DP-Rec improves recommendation efficiency on long user histories
DP-Rec: Towards Dynamic Patching for Efficient Long-Sequence Recommendation
Machine LearningArtificial IntelligenceInformation Retrieval
Summary
Long lists of user activity can help online services recommend things, but looking at every single action can be slow and noisy. The authors propose a new way to group user actions into meaningful chunks, compressing the data to focus on important behavior changes. Their method, called DP-Rec, processes these chunks efficiently to predict what a user might want next while using less computing power. Tests show it works better than existing methods when computers have limited resources.
What this means in practice
- •For recommendation system engineers: Optimize recommendation models to handle very long user histories efficiently without losing accuracy using dynamic patching techniques.
- •For mobile app developers: Reduce computational costs and latency in apps that suggest products or content by using compressed user interaction sequences.
Authors
Dwipam Katariya, Thomas Caputo, Akshat Shreemali, Juan Manuel Origgi, Nikita Seleznev, Pranab Mohanty, Kalanand Mishra, Nam Nguyen, James Montgomery
Abstract
Transformers have redefined sequential recommendation by effectively modeling dynamic user behaviors and long-range dependencies. However, they remain inherently inefficient: standard architectures operate at a fixed rate, allocating comparable computation to every item in a user's history regardless of its information content. This leads to prohibitive computational overhead on long sequences and increased sensitivity to behavioral noise. To address this, practitioners often resort to lossy sequence compression, staged modeling, or truncation. This limits the model's ability to leverage the full context of long histories during inference. Inspired by the recent success of Byte Latent Transformer, we propose DP-Rec, a dynamic latent patching architecture for recommendation. DP-Rec shifts from item-level modeling to patch-level modeling by segmenting interaction sequences using contrastive entropy surprise to identify informative behavioral boundaries. A lightweight patch encoder compresses these temporally contextualized segments into a reduced set of dynamic latent behavior vectors, which are then processed by a larger latent transformer and decoded for next-item prediction. Extensive experiments show that, under constrained computational budgets, DP-Rec scales effectively to long sequences and achieves a superior efficiency-accuracy trade-off over both non-compressed and fixed-size compression baselines.