Latent neuro symbolic reasoning improves personalized long term memory use
Disentangling Long-Term Memory via Latent Neuro-Symbolic Reasoning
Artificial IntelligenceComputation and LanguageMachine Learning
Summary
Personalizing digital assistants over time is hard because memories from past interactions are messy and mixed together. The authors present a new method called LGM that organizes these long-term memories into simpler, meaningful pieces inside a hidden space, instead of fixed graphs. This helps the system understand and use personal preferences better based on the current question or task. Their tests show LGM does better than existing methods at remembering and responding to people’s unique likes and behaviors.
What this means in practice
- •For mobile app developers: Build personalized assistants that dynamically interpret user history to provide accurate, context-aware responses over long periods.
- •For customer support teams: Create systems that better remember previous customer interactions to tailor responses based on explicit and implicit preferences.
Authors
Cai Ke, Xinghao Chen, Xiaoyu Shen, Keyu Chen, Siyu An, Junnan Dong, Ruifeng Xu, Ruizhi Qiao, Xing Sun
Abstract
Personalized agents are required to reason over long-term history interactions to infer both explicit preferences and implicit behavioral evidence. While early flat retrieval methods score memory fragments independently and neglect the distributed information, current structured memory frameworks rely on query-agnostic static graphs that fail to capture the context-dependent relations. Crucially, raw textual memories are inherently entangled and noisy, making fine-grained personalization and cross-session reasoning computationally prohibitive. To this end, we present LGM, a novel neuro-symbolic framework that shifts long-term memory disentanglement into a continuous latent space. Specifically, (i) instead of persisting fixed graphs, we design a tailored latent graph construction with a sparse autoencoder. Subject to each query, it maps historical interactions into latent memory nodes and disentangles the memory traces into sparse concept activations, dynamically synthesizing query-aware relational edge weights. (ii) A graph encoder then treats the query embedding as a conditioning preference to direct non-linear message passing across the task-specific latent subgraph. This yields a highly expressive memory representation for effective activations. Extensive experiments on long-term personalization benchmarks demonstrate that LGM significantly outperforms state-of-the-art baselines in capturing both explicit and implicit preferences while enabling personalized responses.