Preference-Driven Online Adaptation for Personalized Interaction Initiation in Proactive AI Assistants

2026-08-04

AI Summary

This work addresses the challenge that proactive AI assistants struggle to accurately identify highly personalized and sparsely distributed user interaction opportunities. To this end, the paper proposes Evidence-driven Online Preference Adaptation (EOPA), which, for the first time, models interaction timing as measurable contextual evidence. EOPA leverages two types of evidence carriers—temporal preference anchors and activity prototypes—and combines user prior smoothing with uncertainty-guided scaling to fuse evidence online for timely interaction decisions. It further employs high-quality historical responses to steer LLMs toward personalized replies. Notably, EOPA requires neither LLM inference nor model retraining, substantially improving efficiency: on the ProPerSim benchmark, it boosts the F1 score for interaction timing by 19.80 points, reduces daily adaptation time from 11.41 seconds to 0.39 seconds, and lowers inference latency for both silent and interactive steps.

Abstract

AI assistants are typically reactive, relying on users to initiate interactions. Proactive assistants go beyond this paradigm by autonomously initiating interactions based on users' activity contexts. However, appropriate interaction timing is user-specific and difficult to determine in advance, while online feedback offers valuable signals for personalization. Direct feedback-driven adaptation is therefore appealing, but remains challenging due to sparse interaction-worthy moments scattered across fine-grained user states. To address the issues, we propose Evidence-driven Online Preference Adaptation (EOPA), which grounds a user's interaction-timing preferences in measurable contextual evidence through two evidence carriers: temporal preference anchors and evidence-bearing activity prototypes. At each polling step, EOPA derives temporal and activity evidence from the carriers through user-prior-smoothed evidence estimation and uncertainty-guided evidence scaling, and adaptively fuses the evidence for interaction-or-silence decisions. When interaction is selected, an LLM uses high-quality historical responses as demonstrations to generate a context-aware response that better reflects user preferences. EOPA updates its evidence carriers and decision parameters from received online feedback without LLM-based reasoning or retraining. Extensive experiments on a ProPerSim-based benchmark show that EOPA improves the interaction-timing F1 score by 19.80 points over the strongest baseline in our experiments, substantially reduces inference latency for both silence and interaction steps, and lowers the average daily adaptation time from 11.41 to 0.39 seconds.

Problem

• proactive AI assistants
• personalized interaction initiation
• online adaptation
• interaction timing
• user preferences

Innovation

• Preference-Driven Adaptation
• Proactive AI Assistants
• Online Learning
• Contextual Evidence Fusion
• Interaction Initiation