Quantify autoplay contamination in session co-occurrence data

Determine the proportion of scrobbles originating from client-side autoplay or platform radio rather than direct listener selection, and assess the resulting effect on the learned song co-occurrence structure.

Background

The preprocessing heuristic detects mechanically uniform bot-like playback but cannot distinguish naturalistically timed autoplay from tracks selected directly by listeners. Consequently, some of the structure used to train Song2Vec may reflect recommendation logic from originating platforms. The available Last.fm API response lacks playback-source information, so the magnitude of this contamination is unresolved and would require client-side instrumentation or indirect comparison with documented platform sequences.

References

The magnitude of this effect cannot be determined from the present dataset.

— Project Qualia: Recovering Experiential Music Structure from Session Co-occurrence Data  (2609.10862 - Mohammed et al., 9 Sep 2026) in Section 5.2, “Autoplay and the Provenance of Session Structure”