LUCid: Redefining Relevance For Lifelong Personalization

2026-04-28

AI Summary

Existing lifelong personalization approaches rely on semantic similarity to assess relevance, struggling to extract critical user information from interaction histories that are topically unrelated yet contextually relevant. To address this limitation, this work introduces the LUCid benchmark—the first user-centric framework for evaluating contextual relevance—comprising 1,936 real-world queries paired with up to 500 rounds of historical conversations. Through comprehensive experiments across multiple architectures, the study systematically evaluates state-of-the-art models, including Gemini-3-Flash, GPT-5.4, and Claude Haiku, on retrieval recall and response alignment. Results reveal a stark performance gap: on the most challenging samples that are semantically distant yet contextually relevant, model recall approaches zero and response alignment hovers around 50%, exposing a fundamental misalignment between current relevance modeling capabilities and real-world user needs.

Abstract

Current approaches to lifelong personalization operationalize relevance through semantic proximity, causing them to miss essential user information from topically unrelated interactions. To address this gap, we introduce LUCid, a benchmark designed to measure situational user-centric relevance in personalization. The benchmark consists of 1,936 realistic queries paired with interaction histories from up to 500 sessions. Across multiple architectures, our experiments show significant performance collapse when relevant context must be surfaced from semantically distant history: retrieval recall drops to near zero on the hardest instances, and response alignment remains near 50% even for state-of-the-art models such as Gemini-3-Flash, GPT-5.4, and Claude Haiku. These results expose a fundamental mismatch between the notion of relevance encoded by current systems and the situational relevance required for personalization, with direct implications for robustness and safety when critical user attributes remain undetected. LUCid enables the systematic evaluation of whether current models can surface situationally-relevant user information from previous interactions, and serves as a step toward realigning personalization with user-centered relevance.

Problem

• lifelong personalization
• situational relevance
• semantic proximity
• user-centric relevance
• interaction history

Innovation

• lifelong personalization
• situational relevance
• user-centric benchmark
• semantic proximity
• interaction history