MemoryCD: Filtered Cross-Domain Amazon Reviews for Memory-Augmented LLM Evaluation

MemoryCD is a filtered subset of Amazon Reviews 2023 designed for evaluating memory-augmented large language models. The dataset contains interactions from 323 users and metadata for 167,870 unique items across four product categories: personal care, books, electronics, and home. It was created by author WZDavid and last updated on Hugging Face in May 2026.

Use Cases

• Benchmarking memory-augmented LLMs based on cross-domain user-item interaction data.
• Analyzing user behavior patterns across different product categories mentioned in the description.
• Evaluating recommendation algorithms on a dataset with 100% coverage of referenced items.
• Studying cross-domain recommendation challenges using the four provided meta files.

Strengths

• Contains metadata for 167,870 unique items, providing 100% coverage of items referenced by the 323 users.
• Includes cross-domain data from four distinct product categories: personal care (33,475 items), books (48,054 items), electronics (25,441 items), and home (60,900 items).
• Specifically filtered and curated for memory-augmented LLM evaluation, indicating a clear research purpose.

Limitations

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

Provenance

Source

Filtered subset of Amazon Reviews 2023.

Collection Method

Likely filtered and curated from the original Amazon review corpus for a specific evaluation purpose.

Time Range

Based on source, likely contains reviews from around 2023 or earlier.

Freshness

Last updated 2026-05-27 09:20:56; freshness should be verified.