Scott Counts’s research while affiliated with Microsoft and other places

As users navigate online social spaces, they encounter numerous personal profiles, each displaying a unique constellation of attributes. How do users make sense of this information? In our first study, we provide evidence that users spontaneously make personality trait inferences about people from profiles they encounter online, and for certain profiles, preferentially remember this inferred trait content over actual profile content. Study 2 uses several measures of profile coherence to assess how the coherence of user profiles interacts with trait inferences to influence memory for profiles. Findings provide a better understanding of specific profile content that makes profiles memorable and the social-cognitive process utilized when extracting information from profiles.

People form consistent impressions of others given surprisingly little information. With the advent of social networks, impressions now may form online rather than in a face-to-face context. This research explores aspects of online impression formation and discusses the crucial role of user profiles in this process. By examining users' decisions in an experimentally controlled social network, we show that users need only a thin slice of profile information in order to form impressions of others online. Additionally, specific profile attributes are evaluated for their perceived utility (how much do users choose to view these attributes), predictiveness (how well they serve as a proxy for a full profile), and diagnosticity (their ability to help users choose between online profiles). Findings provide design suggestions for better profile displays when space is restricted.

The anti-vaccination movement threatens public health by reducing the likelihood of disease eradication. With social media’s purported role in disseminating anti-vaccine information, it is imperative to understand the drivers of attitudes among participants involved in the vaccination debate on a communication channel critical to the movement: Twitter. Using four years of longitudinal data capturing vaccine discussions on Twitter, we identify users who persistently hold pro and anti attitudes, and those who newly adopt anti attitudes towards vaccination. After gathering each user’s entire Twitter timeline, totaling to over 3 million tweets, we explore differences in the individual narratives across the user cohorts. We find that those with long-term anti-vaccination attitudes manifest conspiratorial thinking, mistrust in government, and are resolute and in-group focused in language. New adoptees appear to be predisposed to form anti-vaccination attitudes via similar government distrust and general paranoia, but are more social and less certain than their long-term counterparts. We discuss how this apparent predisposition can interact with social media-fueled events to bring newcomers into the anti-vaccination movement. Given the strong base of conspiratorial thinking underlying anti-vaccination attitudes, we conclude by highlighting the need for alternatives to traditional methods of using authoritative sources such as the government when correcting misleading vaccination claims.

Microblogging environments such as Twitter present a modality for interacting with information characterized by exposure to information “streams”. In this work, we examine what information in that stream is attended to, and how that attention corresponds to other aspects of microblog consumption and participation. To do this, we measured eye gaze, memory for content, interest ratings, and intended behavior of active Twitter users as they read their tweet steams. Our analyses focus on three sets of alignments: first, whether attention corresponds to other measures of user cognition such as memory (e.g., do people even remember what they attend to?); second, whether attention corresponds to behavior (e.g., are users likely to retweet content that is given the most attention); and third, whether attention corresponds to other attributes of the content and its presentation (e.g., do links attract attention?). We show a positive but imperfect alignment between user attention and other measures of user cognition like memory and interest, and between attention and behaviors like retweeting. To the third alignment, we show that the relationship between attention and attributes of tweets, such as whether it contains a link or is from a friend versus an organization, are complicated and in some cases counterintuitive. We discuss findings in relation to large scale phenomena like information diffusion and also suggest design directions to help maximize user attention in microblog environments

Major depression constitutes a serious challenge in personal and public health. Tens of millions of people each year suffer from depression and only a fraction receives adequate treatment. We explore the potential to use social media to detect and diagnose major depressive disorder in individuals. We first employ crowdsourcing to compile a set of Twitter users who report being diagnosed with clinical depression, based on a standard psychometric instrument. Through their social media postings over a year preceding the onset of depression, we measure behavioral attributes relating to social engagement, emotion, language and linguistic styles, ego network, and mentions of antidepressant medications. We leverage these behavioral cues, to build a statistical classifier that provides estimates of the risk of depression, before the reported onset. We find that social media contains useful signals for characterizing the onset of depression in individuals, as measured through decrease in social activity, raised negative affect, highly clustered egonetworks, heightened relational and medicinal concerns, and greater expression of religious involvement. We believe our findings and methods may be useful in developing tools for identifying the onset of major depression, for use by healthcare agencies; or on behalf of individuals, enabling those suffering from depression to be more proactive about their mental health.

Social media and networking websites, such as Twitter and Facebook, generate large quantities of information and have become mechanisms for real-time content dissipation to users. An important question that arises is: how do we sample such social media information spaces in order to deliver relevant content on a topic to end users? Notice that these large-scale information spaces are inherently diverse, featuring a wide array of attributes such as location, recency, degree of diffusion effects in the network and so on. Naturally, for the end user, different levels of diversity in social media content can significantly impact the information consumption experience: low diversity can provide focused content that may be simpler to understand, while high diversity can increase breadth in the exposure to multiple opinions and perspectives. Hence to address our research question, we turn to diversity as a core concept in our proposed sampling methodology. Here we are motivated by ideas in the "compressive sensing" literature and utilize the notion of sparsity in social media information to represent such large spaces via a small number of basis components. Thereafter we use a greedy iterative clustering technique on this transformed space to construct samples matching a desired level of diversity. Based on Twitter Firehose data, we demonstrate quantitatively that our method is robust, and performs better than other baseline techniques over a variety of trending topics. In a user study, we further show that users find samples generated by our method to be more interesting and subjectively engaging compared to techniques inspired by state-of-the-art systems, with improvements in the range of 15--45%.

Bias can be defined as selective favoritism exhibited by human beings when posed with a task of decision making across multiple options. Online communities present plenty of decision making opportunities to their users. Users exhibit biases in their attachments, voting and ratings and other tasks of decision making. We study bias amongst microblog users due to the value of an author's name. We describe the relationship between name value bias and number of followers, and cluster authors and readers based on patterns of bias they receive and exhibit, respectively. For authors we show that content from known names (e.g., @CNN) is rated artificially high, while content from unknown names is rated artificially low. For readers, our results indicate that there are two types: slightly biased, heavily biased. A subsequent analysis of Twitter author names revealed attributes of names that underlie this bias, including effects for gender, type of name (individual versus organization), and degree of topical relevance. We discuss how our work can be instructive to content distributors and search engines in leveraging and presenting microblog content.

Estimation of disease prevalence at sub-city neighborhood scale allows early and targeted interventions that can help save lives and reduce public health burdens. However, the cost-prohibitive nature of highly localized data collection and sparsity of representative signals, has made it challenging to identify neighborhood scale prevalence of disease. To overcome this challenge, we utilize alternative data sources, which are both less sparse and representative of localized disease prevalence: using query data from a large commercial search engine, we identify the prevalence of respiratory illness in the United States, localized to census tract geographic granularity. Focusing on asthma and Chronic Obstructive Pulmonary Disease (COPD), we construct a set of features based on searches for symptoms, medications, and disease-related information, and use these to identify illness rates in more than 23 thousand tracts in 500 cities across the United States. Out of sample model estimates from search data alone correlate with ground truth illness rate estimates from the CDC at 0.69 to 0.76, with simple additions to these models raising those correlations to as high as 0.84. We then show that in practice search query data can be added to other relevant data such as census or land cover data to boost results, with models that incorporate all data sources correlating with ground truth data at 0.91 for asthma and 0.88 for COPD.