Zechun Niu, Jiayu Chen, Haozhe Duan, Chunzhen Huang, Han Wen, Qing Tan, Kaiqiao Zhan, Kun Gai, Jiaxin Mao · ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 2026
DOI: 10.1145/3799682.3840805
Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).
Ranking is a central problem in information retrieval. Traditional learning-to-rank methods aim to train a ranking model to predict the relevance of each item and then rank them according to the relevance scores. However, many real-world ranking systems must satisfy application-specific constraints to ensure the fairness, safety, or diversity of the ranking, so the ranking decision can no longer be obtained by decreasingly sorting the relevance scores. A straightforward solution is to utilize the predict-then-optimize paradigm: first training a ranking model to predict relevance scores and then feeding the predicted relevance into a downstream optimizer to generate a ranking list that satisfied the constraints. This paradigm can produce feasible rankings when the constrained problem is feasible, but its prediction training is separated from the optimization of the ranking objective, making it difficult to directly optimize the quality of the induced ranking. In this paper, we propose SPORank, a unified framework for constrained learning to rank. By adopting the SPO+ loss, SPORank jointly considers relevance prediction and constrained optimization during training, making it broadly applicable to constrained ranking problems. We instantiate SPORank on two concrete tasks. The experimental results demonstrate SPORank effectively improves ranking performance while satisfying the required constraints.
ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 0 citations
ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 0 citations
ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 0 citations
ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 0 citations
No comments yet — start the discussion below.