Source-linked AI summary
#FoodPorn: Obesity Patterns in Culinary Interactions
Yelena Mejova, Hamed Haddadi, Anastasios Noulas, Ingmar Weber
TL;DR
Static health statistics do not capture everyday dining experiences, food perceptions, or their social setting. The paper analyzes Foursquare and Instagram restaurant data alongside county statistics, finding that fast-food prevalence is more clearly correlated with obesity in its data than in County Health Rankings. It also finds that social approval often favors unhealthy foods, even where users associate #foodporn with healthier cuisines.
Problem
County health statistics are static and do not capture everyday dining outings, people's food perceptions, or the social setting of meals.
Source: §1 · p. 1Method
The study analyzes Foursquare and Instagram restaurant data, including posts and their social interactions, alongside county health statistics.
Source: §1 · p. 1, §2.1 · p. 2, §2.2 · p. 2Results
Takeaways & Limitations
The findings suggest a complex dietary landscape in which fast food, social interactions, and personal perceptions may be at odds.
Source: §8 · p. 8, §7 · p. 8The social-media data are biased toward more visited and shared locations, and the fast-food definition may differ from County Health Rankings or users' labels.
Source: §3.1 · p. 3, §7 · p. 7
Abstract
from arXiv · showhide
We present a large-scale analysis of Instagram pictures taken at 164,753 restaurants by millions of users. Motivated by the obesity epidemic in the United States, our aim is three-fold: (i) to assess the relationship between fast food and chain restaurants and obesity, (ii) to better understand people's thoughts on and perceptions of their daily dining experiences, and (iii) to reveal the nature of social reinforcement and approval in the context of dietary health on social media. When we correlate the prominence of fast food restaurants in US counties with obesity, we find the Foursquare data to show a greater correlation at 0.424 than official survey data from the County Health Rankings would show. Our analysis further reveals a relationship between small businesses and local foods with better dietary health, with such restaurants getting more attention in areas of lower obesity. However, even in such areas, social approval favors the unhealthy foods high in sugar, with donut shops producing the most liked photos. Thus, the dietary landscape our study reveals is a complex ecosystem, with fast food playing a role alongside social interactions and personal perceptions, which often may be at odds.
1. INTRODUCTION
Static obesity and diabetes statistics do not capture everyday dining, people's perceptions of food, or the social setting of meals. This study uses Foursquare and Instagram data to examine restaurant prevalence, dining perceptions, and social approval, finding that online approval can favor unhealthy foods even where users associate #foodporn with healthier cuisines.
1. INTRODUCTION
- Static health statistics miss everyday dining outings, people's thoughts and feelings about food, and the social setting of meals.Prior studies linked neighborhood fast-food prevalence with higher odds of diabetes, obesity, and childhood obesity.Source: §1 · p. 1
- The study analyzes Foursquare and Instagram data to examine fast-food and chain restaurants, dining perceptions, and social approval across 164,753 US restaurants and more than 3 million people.The paper positions these platforms as sources of everyday interactions between people, society, and food [1].Source: §1 · p. 1
- Fast-food prevalence correlates more clearly with county obesity in these data than in the 2013 County Health Rankings, while low-obesity regions use #smallbiz and #eatlocal more often.Fast-food photos also receive twice as many tags associated with unhealthy topics; donut, cupcake, and burger photos often receive social approval.Source: §1 · pp. 1–2
- Social approval often favors unhealthy restaurants such as donut, cupcake, and burger places, despite users associating #foodporn predominantly with Asian cuisines.The paper frames these findings as a window into daily dining that could potentially inform intervention and policy decisions.Source: §1 · p. 2
2. DATA COLLECTION
The study combines Foursquare restaurant locations, Instagram posts, and county health and census data to examine dining and obesity across the United States. It describes how each dataset was assembled and classified, and checks how well Instagram activity tracks county population.
2.1 Foursquare locations
- Foursquare check-ins made public on Twitter from December 2010 through September 2011 yielded 194,752 unique US food places, which were mapped to counties; Figure 2 illustrates Manhattan check-in coverage.Foursquare filters fake check-ins and harmonizes venues to remove duplicates.Source: §2.1 · p. 2, Figure 2 · p. 2
2.2 Instagram pictures
- Mapping Foursquare places to Instagram location IDs yielded 164K unique Instagram locations; recent posts were sampled monthly from September through November 2014.The collection included 20,848,190 posts from 3,367,777 people in 316 counties; location tagging is voluntary, and duplicate images were removed. Figure 3 illustrates a post's hashtags, likes, and comment.Source: §2.2 · p. 2, Figure 3 · p. 2
2.3 County health & census data
- The study uses 2013 County Health Rankings data for county obesity and fast-food percentages, alongside health, demographic, and community measures.Obesity rates in the data range from 13% in Teton County, Wyoming, to 48% in Greene County, Alabama.Source: §2.3 · p. 3
2.4 Fast food restaurants
- Fast-food places are defined as restaurants in Foursquare's Fast Food or Burger categories or in compiled lists of prominent fast-food and fast-casual chains.The compiled lists include the top 50 chains by sales and Wikipedia lists.Source: §2.4 · p. 3
2.5 Population representativeness
- Spearman's ρ is 0.746 between county population and unique users, and 0.9954 between users and pictures, indicating that activity is proportional to population in the sample.The authors note that users may be young and tech-savvy; Figure 1 displays county populations, pictures, and unique users ordered by population.Source: §2.5 · p. 3, Figure 1 · p. 3
3. FAST FOOD
Foursquare’s county-level fast-food share correlates more strongly with obesity than County Health Rankings data, while chain counts show no clear correlation. Local restaurants attract more Instagram activity than chains, although the fast-food measure may reflect Foursquare users’ dining behavior.
3.1 Fast food and obesity
- 0.424 is the Pearson correlation between Foursquare fast-food share and obesity, versus 0.246 for County Health Rankings and 0.267 for population-weighted County Health Rankings.Figure 4 compares the two fast-food measures and county obesity.Source: §3.1 · p. 3, Figure 4 · p. 4
- The Foursquare data show a less prominent fast-food share, possibly because of differing definitions, user labeling, or less frequent check-ins at fast-food restaurants.The authors note that the data may capture the dining behavior of the population active on Foursquare.Source: §3.1 · p. 3
- 0.499 for County Health Rankings and 0.450 for Foursquare are the correlations between fast-food prevalence and the population of children under 18.Among the race, income, and poverty-related variables examined, this was the highest correlation.Source: §3.1 · p. 3
3.2 Local places versus chain restaurants
- Restaurants with names repeated at least 10 times were classified as chains, with matching name variants and locations containing those names included.Manual inspection suggested the threshold distinguished chains from frequently repeated local names while allowing restaurants with a handful of local branches.Source: §3.2 · pp. 3–4
- Counties showed no clear correlation between chain-restaurant counts and obesity, while local restaurants received more photos and users than chains.Restaurants averaged 396 photos and 107 distinct users at local restaurants, compared with 150 photos and 46 distinct users at chains; Figure 5 displays per-restaurant distributions on a log scale.Source: §3.2 · p. 4, Figure 5 · p. 4
4. FOOD PERCEPTION
The section examines what restaurant hashtags reveal about dining perceptions and how those tags vary across restaurant types and obesity-rate areas. Users’ tag patterns distinguish fast-food perceptions from social dining at chains, while health- and local-food tags are more prominent in low-obesity areas.
4. FOOD PERCEPTION
- Asian cuisines and molecular gastronomy dominate #foodporn shares, while picture-volume leaders differ; its correlation with picture count is -0.174.The table lists categories by #foodporn share alongside their overall picture counts.Source: §4 · p. 4, Table 1 · p. 5
- #foodporn is the most frequently used hashtag, but #vscocam is most liked among frequent hashtags.The camera-app hashtag leads in both average and median likes.Source: §4 · p. 4
4.1 Hashtag labeling
- The top 2,000 hashtags were crowd-labeled into four binary dimensions—healthy, unhealthy, social, and emotion—with 92–99% label overlap.Three different annotators labeled each tag.Source: §4.1 · p. 5
4.2 Perception of fast food
- Unhealthy-food tags appear twice as often for fast food as for slow food, while #foodporn and social-occasion tags are more associated with slow-food places.The authors caution that these comparisons depend on the vocabulary covered by the tagging method.Source: §4.2 · p. 5, Figure 6 · p. 5
- Chain restaurants have more social-event tags than local restaurants (0.158 vs. 0.137), suggesting chains also feature in social outings.The difference is significant at p<0.001.Source: §4.2 · p. 5
4.3 Chatter around obesity
- High-obesity areas show more sharing and dessert-related tags, whereas low-obesity areas show more healthy-food and local-food tags.The authors note the association between small local restaurants and healthier communities.Source: §4.3 · pp. 5–6, Table 2 · p. 6
5. SOCIAL APPROVAL
Social feedback varies with obesity region and hashtag use, while the most-liked restaurant categories are largely rich foods and sweets. Donuts and cupcakes are especially popular in low-obesity areas.
5. SOCIAL APPROVAL
- Pictures from high-obesity areas receive 56% fewer likes than those from low-obesity areas, and also tend to receive fewer comments.Source: §5 · p. 6
- Both healthy and unhealthy tags are associated with more likes, with the difference between them more pronounced in high-obesity areas.In low-obesity areas, healthy and unhealthy tags are associated with roughly similar likes; “social” tags also gain popularity relative to “none”.Source: §5 · p. 6, Figure 8 · p. 6
- Pictures tagged #foodporn average 38 likes, compared with 26 for pictures without the hashtag.Comments show a similar difference: 2.5 with #foodporn versus 1.7 without it.Source: §5 · p. 6
- Desserts lead likes overall, followed by mac & cheese, burgers, and French food; donuts average 70 likes per picture in low-obesity areas.The top-liked foods are described as high in sugar, fat, and carbohydrates.Source: §5 · p. 6
6. RELATED RESEARCH
Prior work used social-media language and location data to study health and eating patterns, while this study examines restaurants and food perceptions alongside obesity. Its findings connect fast-food prevalence and hashtag patterns with obesity regions and social interaction.
6. RELATED RESEARCH
- Earlier work used social-media language and check-ins to model health or connect cultural preferences to eating habits [13]; a city-level overweight classifier reached 80.39% accuracy [7].Source: §6 · pp. 6–7
- Desserts lead likes overall, followed by mac & cheese, burgers, and French food; donuts average 70 likes per picture in low-obesity areas.The top-liked foods are described as high in sugar, fat, and carbohydrates.Source: Table 3 · p. 7
- Unlike those efforts, this study focuses on fast-food and chain restaurants, other restaurant categories, and their relation to obesity and food perception.Source: §6 · p. 7
- Prior studies link fast-food availability with obesity, including near restaurants for children and in disadvantaged urban areas; this study's social-media data confirms the link.Source: §6 · p. 7
- Food sharing is associated with high-obesity regions, while mentions of local and small businesses are associated with low-obesity regions; likes on healthy and unhealthy tags echo contradictory pressures.Source: §6 · p. 7
7. DISCUSSION
Social-media data are biased toward visited and shared locations, and the available picture metadata is insufficient for county-wide obesity prediction. The discussion also connects local restaurants with lower-obesity areas and notes that social approval often favors unhealthy foods.
7. DISCUSSION
- Foursquare and Instagram data favor more visited and shared locations than County Health Rankings data, and differing fast-food definitions may help explain their stronger correlation.The authors also note that popularity bias may reflect a relationship between fast-food popularity and local obesity.Source: §7 · p. 7
- The regression model's R2 of 0.013 indicates that picture metadata alone was insufficient to predict county-wide obesity without deeper demographic knowledge.Source: §7 · p. 7
- Users are almost twice as likely to post at local restaurants as at chains, and #eatlocal and #smallbiz are associated with lower-obesity areas.Source: §7 · p. 7
- Small restaurants may be easier to engage as a community in promoting healthier menus, while some non-fast-food chains provide social settings for sharing food that may not be the healthiest choice.Source: §7 · p. 8
- Social approval emphasizes foods high in sugar and fat, and the authors call for more work on users' motivations to understand how interactions might benefit dietary health.Source: §7 · p. 8
8. CONCLUSION
Using millions of Instagram posts and Foursquare data, the study finds local-food and small-business patterns alongside social approval that often favors unhealthy choices. The authors propose comment sentiment analysis and image recognition as future directions.
8. CONCLUSION
- The study analyzes food consumption at scale using Instagram and Foursquare, with millions of posts from restaurants across the United States correlated with national statistics.Source: §8 · p. 8
- Small businesses and local foods are related to obesity patterns and receive more social-media attention, while likes and comments often favor donuts, cupcakes, and other sweets.Source: §8 · p. 8
- Future work will use comment sentiment analysis and image recognition to assess approval and infer picture contents and potential caloric values.The authors say these efforts may increase analysis confidence across locations and improve understanding of the food consumed.Source: §8 · p. 8