Jingwei Kang

PhD candidate · Information Retrieval Lab (IRLab), University of Amsterdam

Portrait of Jingwei Kang

I study recommender systems in streaming media services, where recommendations are shown in carousel interfaces. My work focuses on user behaviour modelling and offline evaluation; I have also worked on learning to rank. I am a PhD candidate at the University of Amsterdam, supervised by prof. dr. Maarten de Rijke and dr. Harrie Oosterhuis.

Research

User behaviour modelling

Offline evaluation

Learning to rank

Publications

2026

  1. Short paper

    Revisiting N2DCG: An Empirically Grounded Reformulation of Carousel Recommendation Evaluation

    Jingwei Kang, Santiago de Leon-Martinez, Maarten de Rijke, Harrie Oosterhuis

    CIKM ’26 · 35th ACM International Conference on Information and Knowledge Management, Rome, Italy, November 2026, pp. 7284–7289

    Shows that N2DCG, the NDCG adaptation for carousels, uses an ideal ranking that violates carousel constraints and a discount that does not match observed browsing; proposes a reformulated metric grounded in eye-tracking data.

  2. Doctoral symposium

    From Click Modeling to Offline and Off-Policy Evaluation in Carousel Recommendation

    Jingwei Kang

    RecSys ’26 · 20th ACM Conference on Recommender Systems, Minneapolis, USA, September 2026, pp. 1932–1938

    Lays out the PhD programme: a click-model design framework built on relations between observed variables, discrete-choice click models, carousel-specific offline metrics, and off-policy evaluation of carousel recommendation policies from logged data.

  3. Full paper

    Following the Eye-Tracking Evidence: Established Web-Search Assumptions Fail in Carousel Interfaces

    Jingwei Kang, Maarten de Rijke, Harrie Oosterhuis

    SIGIR ’26 · 49th International ACM SIGIR Conference on Research and Development in Information Retrieval, Melbourne, Australia, July 2026, pp. 734–744

    Tests four web-search behavioural assumptions on carousel eye-tracking data: the F-pattern holds only vertically, users follow an L-shaped pattern after clicking, the examination hypothesis fails, and users skip carousel headings.

2025

  1. Resource paper

    A Large-Scale Web Search Dataset for Federated Online Learning to Rank

    Marcel Gregoriadis, Jingwei Kang, Johan Pouwelse

    CIKM ’25 · 34th ACM International Conference on Information and Knowledge Management, November 2025, pp. 6387–6391

    A large-scale web search dataset built for studying federated online learning to rank.

  2. Full paper · Best Paper Honorable Mention

    Rethinking Click Models in Light of Carousel Interfaces: Theory-Based Categorization and Design of Click Models

    Jingwei Kang, Maarten de Rijke, Santiago de Leon-Martinez, Harrie Oosterhuis

    ICTIR ’25 · International ACM SIGIR Conference on Innovative Concepts and Theories in Information Retrieval, July 2025, pp. 44–55

    Replaces outdated click-model taxonomies with three mathematically grounded design choices that cover single-list, grid and carousel click models (PGMs and neural), and derives a new carousel click model from them.

  3. Resource paper

    RecGaze: The First Eye Tracking and User Interaction Dataset for Carousel Interfaces

    Santiago de Leon-Martinez, Jingwei Kang, Róbert Móro, Maarten de Rijke, Branislav Kveton, Harrie Oosterhuis, Maria Bielikova

    SIGIR ’25 · 48th International ACM SIGIR Conference on Research and Development in Information Retrieval, July 2025, pp. 3702–3711

    The first public carousel dataset combining eye tracking, clicks and cursor movements (87 users, 3,477 interactions), revealing a golden-triangle, F-shaped browsing pattern.

2024

  1. Short paper

    Estimating the Hessian Matrix of Ranking Objectives for Stochastic Learning to Rank with Gradient Boosted Trees

    Jingwei Kang, Maarten de Rijke, Harrie Oosterhuis

    SIGIR ’24 · 47th International ACM SIGIR Conference on Research and Development in Information Retrieval, July 2024, pp. 2390–2394

    Makes gradient-boosted decision trees usable for stochastic learning to rank by estimating the Hessian of ranking objectives.

Education

  • Ph.D. in Recommender Systems

    University of Amsterdam, Amsterdam, the Netherlands, 2023–present

    Dissertation topic: Full-page Personalization

  • M.Sc. in Data Science

    Radboud University, Nijmegen, the Netherlands, 2021–2023

  • B.Sc. in Data Science

    Central South University, Changsha, China, 2016–2020

Contact

Email is the best way to reach me, whether about a paper, a dataset or a possible collaboration.