ReMem: Rethinking Perception and Memory in Long-Context Recommendation Agents
What it is
ReMem is a recommendation agent framework that uses OCR-based multimodal perception to interpret item pages from screenshots and a time-evolving dynamic memory mechanism to handle long user histories. It employs a chunk-wise sequential memory update strategy to maintain bounded context length while preserving evolving user preferences, along with a multi-memory GRPO variant for training.
The problem it solves
Existing recommendation agents suffer from brittle item perception due to noisy HTML parsing and inefficient long-context reasoning over extended user histories, leading to poor performance and high computational costs.
What you could use it for
- Autonomous recommendation agents that need to process long user interaction histories
- Cross-platform recommendation systems requiring robust item perception from heterogeneous web pages
- Long-context reasoning tasks where standard autoregressive models struggle with extended traces
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.