The spread of fake news on social media became a public concern in the United States after the 2016 presidential election. We examined exposure to and sharing of fake news by registered voters on Twitter and found that engagement with fake news sources was extremely concentrated. Only 1% of individuals accounted for 80% of fake news source exposures, and 0.1% accounted for nearly 80% of fake news sources shared. Individuals most likely to engage with fake news sources were conservative leaning, older, and highly engaged with political news. A cluster of fake news sources shared overlapping audiences on the extreme right, but for people across the political spectrum, most political news exposure still came from mainstream media outlets.

There is a growing consensus that online platforms have a systematic influence on the democratic process. However, research beyond social media is limited. In this paper, we report the results of a mixed-methods algorithm audit of partisan audience bias and personalization within Google Search. Following Donald Trump's inauguration, we recruited 187 participants to complete a survey and install a browser extension that enabled us to collect Search Engine Results Pages (SERPs) from their computers. To quantify partisan audience bias, we developed a domain-level score by leveraging the sharing propensities of registered voters on a large Twitter panel. We found little evidence for the "filter bubble'' hypothesis. Instead, we found that results positioned toward the bottom of Google SERPs were more left-leaning than results positioned toward the top, and that the direction and magnitude of overall lean varied by search query, component type (e.g. "answer boxes"), and other factors. Utilizing rank-weighted metrics that we adapted from prior work, we also found that Google's rankings shifted the average lean of SERPs to the right of their unweighted average.

Social media messages, such as tweets, are frequently used by people during natural disasters to share real‐time information and to report incidents. Within these messages, geographic locations are often described. Accurate recognition and geolocation of these locations are critical for reaching those in need. This article focuses on the first part of this process, namely recognizing locations from social media messages. While general named entity recognition tools are often used to recognize locations, their performance is limited due to the various language irregularities associated with social media text, such as informal sentence structures, inconsistent letter cases, name abbreviations, and misspellings. We present NeuroTPR, which is a Neuro‐net ToPonym Recognition model designed specifically with these linguistic irregularities in mind. Our approach extends a general bidirectional recurrent neural network model with a number of features designed to address the task of location recognition in social media messages. We also propose an automatic workflow for generating annotated data sets from Wikipedia articles for training toponym recognition models. We demonstrate NeuroTPR by applying it to three test data sets, including a Twitter data set from Hurricane Harvey, and comparing its performance with those of six baseline models.

BackgroundTraining is a critical part of health information technology implementations, but little emphasis is placed on post-implementation training to support day-to-day activities. The goal of this study was to evaluate the impact of post-implementation training on key electronic health record activities.MethodsBased on feedback from providers and requests for technical support, we developed two classes designed to improve providers’ effectiveness with the electronic health record. Training took place at Kaiser Permanente, Mid-Atlantic States. The classes focused on managing patient-level information using problem lists and medication lists, as well as efficient documentation and chart review. Both classes used the blended learning method, integrating concrete scenarios, hands-on exercises and take-home materials to reinforce class concepts. To evaluate training effectiveness, we used a case–control study with a 1:4 match on pre-training performance. We measured the usage rate of two key electronic health record functions (problem list and medication list management) for six months before and after training. Change scores were compared using the Wilcoxon sign rank test.Results36 participants and 144 non-participants were included in the training evaluation. Training participants were more likely to manage both medication lists and problem lists after training. Class material is now being incorporated into an enterprise-wide multi-modal training program available to all providers at Kaiser Permanente in the Mid-Atlantic States.ConclusionsOngoing information technology training is well-received by healthcare providers, who expressed a clear preference for additional training. Training improved use of two important electronic health record features that are included as part of the Meaningful Use criteria.

The Islamic State of Iraq and ash-Sham (ISIS) continues to use social media as an essential element of its campaign to motivate support. On Twitter, ISIS’ unique ability to leverage unaffiliated sympathizers that simply retweet propaganda has been identified as a primary mechanism in their success in motivating both recruitment and “lone wolf” attacks. The present work explores a large community of Twitter users whose activity supports ISIS propaganda diffusion in varying degrees. Within this ISIS supporting community, we observe a diverse range of actor types, including fighters, propagandists, recruiters, religious scholars, and unaffiliated sympathizers. The interaction between these users offers unique insight into the people and narratives critical to ISIS’ sustainment. In their entirety, we refer to this diverse set of users as an online extremist community or OEC. We present Iterative Vertex Clustering and Classification (IVCC), a scalable analytic approach for OEC detection in annotated heterogeneous networks, and provide an illustrative case study of an online community of over 22,000 Twitter users whose online behavior directly advocates support for ISIS or contibutes to the group’s propaganda dissemination through retweets.

No abstract

Written by Michelle A. Amazeen, Fabrício Benevenuto, Nadia M. Brashier, Robert M. Bond, Lia C. Bozarth, Ceren Budak, Ullrich K. H. Ecker, Lisa K. Fazio, Emilio Ferrara, Andrew J. Flanagin, Ales-sandro Flammini, Deen Freelon, Nir Grinberg, Ralph Hertwig, Kathleen Hall Jamieson, Kenneth Jo-seph, Jason J. Jones, R. Kelly Garrett, Daniel Kreiss, Shannon McGregor, Jasmine McNealy, Drew Margolin, Alice Marwick, FiIippo Menczer, Miriam J. Metzger, Seungahn Nah, Stephan Lewan-dowsky, Philipp Lorenz-Spreen, Pablo Ortellado, Irene Pasquetto, Gordon Pennycook, Ethan Porter, David G. Rand, Ronald Robertson, Briony Swire-Thompson, Francesca Tripodi, Soroush Vosoughi, Chris Vargo, Onur Varol, Brian E. Weeks, John Wihbey, Thomas J. Wood, & Kai-Cheng Yang

Manual annotations are a prerequisite for many applications of machine learning. However, weaknesses in the annotation process itself are easy to overlook. In particular, scholars often choose what information to give to annotators without examining these decisions empirically. For subjective tasks such as sentiment analysis, sarcasm, and stance detection, such choices can impact results. Here, for the task of political stance detection on Twitter, we show that providing too little context can result in noisy and uncertain annotations, whereas providing too strong a context may cause it to outweigh other signals. To characterize and reduce these biases, we develop ConStance, a general model for reasoning about annotations across information conditions. Given conflicting labels produced by multiple annotators seeing the same instances with different contexts, ConStance simultaneously estimates gold standard labels and also learns a classifier for new instances. We show that the classifier learned by ConStance outperforms a variety of baselines at predicting political stance, while the model's interpretable parameters shed light on the effects of each context.