Source: arXiv

Paper

Measuring Cross-Cultural Style Diffusion Through Era Classification: US and Korean Popular Music — Dasol Lee, Minhee Lee, Seonguk Ju, Daewoong Kim, Harin Lee, Dasaem Jeong (arXiv 11 Aug 2026; accepted at ISMIR 2026)
ID: arXiv:2608.10980

What it does

The authors treat “when does Korean chart audio sound like on a US timeline?” as a measurable MIR problem. They train era-classification CNNs from scratch on Billboard Hot 100 audio (22k tracks), then apply those models to Korean Melon chart songs (2.2k tracks). The gap between inferred Billboard era and actual Melon chart-entry year becomes a cross-cultural era offset — a proxy for how far local charts sit behind (or near) a US-learned chronological axis.

Design choices matter for literacy, not only for accuracy. Tracks primarily credited to Korean acts were removed from the Billboard training set (secondary features by Korean artists were kept) so contemporary K-pop did not dominate the US-era axis. Melon coverage is capped at 2009, before full-scale bidirectional K-pop globalization. Labels sit on a chronological era axis, not genre or mood tags — avoiding taxonomies that disagree across listener cultures.

The breakdown

On Melon songs from the 1960s–1980s, the Billboard-trained models consistently place Korean chart audio earlier on the US timeline — median lag about four to five years — while the same models stay unbiased on held-out Billboard audio. From the 1990s through the 2000s, that offset halves to roughly two to three years and holds. The paper reads the pattern as progressive synchronization of locally adopted global pop styles, not as a claim that every Korean hit copied a US single. Reverse inference and multi-architecture / multi-seed checks support that the narrowing is not a one-run fluke.

Operationally: first chart entry dates define eras; artist-aware splits reduce voice leakage; hierarchical decade-to-year heads keep ordinal time structure. The overseas reader’s takeaway is the timeline story, not the loss function.

Why readers outside the lab should care

If you teach K-pop history, program festivals, or brief investors on “when Korea caught up sonically,” this paper replaces vibe timelines with a chart-audio lag curve. Expats who grew up on 1990s–2000s Melon-era hits should expect those years to sound closer to contemporaneous Billboard than 1960s–80s Korean charts do — a quantitative echo of the Seo Taiji–era industrial shift the authors situate in related work. Cultural journalists and playlist curators can treat “four-to-five years behind, then half that” as a frame for archival listening, not a genre stereotype.

What travelers and expats should watch

• Do hear pre-1990s Korean chart pop as lagging a US-era axis by ~4–5 years in this study’s terms — not as “wrong” or derivative by default.
• Do expect the gap to narrow to ~2–3 years for 1990s–2000s Melon entries under the same models.
• Don’t extrapolate past 2009; the Melon corpus stops there by design.
• Don’t read genre-similarity claims into the result — the axis is chronological era, not shared genre labels.
• Expect the US training domain to exclude Hot 100 tracks primarily credited to Korean acts (not every Korean credit); reverse flow after 2010s globalization is out of scope here.

Context

Read this as a style-diffusion clock, not as a ranking of which country’s pop was “better.” Korelay frame: the story is temporal alignment between chart cultures, not “Korea copied Billboard.” Carry that into overseas listening guides and K-culture briefings — sync speed is the claim, not authenticity score.

Source

arXiv:2608.10980 — abstract and introduction cited (ISMIR 2026 accepted); open the OA PDF for method, offsets, and figures. Do not republish the PDF.