A5000967986- Hate Speech and Cyberbullying Detection
- Misinformation and Its Impacts
- Psychology of Moral and Emotional Judgment
- Social and Intergroup Psychology
- Topic Modeling
- Natural Language Processing Techniques
- Adversarial Robustness in Machine Learning
- Sentiment Analysis and Opinion Mining
- Cultural Differences and Values
- Ethics and Social Impacts of AI
- Computational and Text Analysis Methods
- Social Media and Politics
- Terrorism, Counterterrorism, and Political Violence
- Mobile Crowdsensing and Crowdsourcing
- Cybercrime and Law Enforcement Studies
- Multimodal Machine Learning Applications
- Innovative Teaching Methodologies in Social Sciences
- Explainable Artificial Intelligence (XAI)
- Context-Aware Activity Recognition Systems
- Risk Perception and Management
- Impact of AI and Big Data on Business and Society
- Psychosomatic Disorders and Their Treatments
- Personal Information Management and User Behavior
- Emotions and Moral Behavior
- Experimental Behavioral Economics Studies
Google (United States)
2021-2024
University of Southern California
2018-2023
Virginia Tech
2023
Southern California University for Professional Studies
2019-2021
Snap (United States)
2021
University of California, Berkeley
2020
Bocconi University
2020
University of Michigan
2020
University of Massachusetts Amherst
2020
Pacific Northwest National Laboratory
2020
Abstract Majority voting and averaging are common approaches used to resolve annotator disagreements derive single ground truth labels from multiple annotations. However, annotators may systematically disagree with one another, often reflecting their individual biases values, especially in the case of subjective tasks such as detecting affect, aggression, hate speech. Annotator capture important nuances that ignored while aggregating annotations a truth. In order address this, we investigate...
Research has shown that accounting for moral sentiment in natural language can yield insight into a variety of on- and off-line phenomena such as message diffusion, protest dynamics, social distancing. However, measuring is challenging, the difficulty this task exacerbated by limited availability annotated data. To address issue, we introduce Moral Foundations Twitter Corpus, collection 35,108 tweets have been curated from seven distinct domains discourse hand at least three trained...
Hate speech classifiers trained on imbalanced datasets struggle to determine if group identifiers like “gay” or “black” are used in offensive prejudiced ways. Such biases manifest false positives when these present, due models’ inability learn the contexts which constitute a hateful usage of identifiers. We extract post-hoc explanations from fine-tuned BERT detect bias towards identity terms. Then, we propose novel regularization technique based that encourages models context addition...
A common practice in building NLP datasets, especially using crowd-sourced annotations, involves obtaining multiple annotator judgements on the same data instances, which are then flattened to produce a single “ground truth” label or score, through majority voting, averaging, adjudication. While these approaches may be appropriate certain annotation tasks, such aggregations overlook socially constructed nature of human perceptions that annotations for relatively more subjective tasks meant...
Abstract Social stereotypes negatively impact individuals’ judgments about different groups and may have a critical role in understanding language directed toward marginalized groups. Here, we assess the of social automated detection hate speech English by examining on annotation behaviors, annotated datasets, classifiers. Specifically, first investigate novice annotators’ their hate-speech-annotation behavior. Then, examine effect normative aggregated large corpus. Finally, demonstrate how...
Research has shown that accounting for moral sentiment in natural language can yield insight into a variety of on- and off-line phenomena, such as message diffusion, protest dynamics, social distancing. However, measuring is challenging the difficulty this task exacerbated by limited availability annotated data. To address issue, we introduce Moral Foundations Twitter Corpus, collection 35,108 tweets have been curated from seven distinct domains discourse hand-annotated at least three...
Xisen Jin, Francesco Barbieri, Brendan Kennedy, Aida Mostafazadeh Davani, Leonardo Neves, Xiang Ren. Proceedings of the 2021 Conference North American Chapter Association for Computational Linguistics: Human Language Technologies. 2021.
Abstract Understanding motivations underlying acts of hatred are essential for developing strategies to prevent such extreme behavioral expressions prejudice (EBEPs) against marginalized groups. In this work, we investigate the EBEPs as a function moral values. Specifically, propose may often be best understood morally motivated behaviors grounded in people’s values and perceptions violations. As evidence, report five studies that integrate spatial modeling experimental methods relationship...
Large language models are increasingly being integrated into applications that shape the production and discovery of societal knowledge such as search, online education, travel planning. As a result, will how people learn about, perceive interact with global cultures making it important to consider whose systems perspectives represented in models. Recognizing this importance, work Machine Learning NLP has focused on evaluating gaps cultural representational distribution within outputs....
Societal stereotypes are at the center of a myriad responsible AI interventions targeted reducing generation and propagation potentially harmful outcomes. While these efforts much needed, they tend to be fragmented often address different parts issue without taking in unified or holistic approach about social how impact various machine learning pipeline. As result, it fails capitalize on underlying mechanisms that common across types stereotypes, anchor particular aspects relevant certain...
Online radicalization is among the most vexing challenges world faces today. Here, we demonstrate that homogeneity in moral concerns results increased levels of radical intentions. In Study 1, find Gab—a right-wing extremist network—the degree convergence within a cluster predicts number hate-speech messages members post. 2, replicate this observation another network, Incels. Studies 3 to 5 ( N = 1,431), experimentally leading people believe others their hypothetical or real group share...
Infectious diseases have been an impending threat to the survival of individuals and groups throughout our evolutionary history. As a result, humans developed psychological pathogen-avoidance mechanisms societal norms that respond presence disease-causing microorganisms in environment. In this work, we demonstrate morality plays central role cultural architectures help avoid pathogens. We present collection studies which together provide integrated understanding socio-ecological impacts...
Moral framing and sentiment can affect a variety of online offline behaviors, including donation, pro-environmental action, political engagement, even participation in violent protests. Various computational methods Natural Language Processing (NLP) have been used to detect moral from textual data, but order achieve better performances such subjective tasks, large sets hand-annotated training data are needed. Previous corpora annotated for proven valuable, generated new insights both within...
We present the Gab Hate Corpus (GHC), consisting of 27,665 posts from social network service gab.com, each annotated for presence “hate-based rhetoric” by a minimum three annotators. Posts were labeled according to coding typology derived synthesis hate speech definitions across legal precedent, previous typologies, and psychology sociology, comprising hierarchical labels indicating dehumanizing violent as well indicators targeted groups rhetorical framing. provide inter-annotator agreement...
Given its centrality in scholarly and popular discourse, morality should be expected to figure prominently everyday talk. We test this expectation by examining the frequency of moral content three contexts, using methods: (a) Participants' subjective estimates (N = 581); (b) Human analysis unobtrusively recorded in-person interactions 542 participants; n 50,961 observations); (c) Computational Facebook posts 3822 111,886 observations). In their self-reports, participants estimated that 21.5%...
Humans use language toward hateful ends, inciting violence and genocide, intimidating denigrating others based on their identity. Despite efforts to better address the of hate in public sphere, psychological processes involved remain unclear. In this work, we hypothesize that morality are concomitant language. a series studies, find evidence support hypothesis using from diverse array contexts, including propaganda inspire genocide (Study 1), slurs as they occur large text corpora across...
Lambert Mathias, Shaoliang Nie, Aida Mostafazadeh Davani, Douwe Kiela, Vinodkumar Prabhakaran, Bertie Vidgen, Zeerak Waseem. Proceedings of the 5th Workshop on Online Abuse and Harms (WOAH 2021). 2021.
Acts of hate have been used to silence, terrorize, and erase marginalized social groups throughout history. The rising rates these behaviors in recent years underscores the importance developing a better understanding when, why, where they occur. In this work, we present program research that suggests acts may often be best understood not just as responses threat, but also morally motivated grounded people’s moral values perceptions violations. As evidence for claim, findings from five...
Language is a psychologically rich medium for human expression and communication. While language usage has been shown to be window into various aspects of people's social worlds, including their personality traits everyday environment, its correspondence moral concerns yet considered. Here, we examine the relationship between Care, Fairness, Loyalty, Authority, Purity as conceptualized by Moral Foundations Theory. We collected Facebook status updates (N = 107,798) from English-speaking...
Recent years have seen substantial investments in AI-based tools designed to detect offensive language at scale, aiming moderate social media platforms, and ensure safety of conversational AI technologies such as ChatGPT Bard. These efforts largely treat this task a technical endeavor, relying on data annotated for offensiveness by global crowd workforce, without considering workers' socio-cultural backgrounds or the values their perceptions reflect. Existing research that examines...