Differences Between Niche and Mainstream Trends
A new word or concept can spread at an exponential rate through a society, but a recent study of words appearing in Japanese blogs found that a slower, subexponential early growth phase is surprisingly common. Now an improved mathematical model has shown that this subexponential phase typically occurs for words whose origins are tied to specific social communities [1]. The researcher who developed the model also found that real-world events cause words to abruptly shift between different phases of growth with different rates of spread. The new model could provide a practical framework to help interpret complex growth curves for words and other innovations.
For decades, many social scientists have supposed that a new idea spreads rapidly, as one person tells other persons, who each tell other persons, and so on. This exponential growth eventually slows and saturates as the pool of uninformed individuals becomes depleted. Three years ago, however, computational social scientist Hayafumi Watanabe of Seijo University in Japan found important deviations from this picture, showing that the usage frequency of many new words in Japanese blog posts increased in a slower-than-exponential way [2]. While epidemiologists have occasionally seen this pattern in the early stages of disease outbreaks—often when spatial constraints hold an infectious agent back—Watanabe revealed that subexponential growth occurs in online behavior and often persists throughout a word’s entire diffusion period.
Such slow growth is common in social contexts, but it is often overlooked in systematic studies that tend to focus on rapidly spreading phenomena such as buzzwords, Watanabe says. “This kind of subexponential growth has not received much attention.” To learn more about the origin of this slower growth, he has now taken his analysis further, using an improved and more flexible mathematical framework to better fit the growth data.
Watanabe’s idea was to fit the growth curves for new words using a differential equation in which a parameter determines the type of growth: = 1 corresponds to exponential growth; < 1 corresponds to slower, subexponential growth. Crucially, he also allowed this parameter to take on different values across different time segments, because real-world aspects of the spreading process might sometimes change. For example, the name of a popular artist might begin to spread differently if she were to suddenly change recording labels.
To test this model, Watanabe analyzed a dataset of words appearing in more than one billion Japanese blog posts and selected a subset of 2963 distinct words showing monotonic growth over a sufficiently long period of time to allow robust statistical analysis. He found that the spreading trajectories of about 55% of these words exhibited diffusion patterns consistent with just one or two segments. And the most common value of for single-segment curves was close to 0.5, reflecting the subexponential growth common to online social diffusion.
Watanabe used the data to reveal subtle ways by which the character of a specific word influences how it spreads across society. Using a large language model (Google Gemini) to categorize different topics, he found that the parameter for a particular word tends to reflect its appeal to “in-groups.” Words linked to highly niche topics—such as local nicknames for neighborhoods or a specific subculture terminology, such as AKB48, a Japanese pop music group—tend to have small values close to 0, or even below. These low values reflect slow spreading among people in narrowly defined, preexisting communities. In contrast, other words, like a popular meme or the name of a well-known television personality, grow more rapidly (with values between 0.5 and 1) because people tend to readily pass knowledge of them on to strangers.
“The most striking aspect of the study is that the spread of new words often looks less like an epidemic explosion and more like a slow burn,” says physicist and complex systems expert Matjaž Perc of the University of Maribor in Slovenia. “The shape of a word’s rise carries a social fingerprint: Niche terms tend to circulate within tight communities, while more broadly resonant terms move outward into public conversation.”
In future work, Watanabe hopes to extend these findings beyond words in Japanese blogs using other sources of data. “This perspective may be useful not only for new words but also for understanding various social growth phenomena, such as online memes, product names, cultural trends, and the diffusion of technologies or services,” he says.
–Mark Buchanan
Mark Buchanan is a freelance science writer who splits his time between Abergavenny, UK, and Notre Dame de Courson, France.
References
- H. Watanabe, “Subexponential growth dynamics in complex systems: A piecewise power-law model for the diffusion of new words and names,” Phys. Rev. E 114, 014304 (2026).
- H. Watanabe, “Minor extensions of the logistic equation for growth curves of word counts on online media: Parametric observation of diversity of growth in society,” J. Phys. Complex. 4, 025018 (2023).