Search for: All records

Abstract Wearing masks reduces the spread of COVID-19, but compliance with mask mandates varies across individuals, time, and space. Accurate and continuous measures of mask wearing, as well as other health-related behaviors, are important for public health policies. This article presents a novel approach to estimate mask wearing using geotagged Twitter image data from March through September, 2020 in the United States. We validate our measure using public opinion survey data and extend the analysis to investigate county-level differences in mask wearing. We find a strong association between mask mandates and mask wearing—an average increase of 20%. Moreover, this association is greatest in Republican-leaning counties. The findings have important implications for understanding how governmental policies shape and monitor citizen responses to public health crises. 

Government censorship—internet shutdowns, blockages, firewalls—impose significant barriers to the transnational flow of information despite the connective power of digital technologies. In this paper, we examine whether and how information flows across borders despite government censorship. We develop a semi-automated system that combines deep learning and human annotation to find co-occurring content across different social media platforms and languages. We use this system to detect co-occurring content between Twitter and Sina Weibo as Covid-19 spread globally, and we conduct in-depth investigations of co-occurring content to identify those that constitute an inflow of information from the global information ecosystem into China. We find that approximately one-fourth of content with relevance for China that gains widespread public attention on Twitter makes its way to Weibo. Unsurprisingly, Chinese state-controlled media and commercialized domestic media play a dominant role in facilitating these inflows of information. However, we find that Weibo users without traditional media or government affiliations are also an important mechanism for transmitting information into China. These results imply that while censorship combined with media control provide substantial leeway for the government to set the agenda, social media provides opportunities for non-institutional actors to influence the information environment. Methodologically, the system we develop offers a new approach for the quantitative analysis of cross-platform and cross-lingual communication.

 

This paper introduces the Multimodal Chile & Venezuela Protest Event Dataset (MMCHIVED). MMCHIVED contains city-day event data using a new source of data, text and images shared on social media. These data enables the improved measurement of theoretically important variables such as protest size, protester and state violence, protester demographics, and emotions. In Venezuela, MMCHIVED records many more protests than existing datasets. In Chile, it records slightly more events than the Armed Conflict Location and Events Dataset (ACLED). These extra events are from small cities far from Caracas and Santiago, an improvement of coverage over datasets that rely on newspapers, and the paper confirms they are true positives. While MMCHIVED covers protest events in Chile and Venezuela, the approach used in the paper is generalizable and could generate protest event data in 107 countries containing 97.14% of global GDP and 82.7% of the world's population. 

Existing pruning techniques preserve deep neural networks’ overall ability to make correct predictions but could also amplify hidden biases during the compression process. We propose a novel pruning method, Fairness-aware GRAdient Pruning mEthod (FairGRAPE), that minimizes the disproportionate impacts of pruning on different sub-groups. Our method calculates the per-group importance of each model weight and selects a subset of weights that maintain the relative between-group total importance in pruning. The proposed method then prunes network edges with small importance values and repeats the procedure by updating importance values. We demonstrate the effectiveness of our method on four different datasets, FairFace, UTKFace, CelebA, and ImageNet, for the tasks of face attribute classification where our method reduces the disparity in performance degradation by up to 90% compared to the state-of-the-art pruning algorithms. Our method is substantially more effective in a setting with a high pruning rate (99%). The code and dataset used in the experiments are available at https://github.com/Bernardo1998/FairGRAPE 

Understanding who blames or supports whom in news text is a critical research question in computational social science. Traditional methods and datasets for sentiment analysis are, however, not suitable for the domain of political text as they do not consider the direction of sentiments expressed between entities. In this paper, we propose a novel NLP task of identifying directed sentiment relationship between political entities from a given news document, which we call directed sentiment extraction. From a million-scale news corpus, we construct a dataset of news sentences where sentiment relations of political entities are manually annotated. We present a simple but effective approach for utilizing a pretrained transformer, which infers the target class by predicting multiple question-answering tasks and combining the outcomes. We demonstrate the utility of our proposed method for social science research questions by analyzing positive and negative opinions between political entities in two major events: 2016 U.S. presidential election and COVID-19. The newly proposed problem, data, and method will facilitate future studies on interdisciplinary NLP methods and applications. 

Existing public face image datasets are strongly biased toward Caucasian faces, and other races (e.g., Latino) are significantly underrepresented. The models trained from such datasets suffer from inconsistent classification accuracy, which limits the applicability of face analytic systems to non-White race groups. To mitigate the race bias problem in these datasets, we constructed a novel face image dataset containing 108,501 images which is balanced on race. We define 7 race groups: White, Black, Indian, East Asian, Southeast Asian, Middle Eastern, and Latino. Images were collected from the YFCC-100M Flickr dataset and labeled with race, gender, and age groups. Evaluations were performed on existing face attribute datasets as well as novel image datasets to measure the generalization performance. We find that the model trained from our dataset is substantially more accurate on novel datasets and the accuracy is consistent across race and gender groups. We also compare several commercial computer vision APIs and report their balanced accuracy across gender, race, and age groups. Our code, data, and models are available at https://github.com/joojs/fairface.