Google Play icon

You Tweet What You Eat: Studying Food Consumption Through Twitter

Share
Posted December 16, 2014

It’s probably not so much of a surprise: most of people already publish a direct pieces of evidence of everything (or almost everything) they do on various social networks. Despite inevitable and often disastrous privacy concerns which we subject ourselves by behaving this way there are also good things that may be accomplished by using bits of info from our daily lives.

Image source: mieranadhirah.com

Image source: mieranadhirah.com

Studying our habits is one potential example. Putting psychological studies aside, we could easily track behavior patterns which directly influence our personal and public health as well. Exactly this was the idea behind a new study published on arXiv.org this week. In this paper, a team of scientists from Qatar Computing Research Institute described their attempt to analyze food consumption peculiarities using information available from Twitter posts.

“Until now, large-scale dietary studies of food consumption used questionnaires and food diaries to keep track of the daily activities of their participants, which can be intrusive and expensive to conduct. Alternatively, social media is notorious for providing its users with means of documenting the minutiae of their daily lives, including their dietary choices”, write the authors of the study. And this idea is logical indeed: why to waste money on instruments to investigate food consumption characteristics that are hard (and often expensive) to implement in practice? Why not to use something that is readily available, like social networks?

Daily tweet volume mentioning food-related activity normalized by overall tweet activity, by weekday. Image courtesy of the researchers.

Daily tweet volume mentioning food-related activity normalized by overall tweet activity, by weekday. Image courtesy of the researchers.

One could even say that food and social media is a ‘perfect’ combination. Social media has become a part of our daily lives and many people use it on a intra-daily basis. Food is also a very significant part of our existence. And significant not only in a way that it is indispensable for us to live; it also directly influences our quality of life by affecting our health, longevity, social activity, expression of our culture and beliefs, and our well-being in general. Many illnesses like heart disease, diabetes, osteoarthritis and even cancer are linked to poor dietary choices, the scientists say.

Social media – and Twitter in particular – can be used to get insights into dietary habits of an entire country, the team explains in their paper. But information from twitter alone is not enough. In this research, they performed a large-scale analysis of approximately 502 million tweets generated by 210 thousand Twitter users in the United States. This data was supplemented by a variety of additional sources which allowed the scientists to consider significant details, e.g. the nutritional value of foods mentioned in those tweets, demographic characteristics of people behind those tweets,  their interests and some other scientifically important factors.

Distinguishing foods between rural and urban users. Image courtesy of the researchers.

Distinguishing foods between rural and urban users. Image courtesy of the researchers.

This study is also interesting from another point of view: it clearly shows us how much information can be inferred from our data available online (even if it isn’t very exact). For example, in order to assess the sensitivity of data related to more specific personal dietary variations, the scientists used the demographic information from the 2010 US Census. Based on this data, they extrapolated the gender information of Twitter users using their screen names only (!). User-provided ZIP codes were used to classify particular persons as being either urban or rural.

Caloric value of foods mentioned in tweets versus obesity rates. Image courtesy of the researchers.

Caloric value of foods mentioned in tweets versus obesity rates. Image courtesy of the researchers.

Even though gender-related information obviously lacks precision, the team managed to determine some dietetic associations that matched results obtained from more ‘traditional’ studies. And the variety of those associations is, well, surprising: women generally tweet about less caloric foods than men; higher estimated level of education relates to fewer consumed calories; urban-based users tweet about alcoholic beverages more often, while pizza and chocolate are popular in rural areas. When considering personal interests disclosed by users on their profiles, those who mentioned an interest in cooking tweeted 7.5 calories fewer than average. This is “in line with research showing that lack of cooking at homes gives rise to obesity”, the authors of the study say.

Average predicted percent obesity and average caloric value of tweets of four quartiles divided by education (Bachelor degree attainment, and income level), with 95% confidence intervals. Image courtesy of the researchers.

Average predicted percent obesity and average caloric value of tweets of four quartiles divided by education (Bachelor degree attainment, and income level), with 95% confidence intervals. Image courtesy of the researchers.

The team admits that social-media based sociological studies (as well as the current one) suffer from so called user sampling bias, since social media users generally represent affluent and tech-savvy demographic groups. They also often come from neighborhoods with average household income well above the US average (85,117 vs 51,017 in 2012). There are some other important drawbacks; however, a more fine-grained analysis could improve similar studies, the scientists argue.

A possibility to employ big data analysis is the largest advantage of the food consumption studies through Twitter, the authors say. Keyword mining, detection of words pertaining to self-image expressions and other high-precision techniques can be successfully employed to increase the overall accuracy of such studies, while maintaining a high level of automation.

“More needs to be done to develop sensitive and accurate tools for user characterization, with both textual and social network information available. As a documentation of users’ interests, opinions, and behaviors, this study is another example of the potential Twitter has for public health research”, the scientists conclude.

Written by Alius Noreika

Share

Featured news from related categories:

Technology Org App
Google Play icon
81,745 science & technology articles