From Atom to Community: Structured and Evolving Agent Memory for User Behavior Modeling

TL;DR

This paper introduces STEAM, a structured, evolving memory framework for user behavior modeling that improves recommendation accuracy by capturing multi-faceted interests and leveraging collaborative community signals.

Contribution

STEAM reorganizes user preferences into atomic, linked memory units, incorporates community-based signal propagation, and introduces adaptive evolution mechanisms for better user interest modeling.

Findings

Outperforms state-of-the-art baselines in recommendation accuracy.

Enhances simulation fidelity and diversity.

Effectively captures evolving and multi-faceted user interests.

Abstract

User behavior modeling lies at the heart of personalized applications like recommender systems. With LLM-based agents, user preference representation has evolved from latent embeddings to semantic memory. While existing memory mechanisms show promise in textual dialogues, modeling non-textual behaviors remains challenging, as preferences must be inferred from implicit signals like clicks without ground truth supervision. Current approaches rely on a single unstructured summary, updated through simple overwriting. However, this is suboptimal: users exhibit multi-faceted interests that get conflated, preferences evolve yet naive overwriting causes forgetting, and sparse individual interactions necessitate collaborative signals. We present STEAM (\textit{\textbf{ST}ructured and \textbf{E}volving \textbf{A}gent \textbf{M}emory}), a novel framework that reimagines how agent memory is organized and updated. STEAM decomposes preferences into atomic memory units, each capturing a distinct interest dimension with explicit links to observed behaviors. To exploit collaborative patterns, STEAM organizes similar memories across users into communities and generates prototype memories for signal propagation. The framework further incorporates adaptive evolution mechanisms, including consolidation for refining memories and formation for capturing emerging interests. Experiments on three real-world datasets demonstrate that STEAM substantially outperforms state-of-the-art baselines in recommendation accuracy, simulation fidelity, and diversity.