PersonaMem-v2: 10K-100K Records for Implicit Persona and Agentic Memory

PersonaMem-v2 contains between 10,000 and 100,000 records designed for learning implicit user personas and agentic memory. Developed by researchers at the University of Pennsylvania and updated in March 2026, it addresses AI personalization for tasks with subjective goals.

Use Cases

• Fine-tuning text generation models to reflect implicit user persona features
• Implementing agentic memory retrieval systems using long-context features
• Evaluating question answering performance on tasks with subjective user goals

Strengths

• Scale of 10,000 to 100,000 records
• Permissive MIT license for research and commercial use
• Integration with Polars and Pandas libraries for tabular processing

Limitations

• Unknown column names and schema structure in the current metadata
• Potential for label noise due to the subjective nature of personalized goals

Provenance

Source

University of Pennsylvania (bowen-upenn)

Freshness

Last updated March 2026.

Geography

United States