Towards Realistic Personalization: Evaluating Long-Horizon Preference Following in Personalized User-LLM Interactions
This work addresses the underexplored problem of evaluating long-horizon preference following in personalized user-LLM interactions, which is crucial for developing more user-aware LLM assistants.
This paper introduces RealPref, a benchmark designed to evaluate how well Large Language Models (LLMs) follow user preferences over long-term interactions. The study found that LLM performance significantly decreases with longer context lengths and more implicit preference expressions, and that generalizing preference understanding to new scenarios is challenging.
Large Language Models (LLMs) are increasingly serving as personal assistants, where users share complex and diverse preferences over extended interactions. However, assessing how well LLMs can follow these preferences in realistic, long-term situations remains underexplored. This work proposes RealPref, a benchmark for evaluating realistic preference-following in personalized user-LLM interactions. RealPref features 100 user profiles, 1300 personalized preferences, four types of preference expression (ranging from explicit to implicit), and long-horizon interaction histories. It includes three types of test questions (multiple-choice, true-or-false, and open-ended), with detailed rubrics for LLM-as-a-judge evaluation. Results indicate that LLM performance significantly drops as context length grows and preference expression becomes more implicit, and that generalizing user preference understanding to unseen scenarios poses further challenges. RealPref and these findings provide a foundation for future research to develop user-aware LLM assistants that better adapt to individual needs. The code is available at https://github.com/GG14127/RealPref.