aiqbench analysis
This research introduces Contextual Scalarisation Thompson Sampling (CSTS), a machine learning method designed to dynamically balance competing objectives in public media recommender systems. By adapting objective weights based on real-time context, the approach outperforms static weighting and standard bandit techniques using data from Radio Télévision Suisse. The study addresses critical multi-objective decision-making challenges where audience reach must be reconciled with cultural mandates and editorial constraints.
Contextual Scalarisation Thompson Sampling for multi-objective decisions in public media
arXiv:2605.31291v2 Announce Type: replace-cross Abstract: Recommender systems may operate under multiple, competing objectives. For example, audience reach, cultural values, public service mandate, and operational constraints must be balanced in editorial decisions of public service media. Existing approaches relying on fixed combinations of objectives or Pareto-based optimisation do not adapt to changing priorities across situations. In this paper, we propose Contextual Scalarisation Thompson Sampler (CSTS), a multi-objective contextual bandit method that learns to weight objectives as a function of the observed context. We evaluate CSTS on real programming data from Radio T\'el\'evision Suisse, the Swiss national broadcaster, showing improved contextual relevance and better alignment with expert curation practices compared to fixed weight and standard contextual bandit approaches.
Abstract:Recommender systems may operate under multiple, competing objectives. For example, audience reach, cultural values, public service mandate, and operational constraints must be balanced in editorial decisions of public service media. Existing approaches relying on fixed combinations of objectives or Pareto-based optimisation do not adapt to changing priorities across situations. In this paper, we propose Contextual Scalarisation Thompson Sampler (CSTS), a multi-objective contextual bandit method that learns to weight objectives as a function of the observed context. We evaluate CSTS on real programming data from Radio Télévision Suisse, the Swiss national broadcaster, showing improved contextual relevance and better alignment with expert curation practices compared to fixed weight and standard contextual bandit approaches.
15 pages, 3 figures, 3 tables. Submitted-manuscript version of a paper published at ICPR 2026 (LNCS vol. 16824, Springer). v2 adds the publisher acknowledgement and the DOI of the Version of Record, and corrects bibliography metadata; no other changes
Information Retrieval (cs.IR); Machine Learning (cs.LG)
arXiv:2605.31291 [cs.IR]
(or arXiv:2605.31291v2 [cs.IR] for this version)
Pattern Recognition. ICPR 2026. Lecture Notes in Computer Science, vol. 16824, pp. 357-372. Springer, Cham (2027)
https://doi.org/10.1007/978-3-032-31927-2_24 DOI(s) linking to related resources
Full text under the publisher's open licence · 186 words
Related coverage
Article: Governing AI in the Cloud: A Practical Guide for Architects
InfoQ · Jun 15, 2026
Presentation: Building Evals for AI Adoption: from Principles to Practice
InfoQ · May 29, 2026
[AINews] not much happened today
Latent Space · Oct 3, 2026
Uplifting conversion across the acquisition funnel with personalization using contextual bandits on AWS
AWS · Oct 1, 2026