I'm an Assistant Professor of Sociology at Rutgers University—New Brunswick. I specialize in using computational methods and data from social media to analyze online political discussions, far-right activism, and hate speech. My work has been published in venues including Social Forces, Mobilization, and Socius. My website includes a short overview of my main areas of teaching and research and links to my CV, social media, and Github profile. If you have any questions, please get in touch using the contact details at the bottom of the page.
My research addresses several related questions. Why have far-right, populist actors have been so successful at building remarkably large online audiences? To what extent is online activism driven by offline events like protests, elections, and terrorist attacks versus endogenous processes like ranking and recommendation algorithms? What role do individual opinion leaders play in disseminating information and influencing public opinion in massive online debates? To address these questions, I bring together theories from political sociology, social movements, and public opinion scholarship and novel computational methods to examine data collected from social media, newspapers, and other sources.
My work includes case studies focused on the United Kingdom and comparative studies of political parties across Europe. In a forthcoming article in Mobilization, I argue that the popularity of far-right extremists on social media is attributable to processes endogenous to social media, considering how online engagement and algorithmic feedback loops enable actors to generate and sustain attention from online audiences. In a working paper with Jenny Enos using over a decade of data from Facebook and Twitter in twenty-eight countries, we show that populists have attracted more engagement on Facebook than other parties and that their online advantages appear to be growing. In other work, I compare online debates during the Brexit referendum and 2016 US Presidential election.
A second area of my research focuses on identifying and understanding hate speech on social media. My initial work on the subject focused on the distinction between hate speech and other forms of offensive language, demonstrating how the conflation of the two often resulted in false positives in machine learning classifiers here. This research was covered in Wired Magazine, Tech Republic, and New Scientist. I have subsequently developed theoretical work on to better understand different dimensions of hateful and abusive language (paper) and examined racial bias in hate speech and abusive language detection systems, demonstrating how classifiers designed to detect hate speech tend to predict that tweets written in African-American English are hateful than similar tweets written in Standard American English. You can read the paper, along with coverage in Vox.
I'm currently developing experimental research to understand how social context influences judgments about whether certain content is hateful or abusive and to ascertain the scope of biases in hate speech detection systems, supported by a Foundational Integrity Research award from Meta. A book chapter on the sociology of hate speech detection will appear in the Oxford Handbook on the Sociology of Machine Learning in 2023. I contributed to a recent paper combining large language models and experiments to understand how social contexts inform perceptions of toxicity and offensiveness, to appear in the 2023 Proceedings of the Association for Computational Linguistics. I am interested in informing policy debates on online hate speech and content moderation and have spoken about my research at a policy dialogue organized by the Organization of American States and a working group on content moderation at the European Commission.
In addition to my substantive interests, I also study how computational methods can be applied more generally in sociological research. For example, as part of the Fragile Families Challenge, I examined how neural networks can enable us to predict social outcomes and assessed the extent to which these black box predictive models can be amenable to sociological explanations. I found that these methods do not radically outperform traditional approaches like linear regression but may allow us to use large amounts of data to inductively identify important variables. A paper based on my analysis is published Socius and our paper describing the results of the Fragile Families Challenge mass collaboration was published in PNAS.
I also have experience using computational methods in industry settings. In the summer of 2016 I was an Eric and Wendy Schmidt Data Science for Social Good Fellow at the University of Chicago. I worked on a project to develop an early-warning system to identify police misconduct and helped develop a new model to predict risks at the dispatch level. You can read about our work here, along with media coverage in The Chicago Tribune, NPR, Mother Jones, the Economist, and Forbes. As of spring 2017, our system is being implemented in two police departments, you can read about the progress here. In 2017 I spent the summer in the Data Science Research & Development group at Civis Analytics, a data science consulting and software company based in Chicago. I used natural language processing and machine learning techniques to build a tool to monitor political discussions on Twitter. In 2018 I was a Core Data Science intern at Facebook in Menlo Park. I studied misinformation sharing on the platform and helped evaluate and deploy a new tool related to their ongoing election integrity efforts.
At Rutgers, I teach undergraduate classes on Political Sociology, Sociology of Culture, and Data Science. At the graduate level, I teach Computational Sociology and a second-semester Statistics course. Code and slides for my graduate methods classes are available on Github.
Email me at thomas dot davidson at rutgers dot edu.