Towards Root Memories: Benchmarking and Enhancing Implicit Logical Memory Retrieval for Personalized LLMs

2026-06-22

AI Summary

This work addresses the limitations of existing personalized large language models, whose memory retrieval relies solely on semantic similarity and thus struggles to recall memories with low semantic overlap yet critical logical relevance, compounded by the absence of dedicated evaluation benchmarks. To overcome this, the paper introduces RootMem, a novel framework that pioneers the concept of “root memories.” It structures user history through distillation and integrates a large language model–based routing mechanism to fuse semantic and logical retrieval pathways, thereby explicitly modeling and activating implicit decision logic. The authors also construct IMLogic, the first high-quality benchmark tailored for evaluating implicit logical memory retrieval. Experiments demonstrate that RootMem substantially outperforms state-of-the-art retrieval baselines and consistently enhances task accuracy across multiple memory-augmented agents.

Abstract

Memory systems are essential for personalized Large Language Models (LLMs). However, existing retrieval methods in these systems primarily rely on semantic similarity, potentially missing logically critical memories with limited semantic overlap. Current benchmarks remain inadequate for evaluating this problem. To address this gap, we construct IMLogic, the first high-quality benchmark targeting implicit logical memory retrieval in long-dialogue scenarios. Motivated by this challenge, we introduce root memory, a structured, decision-preserving representation that distills reusable personalized logic from long-term user histories. We then propose RootMem, a plug-and-play framework that first distills raw histories into structured root memories and then uses an LLM-based router to activate logically relevant ones, complementing semantic retrieval with personalized decision logic. Extensive experiments demonstrate that RootMem significantly outperforms the strongest retrieval baselines and consistently boosts the accuracy of existing memory agents. Our benchmark and codes will be available at https://anonymous.4open.science/r/IMLogic-DBB3.

Problem

• implicit logical memory
• personalized LLMs
• memory retrieval
• semantic similarity
• benchmarking

Innovation

• root memory
• implicit logical retrieval
• personalized LLMs
• memory distillation
• IMLogic benchmark