Yelena Mejova

You Tweet What You Eat: Studying Food Consumption Through Twitter

Food is an integral part of our lives, cultures, and well-being, and is of major interest to public health. The collection of daily nutritional data involves keeping detailed diaries or periodic surveys and is limited in scope and reach. Alternatively, social media is infamous for allowing its users to update the world on the minutiae of their daily lives, including their eating habits. In this work we examine the potential of Twitter to provide insight into US-wide dietary choices by linking the tweeted dining experiences of 210K users to their interests, demographics, and social networks. We validate our approach by relating the caloric values of the foods mentioned in the tweets to the state-wide obesity rates, achieving a Pearson correlation of 0.77 across the 50 US states and the District of Columbia. We then build a model to predict county-wide obesity and diabetes statistics based on a combination of demographic variables and food names mentioned on Twitter. Our results show significant improvement over previous CHI research (Culotta 2014). We further link this data to societal and economic factors, such as education and income, illustrating that areas with higher education levels tweet about food that is significantly less caloric. Finally, we address the somewhat controversial issue of the social nature of obesity (Christakis & Fowler 2007) by inducing two social networks using mentions and reciprocal following relationships.

GOP primary season on twitter: "popular" political sentiment in social media

As mainstream news media and political campaigns start to pay attention to the political discourse online, a systematic analysis of political speech in social media becomes more critical. What exactly do people say on these sites, and how useful is this data in estimating political popularity? In this study we examine Twitter discussions surrounding seven US Republican politicians who were running for the US Presidential nomination in 2011. We show this largely negative rhetoric to be laced with sarcasm and humor and dominated by a small portion of users. Furthermore, we show that using out-of-the-box classification tools results in a poor performance, and instead develop a highly optimized multi-stage approach designed for general-purpose political sentiment classification. Finally, we compare the change in sentiment detected in our dataset before and after 19 Republican debates, concluding that, at least in this case, the Twitter political chatter is not indicative of national political polls.

#FoodPorn: Obesity Patterns in Culinary Interactions

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 peoples 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.

Penguins in sweaters, or serendipitous entity search on user-generated content

In many cases, when browsing the Web users are searching for specific information or answers to concrete questions. Sometimes, though, users find unexpected, yet interesting and useful results, and are encouraged to explore further. What makes a result serendipitous? We propose to answer this question by exploring the potential of entities extracted from two sources of user-generated content -- Wikipedia, a user-curated online encyclopedia, and Yahoo! Answers, a more unconstrained question/answering forum -- in promoting serendipitous search. In this work, the content of each data source is represented as an entity network, which is further enriched with metadata about sentiment, writing quality, and topical category. We devise an algorithm based on lazy random walk with restart to retrieve entity recommendations from the networks. We show that our method provides novel results from both datasets, compared to standard web search engines. However, unlike previous research, we find that choosing highly emotional entities does not increase user interest for many categories of entities, suggesting a more complex relationship between topic matter and the desirable metadata attributes in serendipitous search.

Twitter: A Digital Socioscope

How can Twitter data be used to study individual-level human behavior and social interaction on a global scale? This book introduces readers to the methods, opportunities, and challenges of using Twitter data to analyze phenomena ranging from the number of people infected by the flu, to national elections, to tomorrows stock prices. Each chapter, written by leading domain experts in clear and accessible language, takes the reader to the forefront of the newly emerging field of computational social science. An introductory chapter on Twitter data analysis provides an overview of key tools and skills, and gives pointers on how to get started, while the case studies demonstrate shortcomings, limitations, and pitfalls of Twitter data as well as its advantages. The book will be an excellent resource for social science students and researchers wanting to explore the use of online data.

Reuse in the wild: an empirical and ethnographic study of organizational content reuse

We present a large-scale study of content reuse networks in a large and highly hierarchical organization. In our study, we combine analysis of a collection of presentations produced by employees with interviews conducted throughout the organization and a survey to study presentation content reuse. Study results show a variety of information needs and behaviors related to content reuse as well as a need for a personalized and socially-integrated networking tool for enabling easy access to previously generated presentation material. In this paper we describe our findings and outline a set of requirements for an effective content reuse facility.

A relevance-based topic model for news event tracking

Event tracking is the task of discovering temporal patterns of popular events from text streams. Existing approaches for event tracking have two limitations: scalability and inability to rule out non-relevant portions in text streams. In this study, we propose a novel approach to tackle these limitations. To demonstrate the approach, we track news events across a collection of weblogs spanning a two-month time period.

Giving is caring: understanding donation behavior through email

Every day, thousands of people make donations to humanitarian, political, environmental, and other causes, a large amount of which occur on the Internet. The solicitations for support, the acknowledgment of a donation and the discussion of corresponding issues are often conducted via email, leaving a record of these social phenomena. In this paper, we describe a comprehensive large-scale data-driven study of donation behavior. We analyze a two-month anonymized email log from several perspectives motivated by past studies on charitable giving: (i) demographics, (ii) user interest, (iii) external time-related factors and (iv) social network influence. We show that email captures the demographic peculiarities of different interest groups, for instance, predicting demographic distributions found in US 2012 Presidential Election exit polls. Furthermore, we find that people respond to major national events, as well as to solicitations with special promotions, and that social connections are the most important factor in predicting donation behavior. Specifically, we identify trends not only for individual charities and campaigns, but also for high-level categories such as political campaigns, medical illnesses, and humanitarian relief. Thus, we show the extent to which large-scale email datasets reveal human donation behavior, and explore the limitations of such analysis.

Political speech in social media streams: YouTube comments and Twitter posts

Recently, political sentiment on social media websites has been receiving much attention both in research circles and in the news. However, political sentiment analysis has been largely performed on only a single social media source. It is unclear what outcomes would result if more than one source were used. We present a unique comparison of the textual content of two popular social media -- Twitter posts and YouTube comments -- over a common set of queries which include politicians, issues, and events. We show Twitter as a stream driven by news and outside sources with 40% share of its content lacking any sentiment, and YouTube as an outlet for opinionated speech. Specifically in YouTube, we find that the authors political stance and the sentiment of the document do not always match, and should be treated separately in analysis of political documents. We also examine the connection between social media sentiment and that of general population by comparing our findings to the Gallup poll, and show that neither discussion volume or sentiment expressed in the two social media were able to predict the republican Presidential nominee frontrunner.

Using Facebook Ads Audiences for Global Lifestyle Disease Surveillance: Promises and Limitations

Every day, millions of users reveal their interests on Facebook, which are then monetized via targeted advertisement marketing campaigns. In this paper, we explore the use of demographically rich Facebook Ads audience estimates for tracking non-communicable diseases around the world. Across 47 countries, we compute the audiences of marker interests, and evaluate their potential in tracking health conditions associated with tobacco use, obesity, and diabetes, compared to the performance of placebo interests. Despite its huge potential, we find that, for modeling prevalence of health conditions across countries, differences in these interest audiences are only weakly indicative of the corresponding prevalence rates. Within the countries, however, our approach provides interesting insights on trends of health awareness across demographic groups. Finally, we provide a temporal error analysis to expose the potential pitfalls of using Facebooks Marketing API as a black box.