EACL 2024 Workshop on Personalization of Generative AI Systems
Overview
The EACL 2024 Workshop on Personalization of Generative AI Systems was held in conjunction with EACL 2024 (the annual meeting of the European Chapter of the Association for Computational Linguistics). Its proceedings are archived on the ACL Anthology:
- Workshop event page and proceedings (ACL Anthology)
- Resource type: Conference / Workshop
- Section: Conferences, Workshops
- Personalized text generation: adapting tone, content, and style to individual users or audiences.
- User modeling for LLMs: encoding user history, preferences, and context into generation pipelines.
- Adaptive dialogue and assistants: conversational systems that learn user state over multiple turns or long-term interactions.
- Customization and steering of foundation models: prompting, fine-tuning, retrieval, and other techniques for tailoring model behavior.
- Evaluation of personalization: measuring whether personalization actually improves user experience.
- Privacy, fairness, and safety: risks introduced when models memorize or infer personal attributes.
- 2025 SIGIR Workshop on eCommerce
- Activate Activate
- CIKM 2024 1st Workshop on Multimodal Search and Recommendations
- Haystack Haystack
- ICDM MMSR 2025
- KDD 2024 Workshop on Generative AI for Recommender Systems and Personalization
- Original source: EACL 2024 Workshop on Personalization of Generative AI Systems, ACL Anthology.
Scope and Themes
The workshop addresses how generative AI systems — particularly large language models (LLMs) — can be adapted to individual users, communities, and use cases rather than producing generic, one-size-fits-all outputs. Themes relevant to this area include:
System Perspective
From a search and recommendation systems viewpoint, personalization of generative systems intersects with the modern retrieval-and-generation stack:
1. Input and representation — queries, documents, and user context encoded as dense/sparse representations or structured prompts. 2. Core modules — retrievers, re-rankers, planners, memory components, and tool interfaces combined per task. 3. Learning strategies — supervised fine-tuning, contrastive learning, distillation, reinforcement learning, and synthetic data generation. 4. Inference strategies — single-pass retrieval, iterative retrieval, parallel sub-queries, early stopping, and compute budgeting.
Insights for Search / Recommendation / Personalization
1. Architecture: cascaded retrieval + re-ranking + generation remains mainstream, while agentic paradigms increasingly treat retrieval count and policy as learnable objects. 2. Data: high-quality instruction data and click/session logs are both critical; synthetic data requires care against knowledge leakage and distribution shift. 3. Evaluation: the gap between offline metrics and online satisfaction is widening; LLM-as-judge should be cross-validated with human evaluation. 4. Product: latency, cost, interpretability, and safety are hard constraints for industrial deployment — optimizing academic benchmarks alone is insufficient.
Limitations and Notes
Common limitations in this research space include experiment scale bounded by GPU budgets, benchmarks that diverge from real user distributions, English-centric data limiting cross-lingual generalization, and safety risks when agentic systems operate on the open web. Future directions include more efficient test-time compute allocation, deeper integration with knowledge graphs and structured databases, and causal/fairness constraints for recommendation settings.