Patterns and Dynamics of Netflix TV Show Popularity

TL;DR

This paper investigates how global Netflix TV show popularity evolves across 71 countries over 822 days, focusing on cross-country directional patterns. It employs an information-theoretic toolkit—Shannon entropy $H(X_k)$, mutual information $I(X_{k_1},X_{k_2})$, and KL divergence $D_\text{KL}$—to quantify within-country diversity, between-country similarity, and temporal asymmetry in top-ranked shows. Key findings include regional clustering with higher dynamism in North America and Europe, persistent trends in East/Southeast Asia, and directional spread from ESA and CSA toward NAPE, with Korea/Thailand driving intra-ESA; Singapore often acts as a global connector. The framework provides actionable insights for content localization and distribution strategies, while acknowledging data limitations and platform-driven effects that may shape observed dynamics.

Abstract

The rise of platforms like Netflix has expanded the possibility for audiences worldwide to watch the same content simultaneously, motivating research on cross-country media consumption. We investigate the global dynamics of media consumption by analyzing daily top-ranked Netflix TV shows across 71 countries over a span of 822 days. Using an information-theoretic framework, we measure diversity, similarity, and directional relationships in consumption trends using Shannon entropy, mutual information, and Kullback-Leibler (KL) divergence. According to Shannon entropy analysis, North America and Europe have highly dynamic viewing preferences, whereas East and Southeast Asia (ESA) display more persistent trends, with the same shows often dominating for long periods. Mutual information identifies clear regional clusters of synchronized consumption, with particularly strong alignment among ESA countries. To analyze temporal patterns, we introduce a KL-based asymmetry measure that captures directional patterns between countries, applicable to both inter- and intra-regional pairs. This analysis reveals distinct pathways of content spread. We find inter-regional patterns from ESA and South America toward North America and Europe, and intra-regional signals from Korea and Thailand to other ESA countries. We also observe that ESA trends reaching other regions often originate from Singapore. These findings offer insight into the temporal structure of global content spread and highlight the coexistence of global synchronization and regional independence in streaming media preferences.

Figures (7)

• Figure 1: Distribution of the number of distinct shows that were top-ranked over different time intervals. The $x$-axis indicates the total number of days (out of the 822-day observation period) that a show remained at the top rank (not necessarily consecutive), and the $y$-axis indicates how many different shows fall into that category. (a) presents the distribution at daily intervals from day 1 to day 20. (b) presents the distribution at 10-day intervals from day 20 to day 150. See text for details.
• Figure 2: (a) shows the grouping of countries by Shannon entropy, as defined by Eq. (\ref{['e.se']}), displayed in three distinct colors. Red indicates high entropy values (above 5.8), green represents medium entropy values (5.5–5.8), and blue denotes low entropy values (below 5.5). (b) shows the entropy values by country, corresponding to the color coding in (a). The labels 'CSA', 'NAPE', and 'AME' denote the 'Central and South America Group', 'North America and Europe Group', and 'Asia and Middle East Group', respectively. Within each group, countries are ordered by longitude, as detailed in Appendix A.
• Figure 3: (a) shows the mutual information between countries, represented by grayscale shading where darker shades indicate higher values. (b) presents the mutual information between the global dataset and each country. The labels 'CSA', 'NAPE', and 'AME' denote the 'Central and South America Group', 'North America and Europe Group', and 'Asia and Middle East Group', respectively. Countries are ordered by longitude within each group, as detailed in Appendix A. The blue dashed line in (a) marks the boundary dividing the AME group, with western AME countries sharing higher mutual information with NAPE countries compared to East and Southeast Asia (ESA). In (b), red stars highlight ESA countries with low mutual information with the global dataset. Blue squares indicate Colombia and Bolivia, which also exhibit low mutual information with the global dataset.
• Figure 4: Average KL divergence $D_\text{KL}^{k_1,k_2}(\Delta)$ between country pairs $k_1$ and $k_2$, with a time shift $\Delta$ from 0 to 30 days. The pairs are selected from 10 representative countries among the 72 analyzed: 'WO' (World), 'BR' (Brazil), 'CA' (Canada), 'CO' (Colombia), 'FR' (France), 'HK' (Hong Kong), 'JP' (Japan), 'KR' (Korea), 'TH' (Thailand), and 'US' (United States).
• Figure 5: Comparison of $\Delta^{k_1,k_2}_\text{diff}(D_\text{th})$ values defined in Eq. (\ref{['e.deld']}) between country pairs. Each cell shows how well the top-ranked show list of the column country ($k_2$) explains that of the row country ($k_1$), relative to the reverse. Threshold values $D_\text{th}$ are set to (a) 11.5, (b) 11, (c) 9, (d) 8, (e) 6, and (f) 4. The group labels 'NAPE', 'AME', and 'CSA' refer to the 'North America and Pan-Europe Group', 'Asia and Middle East Group', and 'Central and South America Group', respectively. Countries within each group are ordered as listed in Appendix A. See main text for details on the color scheme and interpretation.