Recommender systems now shape what people know, choose, and experience. In the generative AI era, recommendation is becoming an ongoing relationship among people, agents, and platforms. User-facing agents can interpret goals, maintain memory, use tools, and act across multiple steps, while agentic workflows can revise, evaluate, and deploy recommendation models. These capabilities promise more adaptive support, but they may also shift control over how preferences are represented, decisions are executed, and systems evolve. Human-centered recommendation must therefore ask whose goals agents serve, how people can inspect and correct their behavior, and whether their effects remain beneficial over time. Organized around agency, alignment, and long-term value, the 3rd Workshop on Human-Centered Recommender Systems (HCRS@WWW 2027) brings together recommender systems, information retrieval, human-computer interaction, responsible AI, and social computing to develop methods, evaluations, and safeguards that keep agent-mediated recommendation accountable to people.
Important Dates
- · Paper submission deadline January 4, 2027
- · Author notification February 1, 2027
- · Camera-ready February 16, 2027
- ★ Workshop day TBD · during The Web Conference 2027 (May 10–14, 2027)
Timezone: Anywhere On Earth (UTC−12) for paper deadlines; session times will be announced in IST (Irish Standard Time, UTC+1).
Workshop Objectives
This workshop encourages new theoretical frameworks, interdisciplinary approaches, and perspectives for human-centered recommender systems in the generative AI era. We advocate rigorous evaluation through user studies, simulation-based protocols, and off-policy metrics that move beyond engagement to capture satisfaction, trust, and long-term welfare. Participants will explore how large language models and autonomous agents can strengthen, rather than undermine, meaningful human control over recommendation. Rather than centering on agent architectures and system capabilities, the workshop foregrounds the human stakes of this transition: whose goals are served, who retains oversight, and who bears the consequences.
Call for Papers
The workshop is organized around three thematic pillars that address the human stakes of agent-mediated recommendation:
-
Agency: Human control over goals and delegation.
Extending human capabilities while preserving meaningful
control over preference representation, memory, tools,
permissions, and consequential actions. Topics include:
- Personal agents for discovery and decision support.
- Goal expression, revision, and mixed-initiative interaction.
- User control of memory, tools, data, and permissions.
- Consent, contestability, correction, and human override.
- Recommendations interpreted jointly by people and agents.
- Platform access, competition, and user choice.
-
Alignment: Systems aligned with human intentions and values.
Keeping user-facing agents and self-evolving recommendation
systems responsive to human intentions through feedback,
verification, and accountable oversight. Topics include:
- Preference elicitation under ambiguity, conflict, and drift.
- Human and agent feedback, rubrics, and reward models.
- Verified agent harnesses for recommender-system development and deployment.
- Provenance, revision, deletion, and contamination of user and agent memory.
- Safe tool use, reflection, coordination, and self-improvement in recommendation pipelines.
- Reward hacking, capability drift, and failed exploration in adaptive recommender systems.
- Reproducibility, privacy, fairness, and accountable audits of recommender systems.
-
Long-term Value: Benefits for people and ecosystems over time.
Moving beyond one-step relevance and short-term engagement
to assess cumulative effects on users, creators, communities,
and platforms. Topics include:
- Long-horizon user modeling and preference formation.
- Satisfaction, learning, well-being, and autonomy.
- Continual adaptation and delayed feedback.
- Value for users, creators, and communities.
- Simulation-based and longitudinal evaluation.
- Process evaluation, attribution, and red teaming.
- Manipulation, dependency, filter bubbles, and harm.
For detailed submission guidelines, visit the Call for Papers page.
Submission Guidelines
All submitted papers must be formatted as a single PDF document according to the ACM Web Conference 2027 template.
- Length: 4–8 pages, with unlimited pages allowed for references
- Review Policy: Single-blind review
- Evaluation Criteria: Relevance to the workshop, originality, soundness, clarity, and potential value to the human-centered recommendation community
- Best Paper Award: Will be presented at the workshop
Submission Portal: TBA — the OpenReview group will be created following workshop proposal acceptance.
Program Highlights
The workshop is planned as an in-person half-day afternoon session featuring:
- Keynote Talks: Two 30-minute keynote talks (including Q&A) delivered by invited speakers from academia and industry, providing insights into the latest advancements and future directions.
- Paper Sessions: Two oral presentation sessions showcasing accepted research on agency, alignment, and long-term value.
- Poster Session: A dedicated poster session for accepted poster papers, with 10 poster boards requested from the conference.
- Best Paper Award & Closing Remarks: The workshop closes with the Best Paper Award and closing remarks.
See the Program page for the tentative afternoon schedule.
