ProPerSim: Developing Proactive and Personalized AI Assistants through User-Assistant Simulation

TL;DR

ProPerAssistant is proposed, a retrieval-augmented, preference-aligned assistant that continually learns and adapts through user feedback, highlighting the promise of uniting proactivity and personalization.

Abstract

As large language models (LLMs) become increasingly integrated into daily life, there is growing demand for AI assistants that are not only reactive but also proactive and personalized. While recent advances have pushed forward proactivity and personalization individually, their combination remains underexplored. To bridge this gap, we introduce ProPerSim, a new task and simulation framework for developing assistants capable of making timely, personalized recommendations in realistic home scenarios. In our simulation environment, a user agent with a rich persona interacts with the assistant, providing ratings on how well each suggestion aligns with its preferences and context. The assistant's goal is to use these ratings to learn and adapt to achieve higher scores over time. Built on ProPerSim, we propose ProPerAssistant, a retrieval-augmented, preference-aligned assistant that continually learns and adapts through user feedback. Experiments across 32 diverse personas show that ProPerAssistant adapts its strategy and steadily improves user satisfaction, highlighting the promise of uniting proactivity and personalization.

Figures (7)

• Figure 1: Only Proactivity shows initiative but ignores preferences (steakhouse to a vegetarian); Only Personalization fits preferences but lacks initiative. Ours (Proactivity + Personalization) proactively recommends a vegetarian dinner at the right moment.
• Figure 2: Overview of the ProPerSim simulation. The assistant observes the user performing the action of enjoying a strong espresso and responds with a book recommendation. While the recommendation aligns well with the criteria for personal preference, timing, and communication & safety, it exceeds the preferred frequency, resulting in a score of 3 out of 4. Over time, the assistant improves using accumulated recommendations and evaluations.
• Figure 3: 2D projection of 32 personas based on their key attributes. Point size reflects age.
• Figure 4: Daily average recommendation scores by method, with shaded areas indicating the standard error of the mean (SEM).
• Figure 5: Results of ProPerAssistant by persona. Gray: individual personas; blue: average across all personas; green: best-performing persona; red: worst-performing persona.