ScholarWorks@Georgia State University
Recent Submissions
Publication Open Access A Machine-Assisted Framework for Mapping Computer Science Education to Workforce Demands Using NLP and Large Language Models(2026-07-16)Students in computer science programs are routinely required to make consequential decisions about coursework and career preparation without clear, evidence-based guidance connecting individual courses to evolving workforce demand. While prerequisite structures and degree requirements are well defined, the relevance of specific courses to skills and careers is often implicit, informal, or static. At the same time, job roles, required skills, and even occupational categories change on timescales that outpace curricular revision, leaving students to infer meaning from incomplete or biased sources. This dissertation investigates how natural language processing, semantic clustering, and large language models can be used to construct a machine-assisted framework for aligning computer science courses with evolving workforce skill demand. The investigation prioritizes informed student decision-making rather than automated prediction or recommendation. It emphasizes open-vocabulary skill discovery, interpretable intermediate representations, and human-in-the-loop refresh cycles that allow alignments to be revisited as workforce demand changes. Workforce signals are derived from job postings, from which skills are extracted using open-vocabulary methods. Extracted skills are canonicalized through alias-based normalization to reduce fragmentation while preserving the ability to incorporate emerging tools and practices. Skills are weighted based on prevalence and whether they are identified as required or preferred, producing role–skill profiles that balance stability with sensitivity to change. Course-level skill representations are constructed from syllabi, catalog descriptions, and prerequisite context, acknowledging the incompleteness and variability of instructional artifacts. Course–career alignment scores are then computed as comparative relevance signals intended to support exploration and reasoning. The framework was integrated into a deployed, student-facing system that visualizes prerequisite structure and overlays career relevance without prescribing outcomes. Evaluation examined technical properties such as coherence, stability across refresh cycles, and responsiveness to emerging skills, as well as student-reported understanding and confidence. A comparison against the O*NET 29.1 reference database found that between 91% and 99% of extracted skills have no direct O*NET equivalent, and that O*NET descriptors account for only approximately 31% of skill demand by posting volume, providing direct empirical justification for the open-vocabulary approach.Publication Open Access Between Silence and Power: Ethnic Recognition Strategies and Political Stability in Post-Conflict States(2026-07-15)Ethnic conflicts are one of the hardest to resolve, as they are based on identity, power, and histories. In the aftermath of conflicts fought along ethnic lines, there is the question of what institutional arrangements would best promote long term political stability and reduce ethnic tensions. Should states recognize ethnic groups under accommodationist principles, or blur over ethnic differences under integration principles? This dissertation examines the idea of ethnic recognition, propounding two dimensions on which it can be assessed: visibility and institutionalization. Based on these two dimensions, three recognition types are put forward: non-recognition, echoed recognition, and embodied recognition, with different indicators on visibility and institutionalization. Using the outcomes of contentious ethnic mobilization and the masking of minority grievances, this dissertation argues for echoed recognition, with high visibility and low institutionalization, as the most conducive recognition strategy for long term political stability in the aftermath of ethnic conflictItem Open Access Writing Space, Righting Place: Language as a Heterotopic Space in Olaudah Equiano's Interesting Narrative(2012-11-26)Olaudah Equiano or Gustavas Vassa may have had abolitionist motivations when writing The Interesting Narrative of the Life of Olaudah Equiano or Gustavas Vassa, the African, Written by Himself, but the function of the text is much different and self-serving. Specifically, in looking closely at the wording of the text, with its language of we versus they, in group versus out group, ours versus theirs, Equiano clearly feels he at no time belongs fully to any specific group or place; rather, he only partially belongs anywhere, and thus, creates this work of autobiography and appropriation of fiction and oral tradition to negotiate and cultivate his own liminal, or even heterotopic, space. In other words, I suggest he may have used the writing of this text to define his sense of self, creating a space in which he was both in control and fully belonged.Publication Open Access When Algorithms Enter the Room: Human Responses to Algorithm-Generated Content in Digital Content Markets(2026-07-10)As algorithms become increasingly integrated into digital content markets, algorithm-generated and human-generated content increasingly coexist within the same marketplace, and both are presented to market audiences simultaneously. Because humans must decide when and how to use algorithms to replace or augment them in tasks such as content generation, it is critical to understand how market participants respond to algorithm-generated content under this coexistence. This dissertation investigates market participants’ responses across two distinct content types and market contexts. Essay 1 studies creative content in the digital art market, where artworks generated by artificial intelligence (AI) coexist with non-AI artworks (i.e., human-generated artworks created without AI tools). I examine how art audiences respond to artists’ adoption of AI tools. The findings show that AI adoption does not significantly change the overall amount of audience feedback on artists’ non-AI artworks, but it increases the amount of audience feedback on less-experienced artists’ non-AI artworks. This effect is likely driven by a feedback-driven learning mechanism: less-experienced artists learn from audience feedback on their AI-generated artworks and improve their subsequent non-AI artworks. Essay 2 studies evaluative content in the online review market, where algorithm-generated ratings (AGRs) coexist with human reviewer ratings. I examine how human reviewers respond to AGRs under increased public scrutiny. The findings show that increased public scrutiny drives reviewers to lower their ratings, become less likely to rate above AGRs, and align more closely with AGRs. These effects are stronger among low-performing reviewers. Together, the two essays contribute to research on human–algorithm collaboration by showing that algorithm-generated content can complement humans in different ways: it facilitates learning in creative work and serves as a benchmark in evaluative work.Publication Open Access Neural Correlates of Empathy-Related Responding in Middle Childhood(2026-07)Mother’s emotions provide a key context for children’s emotional development. Children interpret these emotions through the lens of their relationship, with empathy, guilt, and reparative behaviors toward mothers emerging early in life. Over time, these interactions are thought to support regulated empathic responses, though far less is known about children’s responses to mothers’ emotions beyond early childhood, especially neural activity and responses to positive emotions. Middle childhood is a period in which children’s empathic responses become more differentiated and shaped by relational meaning and personal relevance, yet this developmental window remains understudied. The current study used an fMRI design to examine blood-oxygen-level-dependent (BOLD) activity associated with children’s empathic responses to their mothers’ emotions, capturing variation by relational context (mother vs. stranger), emotional valence (happiness vs. sadness), and personal responsibility (child-related vs. child-unrelated). Children (N = 33) viewed images of their mothers and a stranger expressing happiness or sadness while hearing audio recordings describing the cause of the emotion that was related or unrelated to their behavior. BOLD activity was examined in regions engaged during affect sharing, mentalizing, forming an empathic understanding, self-referential processing, emotion regulation, and reward processing. Across all emotional conditions, the mPFC showed consistently greater BOLD activity than a non‑emotional control, suggesting children process others’ emotions through a broadly self‑referential lens. The amygdala, dmPFC, and vmPFC showed greater activity when children viewed their mother’s sadness (vs. control), and the amygdala responded more strongly to maternal than stranger sadness, highlighting the salience of caregiver distress. Personal responsibility further modulated BOLD activity: the dmPFC had greater activity during child-related than child-unrelated maternal sadness, and both dmPFC and mPFC had greater activity for child-related maternal sadness than happiness, reflecting greater mentalizing and self-referential processing when children perceived themselves as responsible for their mother’s distress. Other regions showed minimal differentiation across conditions. Overall, children’s BOLD activity when viewing their mother’s emotions in middle childhood was selective, context‑sensitive, and varied by emotional valence, relational context, and personal responsibility, with a central role of self-referential processing. These results provide insight into how children integrate relational and causal information when interpreting their mothers’ emotions.
Communities in ScholarWorks@GSU
Select a community to browse its collections.