Projects in Computational Social Science

Below I detail contributions to publications in Computational Social Science, focusing on contributions in which I applyied NLP/ML algorithms or conducted statistical analysis.

The (Moral) Language of Hate

• Under Review
• My contributions: Lead author; formulated and designed the main hypothesis of the paper; designed three studies to measure the relationship between hate and morality in different language modalities, including word embeddings, annotated social media posts, and historical records of Nazi propaganda speeches and writings; coordinated data collection and annotation; performed all NLP tasks (e.g., text classification, word embeddings); performed statistical analysis; and led writing and editing.

Moral Concerns are Differentially Observable in Language

• Published in: Cognition (Impact Factor: 3.65)
• My contributions: Lead author and designer of the studies; formulated main hypotheses; performed exploratory data analysis and visualization; applied a suite of NLP methods for supervised text prediction (i.e., text regression) for 5 different psychographic variables (i.e., user-reported moral concerns as measured by a validated questionnaire taken online by voluntary participatns, who then volunteered their Facebook data); performed all statistical analysis; led in writing and editing.

Pathogens are linked to human moral systems across space and time

• Published in: Current Research in Ecological Science (Impact Factor: 6.81)
• My contributions: Conducted a study in which I compared the “moral loading” of pathogen-related words in a variety of different languages. The goal of this study was to determine which, among the five “Moral Foundations” (Care, Fairness, Loyalty, Authority, and Purity), are most strongly related to pathogens in a semantic sense. The technique we use is to first generate word embeddings using pre-trained FastText embedding models, average embeddings together, and compute cosine similarity between the pathogen word centroids and the moral word centroids (see Garten et al., 2018).

Morally Homogeneous Networks and Radicalism

• Published in: Social Psychological and Personality Science (Impact Factor: 3.61)
• My contributions: contributed to the construction and execution of a study examining the relationship between “moral centrality” in a social network and the likelihood of posting hate speech. Moral centrality, in this case, was measured using the use of moral language, as indicating by lexicon-based representations (see above). Individuals in a given social network were considered to be in close proximity if they used a similar distribution of moral language.

Hate Speech Classifiers Learn Human-Like Social Stereotypes

• Under revision in: Transactions of the ACL (14.27)
• My contributions: Writing and editing; aided in study design, in which we evaluated the presence of bias in individual data annotators’ annotations and the subsequent effect of such biases in downstream classifiers (trained on these annotations)