Aug 2026· Proceedings of the 32nd ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.2· 0 citations· 13 references
TL;DR
PersMem is proposed, a user profile memory framework for LLM personalization that addresses three key questions: what user information to store, how to organize it, and how to use it effectively during generation that consistently improves personalization effectiveness while reducing prompt length.
Abstract
Large language models (LLMs) are increasingly used in personalized applications, where responses must align with individual user preferences, histories, and profiles. A common approach is to inject user information into the prompt at inference time. However, existing methods typically rely on flat profile representations and similarity-based Top-K retrieval, which often introduce noisy or marginally relevant context under limited prompt budgets. In this paper, we propose PersMem, a user profile memory framework for LLM personalization that addresses three key questions: what user information to store, how to organize it, and how to use it effectively during generation. First, we construct a structured user profile memory by extracting entities and concepts from user profiles. Second, we organize the memory into a hierarchical structure that separates short-term interactions from long-term knowledge, enabling scalable and coarse-to-fine retrieval. Third, we introduce a learning-based memory gating mechanism that adaptively selects long-term memory units under prompt budget constraints and directly optimizes answer quality through evaluated signals. Extensive experiments on multiple personalization tasks demonstrate that our approach consistently improves personalization effectiveness while reducing prompt length, compared with existing prompt-based and retrieval-based baselines.
Large language model (LLM) agents increasingly use external memory systems to support personalization by drawing on long and evolving interaction histories, in which user preferences may be distributed across time, change with context, and conflict with earlier evidence. However, existing systems face three limitations: fixed-turn, fixed-token, or session-based boundaries can mix unrelated dialogue or split an event from its causes, decisions, and outcomes; storing multiple pieces of user information from the same interaction as a single memory binds together items that serve different functions and should be independently retrievable; and treating the current task as a single top-$k$ retrieval query can return fragments that are individually relevant but fail to jointly capture preference evolution, temporal validity, and contextual applicability. We introduce \textsc{QUMem}, a structured memory framework for query-conditioned user-state inference. \textsc{QUMem} first segments interaction histories into variable-length episodes according to semantic continuity, then decomposes each episode into independently retrievable factual, preference, and transferable insight memories while preserving temporal positions and source evidence. At inference time, three sequential agents identify task-specific information needs, plan multi-query retrieval over the typed memory stores, and jointly infer a temporally and contextually valid user state for downstream response generation. \textsc{QUMem} achieves state-of-the-art performance on both PersonaMem and KnowU-Bench, demonstrating the effectiveness of query-conditioned user-state inference for long-term personalization.
Heng Wang, Yifei Li, Lingling Zhang et al.· 0 citations
Large Language Models (LLMs) have demonstrated remarkable ability in generating personalized content by leveraging user histories and contextual cues. However, most existing personalization approaches rely on implicit representations within model parameters, making it difficult to interpret user-specific preferences or effectively handle long-context dependencies. To address these challenges, we propose PrefReward, a novel preference-aware generative framework that explicitly models user styles through a structured preference matrix and integrates it into the decoding process as a reward signal. PrefReward consists of two stages: (1) extracting a user-specific preference matrix that summarizes individual stylistic tendencies, and (2) using the matrix to guide generation via a KL-divergence-based reward function. Experiments on the LongLaMP dataset show that PrefReward outperforms non-personalized and retrieval-based baselines in both generation quality and personalization interpretability.
Yue Wu, Chengbing Wang, Yimeng Bai et al.· 0 citations
This work proposes continuous-query LMLM (CO-LMLM), where the KB pairs continuous keys with textual knowledge values, a significant departure from prior reliance on relational KB and queries, which outperforms prior LMLMs and vanilla LLMs in both perplexity and factual precision.
Yair Feldman, Lin-Xi Zhao, Nathan Godey et al.· 0 citations
Personalized generation systems retrieve user history by request--memory relevance and inject it into the model context. Yet relevant history may concern the wrong preference aspect, duplicate public information, or provide insufficient support. We argue that personal memory should be used only when it adds utility beyond a public-only response. We propose TRACE-Memory, a two-stage framework for selective personalization. Stage 1 queries for user-specific information missing from the request and public context, then retrieves a coverage-oriented candidate pool. Stage 2 admits a compact subset of source-traceable evidence units, or the empty set, according to response-level incremental utility. We progressively train the query-generation and evidence-admission policies through structured SFT initialization, reduced-space stage-wise GRPO warm-up, and nested multi-sample Joint GRPO. Across 4,500 Controlled and Natural tasks from Goodreads, Amazon Reviews, and Reddit, TRACE-Memory consistently outperforms random and lexical memory use, improves over semantic retrieval, remains competitive with frontier-LLM memory pipelines as local generator capacity increases, and conditions evidence admission on public-context sufficiency, supporting selective rather than default personalization.
Jing Wang, Zhuyuan Wang, Yifan Guo et al.· 0 citations
Conversational search systems help users satisfy complex information needs through natural language interactions, yet incorporating user preferences into ranking remains challenging. Existing rewrite-then-rerank pipelines capture topical relevance but struggle with fine-grained constraints such as negative preferences or formatting requirements. Instruction-following retrieval approaches are promising for enforcing such constraints, yet their use in personalised conversational search remains underexplored, since ''instructions'' within this context are implicit and situated within user history and profiles, rather than being explicitly stated. We show that instruction-following models can assist with complex queries but introduce noise and latency on simpler keyword queries. To address this issue, we propose Adaptive Personalised Reranking (APR), a framework that routes queries based on intent. APR uses efficient similarity-based reranking for simple queries and dynamically generates tailored instructions to guide an instruction-following reranker for constraint-heavy contexts. Oracle analysis on TREC iKAT 2023 and 2024 shows that instruction-following provides a ''rescue'' potential for hard queries. We also show that APR trained with synthetic data performs competitively against strong baselines such as MonoT5 while offering promising new research avenues.
Shen Dong, I. Ounis, Debasis Ganguly· Annual International ACM SIG...· 0 citations
Large language models can improve recommendation quality by reasoning explicitly over user history and candidate items - for example, extracting a user's preferences or explaining why one item fits better than another - rather than mapping history directly to a ranked list. This reasoning, however, is expensive to repeat on every ranking request and, once produced, is typically consumed once and discarded, leaving it neither reusable across future requests nor easy to inspect or correct as user tastes drift. Our insight is that reasoning does not need to be regenerated at every call if it can instead be compressed once into a compact, structured memory that a lightweight model retrieves from. We propose rEDMRec, which distills a teacher LLM's reasoning into four typed, editable experience channels - long-term preference, short-term context, item-perception, and counterfactual hard-negative comparisons - maintained by an LLM memory controller that performs Add/Delete/Modify/Keep operations and refines entries via K-agent debate. A lightweight student LLM then ranks candidates purely by retrieving from this memory, without invoking the teacher again, decoupling online inference cost from reasoning depth. Across ML-1M, Amazon Beauty, and Steam and ten student backbones, rEDMRec improves HR@1 over zero-shot, few-shot, and RAG on every backbone, and over GraphRAG on most backbones, with Impv up to 13.3% vs. the second-best baseline on ML-1M. Channel ablations show that short-term context is the only channel that helps consistently across capacity tiers, whereas long-term, item-perception, and counterfactual contributions are capacity-dependent (and can reverse on the strongest students); debate-based memory optimization lowers bank duplication by 7.4 percentage points while raising downstream HR@1 by up to +0.029 over six optimization epochs.
Minh Hoang Nguyen, Tung Le, Huy-Tien Nguyen· 0 citations