Analyzing the Spotify Top 200 Through a Point Process Lens

Michelangelo Harris, Brian Liu, Cean Park, Ravi Ramireddy, Gloria Ren, Max Ren, Shangdi Yu, Andrew Daw, Jamol PenderarXiv 2019 · arXiv

Abstract

Researchers at Cornell University applied stochastic point process models to 20 months of Spotify Top 200 data to decode the "hype cycle" of popular music. By modeling song popularity as a series of jumps and exponential decays, they successfully clustered tracks into distinct behavioral groups—ranging from "mega-hits" and "late bloomers" to the unique seasonal surge of Christmas music.

Introduction

The music industry has shifted from a "buy-and-own" model to a "stream-and-access" ecosystem, where a song's success is measured by daily attention rather than one-time sales. However, the dynamics of this attention are volatile. Why do some songs debut at #1 and vanish, while others climb slowly over months?

This research moves beyond simple chart-watching by providing a rigorous mathematical framework to model the "jump-and-decay" patterns of streaming counts. The authors seek to understand how external shocks—like a music video release, a viral TikTok, or a shout-out at an awards show—quantifiably alter a song’s trajectory and longevity.

Methodology

The core of the study involves the application of Stochastic Point Processes, specifically modeling the number of streams as a non-stationary Poisson process.

  1. Intensity Function: The authors define a rate function $\lambda(t)$ that incorporates a baseline streaming rate, an initial "jump" at debut, and subsequent jumps caused by external events.
  2. Exponential Decay: Each jump is followed by an exponential decay, capturing the intuition that "hype" is highest immediately after an event and fades over time.
  3. Parameter Estimation: Maximum Likelihood Estimation (MLE) is used to find the best-fit parameters (decay rates and jump sizes) for each song.
  4. Clustering: Finally, k-means clustering is applied to the estimated decay rates and the $R^2$ (a measure of model fit) to group songs with similar "mathematical fingerprints."

Data Overview

The study utilizes a dataset scraped from the Spotify Top 200 (United States) rankings from January 1, 2017, to September 12, 2018 (approx. 20 months). This period covers daily records including rank, song title, artist, and total stream counts, capturing the commercial peak of artists like Drake and the rise of "streaming-first" hip-hop.

Analysis and Insights

The Power Law of Popularity

The relationship between a song's rank and its average daily streams follows a power-law distribution ($f(x) = ax^{-b}$). This confirms a "diminishing returns" effect: it takes exponentially more streams to move from #2 to #1 than it does to move from #200 to #199.

Figure 1: The relationship between song ranking and daily stream counts follows a power-law distribution.

Longevity vs. Rank

There is a strong correlation between a song's peak rank and its "first life" duration on the charts. Songs that hit #1 averaged nearly 250 days on the charts, while those that never cracked the top 190 often vanished within days.

Figure 4: Chart duration in days plotted against the highest ranking achieved, showing that higher-ranked songs persist significantly longer.

Behavioral Archetypes

The $k$-means algorithm identified distinct archetypes based on these mathematical patterns:

  • Mega-Hits: High-volume tracks with stable decay (e.g., "God's Plan").
  • Christmas Cluster: Tracks like Mariah Carey’s "All I Want for Christmas Is You" showed exponential growth leading up to Dec 25, followed by an immediate crash.
  • Late Bloomers: Songs like "1-800-273-8255" that experienced massive jumps months after release due to external performances (e.g., the VMAs).

Conclusion and Future Work

While the current model uses a simple exponential decay, the authors acknowledge that some songs exhibit "heavy-tailed" behavior (decaying slower than an exponential), which might require more complex power-law decay functions. Future work could investigate the "long tail" of independent music that never reaches the charts or explore the "optimal external event strategy"—modeling how to perfectly time remixes and videos to maximize a song's streak at #1, as seen with artists like Lil Nas X.

Summary updated

Abstract

Every generation throws a hero up the pop charts. For the current generation, one of the most relevant pop charts is the Spotify Top 200. Spotify is the world's largest music streaming service and the Top 200 is a daily list of the platform's 200 most streamed songs. In this paper, we analyze a data set collected from over 20 months of these rankings. Via exploratory data analysis, we investigate the popularity, rarity, and longevity of songs on the Top 200 and we construct a stochastic process model for the daily streaming counts that draws upon ideas from stochastic intensity point processes and marked point processes. Using the parameters of this model as estimated from the Top 200 data, we apply a clustering algorithm to identify songs with similar features and performance.
BibTeX
@misc{harris2019analyzing,
  author = {Michelangelo Harris and Brian Liu and Cean Park and Ravi Ramireddy and Gloria Ren and Max Ren and Shangdi Yu and Andrew Daw and Jamol Pender},
  title = {{Analyzing the Spotify Top 200 Through a Point Process Lens}},
  year = {2019},
  eprint = {1910.01445},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1910.01445},
}
Figure