Features

XX.1 January + February 2013
Page: 48
Digital Citation

You are what you tweet


Authors:
Natalie Dixon

Every day is a living torment. Nobody understands my suffering. The only thing keeping me going is the pancakes I have for breakfast. —Twitter user

Food is a deeply emotional, political, and commercial subject. In recent years its story has been increasingly told, scrutinized, and debated. Arguably we have never been more aware of what we eat and the effect it has on our bodies (and the planet). Yet, rarely enough are we given a chance as citizens, governments, or campaigners to reflect on our daily food practices in a meaningful way.

The dynamics of everyday life have been drastically altered since the advent of digital technology. Emotions previously expressed in face-to-face encounters, letters, and analog technology are now channeled to the digital world first. We rant on Facebook status updates; we celebrate victories on Twitter; we text and email our friends to share our good mood. We feel happy, sad, frustrated, or disgusted, and we want to share and often validate our feelings with others online. By combining the subjects of food and emotion (or in this case, sentiment) using social media data, we can help facilitate a unique reflection on food practices.

Born out of a healthy love for data and a curiosity about food cultures, FoodMood (www.foodmood.in) is a data-visualization project that captures the content and sentiment of global English-language tweets about food. Acting like a global food-sentiment barometer, FoodMood aims to better understand food-consumption patterns and their impact on the daily emotional well-being of people. This is contextualized against the backdrop of country data such as gross domestic product (GDP) and obesity levels.

The FoodMood user experience is meant to be a playful and explorative one. Users can search for foods, discover new ones, or browse their own country’s food tweets, sorting by emotion or by tweet quantity, and zoom down into the content of individual tweets. By engaging citizens with their own data about food, FoodMood comes at a highly relevant time for reflection and self-awareness. It provides an interface to engage and explore social data to lend important insights into food practices, not only for interested foodies but also for advocacy and campaign groups like Oxfam Grow, the Slow Food movement, and FairFood.

Measuring Global Food Sentiment, One Tweet at a Time

FoodMood’s architecture is depicted in Figure 1. The system continuously gathers live data about food from Twitter by querying the standard Twitter API with terms such as for dinner, for lunch, for breakfast, I ate, and I’m eating. The gathered tweets are analyzed to determine whether they contain food types, and if they do, they are processed further. For each relevant tweet, the FoodMood system consults the geolocation component to determine the location of the Twitter user. A combination of the tweet, the recognized food, the Twitter user location, the Twitter user identity, and the sentiment orientation is stored in the database.

In addition to live data from Twitter, FoodMood uses static data from the CIA World Factbook and the WHO for a country’s GDP per capita and obesity levels, respectively. A total of 138 countries are present in the database, with the annotated GDP per capita and obesity levels.

On request, database information from Twitter and the additional country information are combined, processed, and provided to the front end for visualization.

Visualization

In the visualization, color indicates range of emotion (yellow is most happy; blue is least happy), while sizes of the blocks indicate quantity of tweets (Figure 2). Icons, stick men and money bags, represent country obesity and GDP, respectively (Figure 3). These were adjusted for size to depict a range of quantities.

The treemap design of FoodMood empowers users to first gain an overview and broad awareness of a food landscape, then zoom in and filter for more granularity, and, finally, analyze the visual data with details on demand [1], including the following:

  • At a glance, users can see which foods are most tweeted about per country, as well as the foods for which tweeters express the highest sentiment.
  • Users can view “most obese countries” and “least GDP countries” and select an interval or a point in time on the timeline. This option can be used to search for specific days of interest. Will chicken have a low sentiment on Twitter during an avian flu outbreak? Will tuna’s sentiment slowly decline over time as campaign groups reach more people about its level of endangerment?

As with most data-visualization projects, it is practically impossible to ensure complete objectivity in the process. As visualization researcher Enrico Bertini states, “[E]ven if the author tries to be as neutral as possible, the data itself can offer only a partial view on the phenomenon” [2]. In the case of FoodMood, data was cherry-picked to serve our purpose of investigating global food engagement and sentiment patterns as discussed by Twitter users. So, we have also chosen to disregard certain other data [1]. The data pool consists of tweets by people who have access to the Internet and tweet in English. This undoubtedly creates a large bias in the visualization, so it should be kept in mind that the snapshot of data will obviously not be completely representative of the entire world, and especially so for the non-English-speaking countries.

Limiting our initial dataset to search only for tweets containing the words for breakfast, for lunch, for dinner, I ate, and I’m eating is inevitable considering the scope of this project. At this first iteration of the FoodMood application, readers should keep in mind we have essentially invented a system to detect food items that has not been scientifically tested against other terms that could potentially be used. However, the current approach works successfully, as the phrases chosen refer to eating as an individual action performed in near real time. Though adding more phrases could actually worsen the performance of FoodMood, it is something that forms a basis for experimentation in the future.

Ultimately, what the data tells us is a product of the data-visualization tool, a hybrid of the raw data we find, and the specific ways we look for, categorize, and process it.

The Data Story

RT @connnniiiee: a nice bagel and hot chocolate for breakfast the perfect start to a Saturday morning smiley.gif

Both countries with high levels of obesity and with healthy body mass index (BMI) averages showed an overwhelmingly high sentiment for fast foods, high-fat foods, and high-sugar foods. It is a significant implication that the most tweeted about foods in the world are mostly foods that contribute to obesity. Beyond discovering trends like these, this data visualization can be used to cultivate self-awareness among users and/or advocacy groups to support existing research into why high food engagement in our society is predominantly with “unhealthy” foods.

Echoing the findings of the WHO report on the global issue of rising obesity levels [3], 58 percent of the 50 most tweeted about foods globally contribute to obesity, as shown in Figure 4. These include fast-food brands tweeted by name (e.g., McDonald’s, Burger King, KFC, Taco Bell) and foods with a high glycemic index, fat, and sugar content (e.g., pie, brownies, chocolate, fried chicken).

The ubiquitous popularity of fast food stands out as a major trend in FoodMood. One or more of these four major multinationals is present in nearly all of the country views: McDonald’s, KFC, Burger King, and Chipotle. McDonald’s is absent only when there are no outlets (e.g., some African countries), indirectly revealing the global footprint of the fast-food business. This not only represents the success of these brands among consumers but also a powerful view on the vast globalization and flattening of food culture.

Overwhelmingly, the sentiment expressed by FoodMood’s collected tweets is positive, with some foods showing peaks at certain times of year, such as chocolate near Easter. The sentiment scale on the visualization thus reflects only a rating from “least happy” to “most happy.” In line with previous research findings, pleasant emotions such as satisfaction, enjoyment, and desire were reported more often than negative ones in response to eating and tasting food [4].

Meat in many countries in this visualization received an overwhelmingly high sentiment rating (70 percent and higher). It was also widely tweeted about. This points to the vast global consumption of meat and that people, on average, enjoy it. That people are widely and happily consuming meat has large implications for health and natural resources, among other concerns. (For every one kilogram of beef produced 16,000 kg of water are required [5].)

Instances of cultural specificity were revealed in FoodMood. In each of the country views, very unique, nation-specific foodstuffs are represented in the visualization [6]. However, there are some foods that enjoy near-ubiquitous popularity, such as pancakes, eggs, and pizza. These are also foods that often appear on a country’s top happiest food list. This may represent the universal enjoyment levels of these foods or simply be a symptom of food globalization. By the same token, the presence of certain ethnic foods in countries points to the influence of immigrant communities that reflect cultural diversity across the globe. For example, in Argentina, foodstuffs represented include sisig and nilaga, traditional Filipino dishes. Asian Latin Americans have a centuries-long history in the region, starting with Filipino immigrants in the 16th century. FoodMood often reflects a country’s political history through food, revealing traces of former colonizers. Turn to the country view of Vietnam to see mentions of croissants, a legacy of previous French rule.


Overwhelmingly, the sentiment expressed by FoodMood’s collected tweets is positive, with some foods showing peaks at certain times of year, such as chocolate near Easter.


As the tweet dataset grows, it will allow researchers the chance to answer questions about why some foods are more engaging than others. So too, other datasets can be applied to the visualization to reveal patterns related to healthcare expenditure per country, sustainability, fair trade food, or number of calories consumed. Presenting this information about food consumption in a visually engaging way can help distill the essential changes that could then impact our food-purchasing choices and improve health.

FoodMood uniquely reveals consumption patterns among Twitter users and creates the basis for further analysis. It urges users, through its playful and interactive format, to ask questions about everyday food practices. While it is true that Twitter does not necessarily represent the population of a country, and that FoodMood cannot collect data from everyone on Twitter, it is also true that a person has to be engaged (emotionally affected to the point of action) to share something about food on Twitter. FoodMood in fact measures, implicitly, not only cultural ties of food to countries, but food engagement too [1]. Measuring these engagements can help us better reflect on the largely unconscious and habitual way we are buying, eating, and enjoying food.

Acknowledgments

The author would like to acknowledge fellow FoodMood team members: Bruno Jakić, Roderik Lagerweij, Mark Mooij from AI Applied, and Ekaterma Yudin from Affect Lab.

References

1. Dixon, N., Jakić, B., Lagerweij, R., Mooij, M., and Yudin, E. FoodMood: Measuring global food sentiment one tweet at a time. Proc. of Sixth International AAAI Conference on Weblogs and Social Media (June 4, 2012). AAAI Publications, 2012.

2. Bertini, E. Data visualization and influence. 2011; http://fellinlovewithdata.com/reflections/data-visualization-influence-1.

3. World Health Organisation Europe Fact and Figures; http://www.euro.who.int/en/what-we-do/health-topics/disease-prevention/nutrition/facts-and-figures.

4. Desmet, P. and Schifferstein, H. Sources of positive and negative emotions in food experience. Appetite 50, 2-3 (2008), 290–301.

5. Hoekstra, A.Y., ed. Virtual water trade. Proceedings of the International Expert Meeting on Virtual Water Trade. Value of Water Research Report Series 12. Delft, The Netherlands, 2003.

6. A few examples include: Vegemite (a popular Australian spread), biltong (a cured meat specialty unique to South Africa), biryani and chapati (traditional Indian rice dish and side bread), bun cha (a Vietnamese grilled pork noodle soup), and lechon (a traditional Argentinean dish of roasted suckling pig).

Author

Natalie Dixon is the director of Affect Lab (affectlab.org), a not-for-profit research lab based in Amsterdam that measures, investigates, and builds experiments with expressions of emotion. The lab’s projects aim to playfully facilitate awareness, debate, and reflection on people’s emotions to highlight their impact on behavior and everyday life. Dixon’s Ph.D. research takes place at the Department of Media and Communications at Goldsmiths College, London.

Figures

F1Figure 1. Graphic depiction of the data-collection process of FoodMood.

F2Figure 2. FoodMood visualization depicting the Netherlands country view (modified slightly for print).

F3Figure 3. FoodMood legend depicting GDP levels (money bags), average sentiment (color of stick-man head), and body size for obesity (modified slightly for print).

F4Figure 4. Foods with most tweets (global).

©2013 ACM  1072-5220/13/01  $15.00

Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee.

The Digital Library is published by the Association for Computing Machinery. Copyright © 2013 ACM, Inc.

Post Comment


No Comments Found