Mining social media data for studying the human condition has created new and unique challenges. When analyzing social media data from marginalized communities, algorithms lack the ability to accurately interpret off-line context, which may lead to dangerous assumptions about and implications for marginalized communities. To combat this challenge, we hired formerly gang-involved young people as domain experts for contextualizing social media data in order to create inclusive, community-informed algorithms. Utilizing data from the Gang Intervention and Computer Science Project—a comprehensive analysis of Twitter data from gang-involved youth in Chicago—we describe the process of involving formerly gang-involved young people in developing a new part-of-speech tagger and content classifier for a prototype natural language processing system that detects aggression and loss in Twitter data. We argue that involving young people as domain experts leads to more robust understandings of context, including localized language, culture, and events. These insights could change how data scientists approach the development of corpora and algorithms that affect people in marginalized communities and who to involve in that process. We offer a contextually driven interdisciplinary approach between social work and data science that integrates domain insights into the training of qualitative annotators and the production of algorithms for positive social impact.

Researchers in computer science have spent considerable time developing methods to increase the accuracy and richness of annotations. However, there is a dearth in research that examines the positionality of the annotator, how they are trained and what we can learn from disagreements between different groups of annotators. In this study, we use qualitative analysis, statistical and computational methods to compare annotations between Chicago-based domain experts and graduate students who annotated a total of 1,851 tweets with images that are a part of a larger corpora associated with the Chicago Gang Intervention Study, which aims to develop a computational system that detects aggression and loss among gang-involved youth in Chicago. We found evidence to support the study of disagreement between annotators and underscore the need for domain expertise when reviewing Twitter data from vulnerable populations. Implications for annotation and content moderation are discussed.

Gang violence is a severe issue in major cities across the U.S. and recent studies have found evidence of social media communications that can be linked to such violence in communities with high rates of exposure to gang activity. In this paper we partnered computer scientists with social work researchers, who have domain expertise in gang violence, to analyze how public tweets with images posted by youth who mention gang associations on Twitter can be leveraged to automatically detect psychosocial factors and conditions that could potentially assist social workers and violence outreach workers in prevention and early intervention programs. To this end, we developed a rigorous methodology for collecting and annotating tweets. We gathered 1,851 tweets and accompanying annotations related to visual concepts and the psychosocial codes: aggression, loss, and substance use. These codes are relevant to social work interventions, as they represent possible pathways to violence on social media. We compare various methods for classifying tweets into these three classes, using only the text of the tweet, only the image of the tweet, or both modalities as input to the classifier. In particular, we analyze the usefulness of mid-level visual concepts and the role of different modalities for this tweet classification task. Our experiments show that individually, text information dominates classification performance of the loss class, while image information dominates the aggression and substance use classes. Our multimodal approach provides a very promising improvement (18% relative in mean average precision) over the best single modality approach. Finally, we also illustrate the complexity of understanding social media data and elaborate on open challenges. The annotated dataset will be made available for research with strong ethical protection mechanism.

How should we interpret gun images on social media? Take for example the shooting at Stoneman Douglas High School. Media articles revealed that the gunman, a white adolescent male, posted images of firearms and other weapons on his social media profile prior to the shooting. On the other hand, when it comes to Black communities, digital policing strategies often intercept images with guns and individuals thought to be associated with gangs before a crime is ever committed. In this study, we use a mixed methods approach, situated in social systems theory, to make meaning of gun posting behavior among Black youth who associate with gangs in Chicago. We collected and examined a corpus of Twitter images (1851) through snowball sampling of a well-known deceased gang member in Chicago and users in their Twitter network. We identified 560 images that contain guns and asked two distinct groups to annotate images: formerly gang-involved outreach workers, known as community domain experts, at a local Chicago violence prevention organization and Master of Social Work students at Columbia University. After comparing their results, findings highlighted the prevalence and frequency of gun image posting within this corpus and critical differences in how community domain experts and social work annotators perceive guns. The various underlying intents provide a rich source of knowledge for understanding the symbolic nature of guns in the digital age.

Gang violence is a severe issue in major cities across the U.S. and recent studies [23] have found evidence of social media communications that can be linked to such violence in communities with high rates of exposure to gang activity. In this paper we partnered computer scientists with social work researchers, who have domain expertise in gang violence, to analyze how public tweets with images posted by youth who mention gang associations on Twitter can be leveraged to automatically detect psychosocial factors and conditions that could potentially assist social workers and violence outreach workers in prevention and early intervention programs. To this end, we developed a rigorous methodology for collecting and annotating tweets. We gathered 1,851 tweets and accompanying annotations related to visual concepts and the psychosocial codes: aggression, loss, and substance use. These codes are relevant to social work interventions, as they represent possible pathways to violence on social media. We compare various methods for classifying tweets into these three classes, using only the text of the tweet, only the image of the tweet, or both modalities as input to the classifier. In particular, we analyze the usefulness of mid-level visual concepts and the role of different modalities for this tweet classification task. Our experiments show that individually, text information dominates classification performance of the loss class, while image information dominates the aggression and substance use classes. Our multimodal approach provides a very promising improvement (18% relative in mean average precision) over the best single modality approach. Finally, we also illustrate the complexity of understanding social media data and elaborate on open challenges.