MEMOIR: Temporal Behavioral Memory for Recommendation Across the Preference-Drift Spectrum
2026-07-27 • Information Retrieval
Information RetrievalComputation and Language
AI summaryⓘ
The authors present MEMOIR, a system that breaks down user activity into time segments and uses a large language model to summarize behavior during each period. They tested MEMOIR on Amazon product reviews and found it performs about as well as the best existing method overall, with different strengths on different metrics. Their analysis shows that no single part of MEMOIR explains its improvement, but the gains are most clear for users with extreme changes in preferences over time. This suggests MEMOIR is especially useful for capturing how some user interests evolve, although why this happens remains unclear and is left for future study.
MEMOIRlarge language model (LLM)user interaction historytemporal window segmentationbehavioral memoryrecommendation systemsranking metricspreference driftNDCG@10contrastive loss
Authors
Younggue Bae
Abstract
We propose MEMOIR, a framework that segments user interaction histories into temporal windows, generates semantic behavioral memory for each period using an LLM, and aggregates current state, evolution direction, and predicted future into a single user representation. On the Electronics and Clothing_Shoes_and_Jewelry categories of Amazon Reviews 2023, MEMOIR is statistically tied with UniSRec, the strongest baseline, on aggregate NDCG@10 (0.0643 vs. 0.0641), splitting the four reported metrics 2-2: MEMOIR leads NDCG@10 and MRR, UniSRec leads HR@10 and HR@20. An ablation study finds that no single architectural component - the evolution-preserving contrastive loss, its directional-consistency term, or temporal window segmentation itself - individually explains much of MEMOIR's approximately 18% relative gain over ID-based SASRec; all four ablations land within 2% of the full model on aggregate NDCG@10. Stratifying test performance by a composite preference-drift score instead reveals where the gain concentrates: MEMOIR leads on ranking-quality metrics (NDCG@10, MRR) specifically among users at the high- and low-drift extremes of the distribution, while UniSRec leads the volume-oriented HR@10/HR@20 metrics across all drift strata and edges out MEMOIR on ranking quality in the middle band. We report this drift-stratified pattern, rather than the near-tied aggregate numbers or any single ablated component, as MEMOIR's most substantive and reproducible finding, and surface why it holds as an open question for future work.