DynamicMem: Synthetic Multi-Month User Trajectories for Memory System Benchmarking

DynamicMem provides synthesized multi-month user trajectories for evaluating long-horizon memory systems. At ordered checkpoints, systems must reconstruct a user's current state and act on it. The benchmark was created by xiewenya and includes code and baselines on GitHub.

Use Cases

• Benchmarking state completion models based on synthesized user trajectories.
• Evaluating personalized service algorithms based on reconstructed user states.
• Training memory systems to handle long-horizon, multi-event sequences.

Strengths

• Synthesized trajectories span multiple months, providing a long-horizon testbed.
• Includes tasks for both state reconstruction and action generation, offering a multi-faceted benchmark.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

huggingface

Collection Method

Synthesized user trajectories.

Freshness

Last updated 2026-06-12 20:53:40; freshness should be verified.