Get Started
Home
Topics
Search
Library
Research questionHow can recommenders adapt to changing preferences without trusting unreliable inputs?User preferences change, and the information used to rank items can be noisy or outdated. Some approaches learn better graph connections; others update memory while keeping model parameters fixed.
AI
AI Memory
Evaluation & Benchmarks
Information Retrieval
Machine Learning
Multimodal Models
Research Paper
Statistical Machine Learning
Latest papersRecent research connected to this question, newest first.Continual Graph Memory for Adaptive Recommendation under Intent DriftThe source studies frozen-parameter, one-pass reranking in which adaptation occurs through updates to semantic and episodic graph memories. Evidence spans several recommendation settings, including sampled-candidate reranking and metadata-rich ML-100K, and is limited to the evaluated systems and protocols.research paper · Sep 4, 2026MURAL: Multimodal Uncertainty-aware Recommendation via Adaptive edge LearningThe source contributes an adaptive multimodal graph recommender that uses interaction data and content features, with evaluations on TikTok and Amazon datasets. It also examines modality influence and corrupted inputs; the evidence does not establish behavior in live deployments.research paper · Sep 4, 2026GenCAR: Generative Counterfactual Alignment with Risk-Controlled Selection for Out-of-Distribution RecommendationThe study evaluates counterfactual candidate recovery and realized proxy false discovery proportions across diverse recommendation benchmarks. Its formal guarantees assume exchangeability and positive regression dependence, with a Benjamini–Yekutieli guarantee under arbitrary dependence; the approach also uses preference anchors, trust-radius filtering, offline language-model proposals, and conformal selection.research paper · Sep 2, 2026
Related questions
How can production recommender systems continually optimize retrieval, ranking, and serving as users and content change?How stable are recommender-system evaluation and model-selection conclusions across training seeds?How can AI agents adapt to each user’s evolving quality criteria on open-ended tasks?How can online time-series forecasters adapt to evolving, unseen patterns without costly continual parameter updates?