Unexpected Values in played_ratio_pct Exceeding 100%
Thank you for releasing the Yambda-5B dataset.
While analyzing the listens.parquet file, I noticed that some records contain played_ratio_pct values exceeding 100%.
Could you kindly clarify whether these values are expected by design (e.g., cumulative play across sessions), or if they reflect noise in the logs? Additionally, is there a recommended strategy for preprocessing or filtering these values before downstream use?
Thank you for your question!
A played_ratio_pct greater than 100% indicates that the user rewound and replayed sections of the track—so the total time listened exceeds the original track length. These values are expected behavior and not log noise.
For most downstream tasks, if your use case is focused on unique listens up to 100% of the track, you may wish to cap these values at 100%. However, if it’s relevant to capture replays and repeated listening behavior, retaining the original values could be informative. As an additional note, in our research, we binarize listens and skips using a 50% threshold: a listen is considered valid if played_ratio_pct is at least 50%.