Using geotagged facial expressions to visualize and characterize different demographic groups’ emotion in theme parks

Xiaoqing Song, Haoze Wu, Wei Jiang, Junjun Zhi, Xinyu Xia, Yi Long, Qin SuScientific 2024 · DOI

TL;DR

Researchers leveraged AI-driven facial recognition on thousands of geotagged social media photos to map the emotional landscape of Shanghai Disney Resort. By analyzing valence and arousal across different age and gender groups, the study reveals that tourists are far from a monolith: teen girls crave adventure, adult men are driven by nostalgia, and older visitors are often the most price-sensitive.

What Problem Does This Paper Address?

Traditional tourism research often treats "the tourist" as a single entity or relies on self-reported surveys and interviews. However, surveys are prone to "memory distortion"—we often remember trips more fondly (or poorly) than we felt in the moment—and people may provide biased answers.

Furthermore, while we know theme parks evoke strong emotions, we lack granular data on how specific demographic groups (e.g., a teenage boy vs. an older woman) react to specific "scenic spots" within a park. This paper addresses this gap by using "social sensors"—publicly shared, geotagged photos—to capture real-time emotional expressions across diverse demographic segments.

Methods

The researchers developed a framework to move from raw social media data to "Emotion Distribution Graphs."

  1. Data Extraction: They collected over 227,000 geotagged microblogs from Sina Weibo (China’s equivalent to X/Twitter) located within Shanghai Disney.
  2. Facial Recognition: Using the FaceReader Cognitive Algorithm, they identified 148,132 faces, extracting gender, age, and emotional state. Crucially, they didn't just look at "happy" or "sad," but quantified Valence (pleasure vs. displeasure) and Arousal (excitement vs. calmness) based on the circumplex model of affect.
  3. Spatial Analysis:
    • Standard Deviational Ellipse (SDE): Used to calculate the central tendency and dispersion of emotions for each group.
  4. DBSCAN (Clustering): An algorithm used to identify "emotional hotspots" where scenic spots evoked similar reactions.
  5. Intensity Indices: They created new metrics, the Valence Intensity index (VIT) and Arousal Intensity index (AIT), to compare theme lands (e.g., Tomorrowland vs. Treasure Cove).

Datasets

The study utilized a custom dataset derived from Sina Weibo microblogs posted between January 2019 and December 2020. After rigorous cleaning to remove noise and non-tourist data, they retained 42,988 microblogs. From these, they successfully categorized 42,595 facial expressions into six demographic groups:

  • Teen Male (TM) & Teen Female (TF)
  • Adult Male (AM) & Adult Female (AF)
  • Older Male (OM) & Older Female (OF)

Key Insights and Ideas

  • The Adventure Gap: Teen females (TF) showed significantly higher pleasure and excitement in adventure-themed areas like Treasure Cove compared to teen males (TM). Text analysis suggested TFs found the rides "exciting," while TMs were less impressed.
  • Nostalgia is a Male Driver: Adult males (AM) showed a direct correlation between their emotional pleasure and the release date of the Disney IP. They were happiest at attractions based on older movies (like Mickey or Peter Pan) they watched as children, showing that nostalgia is a powerful emotional anchor for this group.
  • The "Little Girl Stuff" Critique: Older females (OF) exhibited clear negative valence in the Gardens of Imagination. Social media comments revealed they found the romantic, "princess-themed" aesthetics "naïve" or "full of little girl stuff," preferring instead to spend time in Disneytown for dining and rest.
  • Price Sensitivity: Adult and older males were the most likely to express negative emotions in Disneytown. The data suggests this is due to the "burden of earning"; they were more sensitive to the high prices of food and merchandise than teenagers using their parents' money.

Key Results

The study proved that emotional distributions are statistically different ($p < 0.01$) across all demographic groups.

  • Teen Males (TM) had the most scattered emotional responses (the largest SDE area), meaning their reactions were highly inconsistent across the park.
  • Adult Females (AF) had the most consistent emotional "clusters," suggesting the park's design aligns well with their expectations.
  • Optimized Routing: The researchers used these insights to generate "optimized travel routes" for each demographic, connecting the spots that evoked the highest positive valence and arousal for that specific group.

Limitations and Future Work

The authors acknowledge the "social media mask": people often over-exaggerate positive emotions for the camera or "pose" in ways that don't reflect their internal state. The study also focused solely on one cultural context (Shanghai Disney).

Future research aims to develop algorithms to detect "distorted" or "fake" facial expressions and to expand the study cross-nationally to see if these demographic emotional patterns hold true across different cultures and park types.

Summary updated

Abstract

Tourism is an emotional sphere, and researchers focus on emotions to optimize tourism experiences. Tourism studies on emotions mostly ignore differences in emotions across demographic tourist groups by gender and age, thus limiting the understanding of emotions to the explicit characteristics of tourists' emotions. On the basis of geotagged facial expressions on social media platforms, this study aims to visualize the emotions of groups in scenic spots and then reveal the variations between groups' emotions within theme parks. By employing a facial recognition algorithm, an emotion distribution graph was proposed to represent groups' emotions in detail. Some analytical methods were combined to characterize of the emotion distribution of each group. Through a comprehensive comparison, the results suggest that there are unique characteristics of emotion distribution for each group and considerable variations between them. This study helps researchers achieve a deeper understanding of tourists' emotional differences and enhances the theorization of emotions. This research also highlights the advantages and significant practical implications of our method framework.
BibTeX
@inproceedings{song2024using,
  author = {Xiaoqing Song and Haoze Wu and Wei Jiang and Junjun Zhi and Xinyu Xia and Yi Long and Qin Su},
  title = {{Using geotagged facial expressions to visualize and characterize different demographic groups’ emotion in theme parks}},
  booktitle = {Scientific},
  year = {2024},
  doi = {10.1038/s41598-024-69555-5},
}
Figure