussion of the literature in the realm of trustworthy recommender systems, along with a presentation of open challenges and a perspective on future directions. Yu et al. deal with the interpretability of recommendations for complex scenarios such as online games, and they propose a multi-source heterogeneous graph attention network for explainable recommendation for the case without enough a priori knowledge or corpus of user comments. Ren et al. advocate knowledge-enhanced conversational recommendations to make recommenders trustworthy and engaging to users, where semantic alignment between the embedding spaces of utterances and KG entities and explicit KG reasoning jointly facilitate accurate recommendation and quality dialogue generation. Lu et al. design an end-to-end deep adversarial multi-channel transfer network for cross-domain recommendation. They aim to overcome three challenges, namely heterogeneous data, negative transfer, and data sparsity, to improve the robustness of recommender systems.
The guest editors would like to thank all the authors for contributing to the special issue and the reviewers for offering their constructive reviews to uphold the quality of publications in the journal. We also thank the journal's editorial team for their professional support and collaboration.
CSIRO's Data61, Eveleigh, Australia and University of New South Wales, Australia
University of California, San Diego, CA, USA
University of Technology Sydney, Sydney, Australia
Dietmar Jannach
University of Klagenfurt, Klagenfurt, Austria
Guest Editors

Index Terms

  1. Special Issue on Responsible Recommender Systems Part 1

    Recommendations

    Comments

    Information & Contributors

    Information

    Published In

    cover image ACM Transactions on Intelligent Systems and Technology
    ACM Transactions on Intelligent Systems and Technology  Volume 15, Issue 4
    August 2024
    563 pages
    EISSN:2157-6912
    DOI:10.1145/3613644
    • Editor:
    • Huan Liu
    Issue’s Table of Contents

    Publisher

    Association for Computing Machinery

    New York, NY, United States

    Publication History

    Published: 28 August 2024
    Online AM: 15 June 2024
    Accepted: 30 April 2024
    Revised: 30 April 2024
    Received: 30 April 2024
    Published in TIST Volume 15, Issue 4

    Permissions

    Request permissions for this article.

    Check for updates

    Author Tags

    1. Recommender systems
    2. fairness
    3. privacy

    Qualifiers

    • Introduction

    Contributors

    Other Metrics

    Bibliometrics & Citations

    Bibliometrics

    Article Metrics

    • 0
      Total Citations
    • 450
      Total Downloads
    • Downloads (Last 12 months)450
    • Downloads (Last 6 weeks)84
    Reflects downloads up to 12 Feb 2025

    Other Metrics

    Citations

    View Options

    View options

    PDF

    View or Download as a PDF file.

    PDF

    eReader

    View online with eReader.

    eReader

    Login options

    Full Access

    Figures

    Tables

    Media

    Share

    Share

    Share this Publication link

    Share on social media