KwaiRec: Fully-Observed Interaction Matrix for Recommender Systems

KwaiRec provides a fully-observed user-item interaction matrix for recommender system evaluation, created by chongminggao and updated in January 2026. It addresses the missing-not-at-random (MNAR) problem by offering a dense set of observations for a specific subset of users and items, though the exact record count is not specified in the metadata.

Use Cases

• Evaluating recommendation algorithms using the fully-observed user-item interaction matrix
• Benchmarking debiasing techniques by comparing predicted scores against the dense ground truth labels
• Testing collaborative filtering models against a dense set of user-item pairs to study the impact of data density

Strengths

• Fully-observed interaction matrix for unbiased evaluation
• CC-BY-SA-4.0 license allows for sharing and adaptation
• Updated as of 2026

Limitations

• Specific column headers and row counts are not explicitly defined in the provided metadata
• Likely limited to a specific subset of users and items to maintain the fully-observed property

Provenance

Source

chongminggao via GitHub

Freshness

Last updated in January 2026.