CUPID: Evaluating Personalized and Contextualized Alignment of LLMs from Interactions

2025-08-03

AI Summary

Existing LLM personalization methods assume static, global user preferences, neglecting their context dependence and dynamic evolution. Method: We introduce CUPID, the first benchmark for context-aware personalized alignment evaluation, comprising 756 human-annotated multi-turn dialogues grounded in real user feedback to model context-sensitive preferences. Using precision and recall, we quantitatively measure models’ ability to identify and respond to historical contextual preferences. Contribution/Results: We systematically evaluate ten mainstream LLMs and find that current models exhibit weak contextual preference recognition (precision <50%, recall <65%), revealing a critical deficiency in dynamic preference reasoning. CUPID establishes the first standardized, reproducible benchmark and methodology for evaluating and advancing context-aware personalized alignment of LLMs.

Technology Category

• Machine Learning: Learning Preferences or Rankings
• Humans and AI: Learning Human Values and Preferences
• Knowledge Representation and Reasoning: Preferences

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Personalized, context-aware and across-device search
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Personalization of Large Language Models (LLMs) often assumes users hold static preferences that reflect globally in all tasks. In reality, humans hold dynamic preferences that change depending on the context. As users interact with an LLM in various contexts, they naturally reveal their contextual preferences, which a model must infer and apply in future contexts to ensure alignment. To assess this, we introduce CUPID, a benchmark of 756 human-curated interaction session histories between users and LLM-based chat assistants. In each interaction session, the user provides a request in a specific context and expresses their preference through multi-turn feedback. Given a new user request and prior interaction sessions, our benchmark assesses whether LLMs can infer the preference relevant to this request and generate a response that satisfies this preference. With CUPID, we evaluated 10 open and proprietary LLMs, revealing that state-of-the-art LLMs struggle to infer preferences from multi-turn interactions and fail to discern what previous context is relevant to a new request -- under 50% precision and 65% recall. Our work highlights the need to advance LLM capabilities for more contextually personalized interactions and proposes CUPID as a resource to drive these improvements.

Problem

• Evaluating dynamic user preferences in LLM interactions
• Assessing contextual alignment from multi-turn feedback
• Benchmarking LLMs on personalized response generation

Innovation

• Dynamic preference inference from multi-turn interactions
• Contextual alignment using session histories
• Benchmarking LLMs with personalized feedback data