🧬 biology

Circadian Modulation of Semantic Exploration in Social Media Language

By analyzing large-scale Reddit data, this study demonstrates that human semantic exploration exhibits a robust circadian rhythm characterized by a morning peak in local linguistic diversity and an afternoon peak in global topic accumulation, revealing a distinct cognitive dimension of language use driven by biological rhythms rather than mood.

Original authors: Vuong Hung Truong, Mariana Gabrielle Cangco Reyes, Masatoshi Koizumi, Jihwan Myung

Published 2026-01-22

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Vuong Hung Truong, Mariana Gabrielle Cangco Reyes, Masatoshi Koizumi, Jihwan Myung

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

Imagine your brain as a vast library. Every time you post on social media, you're picking a book from that library to share with the world. This study asks a simple question: Does the time of day change which books you pick, or how creatively you mix them together?

The researchers looked at millions of posts from Reddit across four countries (the US, India, the UK, and Canada) to see if our "semantic exploration"—a fancy way of saying how creatively we wander through ideas and words—follows a daily rhythm.

Here is the breakdown of their findings using everyday analogies:

1. The Morning "Wanderer" vs. The Afternoon "Clusterer"

The study found two distinct patterns that happen at different times of the day, like a two-act play:

Act 1: The Morning Wander (Local Exploration)
In the early morning (roughly 3:00 AM to 5:00 AM local time), people are like explorers setting out on a hike. They are picking up random, diverse items from the forest floor.
- What this means: Individual posts are very different from one another. People are talking about a wide variety of unique topics. They are "exploring" new semantic territory.
- The Metaphor: Imagine a morning market where every vendor is selling something completely different and unexpected.
Act 2: The Afternoon Huddle (Global Exploitation)
As the day goes on and more people wake up and start posting, the pattern shifts. The "exploration" slows down, and the "huddling" begins.
- What this means: While the total number of posts grows, people start talking about the same popular topics. New posts pile up around ideas that are already famous.
- The Metaphor: This is the "Rich-Get-Richer" effect. Imagine a popular street performer in a city square. At first, only a few people stop to watch. But as the day goes on, a massive crowd gathers around that one performer, and everyone else ignores the other street corners. The crowd gets bigger, but the variety of where they are looking shrinks.

2. It's Not About Mood

You might think, "Maybe people are just more creative when they are happy and less creative when they are sad." The researchers tested this, and the answer was no.

The Finding: The time of day when people are most creative (morning) doesn't match the time when they are most happy or sad.
The Metaphor: Think of your brain's "creativity engine" and your "mood engine" as two different cars. They run on different schedules. You can be in a bad mood but still have a very creative, wandering mind, or be in a good mood but stick to boring, repetitive topics. The study shows that the "creativity engine" has its own internal clock that isn't just driven by how you feel emotionally.

3. The Sun is the Conductor

The study suggests that our internal "creativity clock" is tuned to the sun, not just our social schedules.

The Finding: The timing of when people start their "morning wander" shifts slightly depending on the season and the length of the day (sunrise/sunset).
The Metaphor: Think of the sun as a conductor for an orchestra. Even if the musicians (people) are in different countries, the conductor (sunlight) signals them all to start playing their most creative solos at roughly the same biological time relative to the sunrise.

4. The Biological "Why"

Why does this happen? The paper suggests it's linked to dopamine, a chemical in our brain associated with motivation and the drive to explore new things.

The Theory: In humans (who are active during the day), our dopamine levels are likely higher in the morning, making us more eager to try new ideas and explore. As the day wears on, we settle into "exploitation" mode—sticking to what we know works.
The Metaphor: Your brain is like a squirrel in the morning. It's hyper-active, running in every direction to find new nuts (ideas). By afternoon, the squirrel is tired and just eats from the pile of nuts it already found.

Summary

This paper tells us that human language isn't just random noise. It follows a biological rhythm:

Morning: We are explorers, generating unique, diverse ideas.
Afternoon/Evening: We are collectors, gathering around popular, established topics.
The Driver: This rhythm is likely controlled by our internal body clock and sunlight, and it operates independently of whether we are feeling happy or sad.

It's a reminder that our brains have a natural "creative peak" in the morning, regardless of our mood, before we settle into the comfort of familiar topics later in the day.

Technical Summary: Circadian Modulation of Semantic Exploration in Social Media Language

Problem Statement
While human cognition is known to exhibit strong circadian modulation affecting attention, affect, and decision-making, the influence of these biological rhythms on high-dimensional semantic behavior remains poorly understood. Previous large-scale analyses of social media have established daily rhythms in emotional expression and word frequency (e.g., positive affect peaking in the morning), but it is unclear whether these fluctuations reflect changes in semantic exploration—the cognitive flexibility to traverse new ideas versus sticking to familiar expressions. Furthermore, existing research has not sufficiently distinguished between individual-level semantic variability and population-level semantic diversity, nor has it determined if semantic exploration is merely a proxy for mood or a distinct cognitive dimension.

Methodology
The authors analyzed a large-scale corpus of 2,921,362 Reddit submissions from January to December 2024, focusing on four countries with high submission volumes: the United States, India, the United Kingdom, and Canada. The study employed a multi-stage technical pipeline:

Data Preprocessing & Geolocation: Submissions were filtered to exclude NSFW content, bots, and non-English text. Geographic attribution was performed using a hierarchical strategy (URL top-level domains, subreddit names, author flair, and Named Entity Recognition) to assign posts to specific cities and time zones. Timestamps were converted to local civil time, accounting for Daylight Saving Time (DST).
Semantic Embedding: Text (title + selftext) was embedded using the pretrained transformer model sentence-transformers/all-mpnet-base-v2 to generate normalized high-dimensional vectors.
Entropy Metrics: Two complementary measures of semantic exploration were derived from Shannon's information entropy:
- Local Semantic Entropy ( $H_{local}$ ): Calculated for individual posts using a $k$ -nearest-neighbor ( $k=10$ ) distance in embedding space ( $H_{local}(x_i) = \log(r_k(x_i))$ ). This measures how distinct a specific post is from others in the same hourly bin, reflecting individual-level exploration.
- Global Semantic Entropy ( $H_{global}$ ): Calculated as the differential entropy of the distribution of all embeddings within a specific hour for a country, using the determinant of the covariance matrix ( $H_{global} = \frac{1}{2}\log((2\pi e)^d \det(\Sigma))$ ). This measures the overall breadth of the semantic space covered by the population.
Sentiment & Validation: Sentiment was quantified using the VADER analyzer. To validate the semantic entropy measure against established rhythms, the authors digitized standard deviation profiles of positive emotion and sadness word usage from a previous Twitter study (Dzogang et al., 2017).
Statistical Modeling: Circadian rhythmicity was assessed using single-component 24-hour cosinor models. Seasonal entrainment was analyzed by correlating entropy peaks/troughs with sunrise and sunset times. Cluster analysis (HDBSCAN) and power-law fitting were used to examine the temporal organization of semantic topics.

Key Results

Robust Circadian Rhythmicity: All four countries exhibited significant 24-hour rhythmicity in local semantic entropy ( $P_{fdr} < 0.001$ ). The peak of semantic exploration (acrophase) consistently occurred in the early morning hours (approx. 3:00–5:00 AM local time), with $R^2$ values between 0.73 and 0.78.
Seasonal Entrainment: In the US, the timing of the semantic entropy trough was positively correlated with sunset times ( $r=0.68$ ), suggesting that seasonal light cues (photoperiod) may entrain the phase of semantic activity.
Dissociation from Sentiment: The relationship between semantic exploration and sentiment was weak and inconsistent. While semantic entropy correlated with the variability of sadness-related word usage (from the reference study), it did not correlate with mean sentiment scores. In some countries, the correlation was negative; in others, positive. This indicates that semantic exploration captures a cognitive dimension distinct from affective valence.
Temporal Dissociation (Local vs. Global): A systematic temporal dissociation was observed:
- Local entropy (individual exploration) peaked in the early morning.
- Global entropy (population diversity) peaked later in the day (approx. 14:00–17:00), closely tracking total submission volume.
"Rich-Get-Richer" Dynamics: The marginal gain in global semantic entropy per additional post declined significantly as the day progressed. Early posts contributed substantially to semantic diversity, while later posts showed diminishing returns, accumulating within established clusters. This follows a power-law distribution ( $P(s) \sim s^{-1.60}$ ), indicating that the population increasingly engages with existing semantic hubs rather than exploring novel regions.

Key Contributions

Quantification of Semantic Exploration: The study introduces a method to quantify semantic exploration-exploitation trade-offs in naturalistic language using transformer embeddings and entropy measures, moving beyond simple word frequency.
Scale-Specific Circadian Patterns: It demonstrates that circadian rhythms operate differently at the individual level (early morning exploration) versus the population level (afternoon accumulation/exploitation), revealing a two-stage daily process.
Independence from Affect: The work provides evidence that semantic diversity is not merely a byproduct of mood fluctuations but represents a distinct cognitive process, potentially linked to neuromodulatory systems regulating cognitive flexibility.
Cross-National Consistency: The findings establish that these circadian semantic patterns are robust across diverse cultures and geographies, suggesting a general property of human cognitive dynamics.

Significance and Claims
The paper claims that biological circadian rhythms extend into the semantic domain, influencing not just what people say, but how diversely they express ideas. The observed early-morning peak in local semantic exploration aligns with known diurnal patterns in neuromodulatory systems, specifically dopaminergic tone, which is associated with behavioral flexibility and exploration in awake states.

The authors posit that the temporal structure of social media language reflects a fundamental cognitive mechanism: the day begins with a phase of broad semantic exploration (high local entropy), followed by a phase of consolidation where the community collectively expands the semantic space but increasingly concentrates on established topics (high global entropy, diminishing marginal returns). This pattern is consistent with "rich-get-richer" network dynamics and suggests that circadian biology shapes the fundamental architecture of human communication and cognitive flexibility in naturalistic settings.