Steph Buongiorno

Steph Buongiorno, PhD, is a researcher and a teacher. She designs computational methods for knowledge production in the digital humanities. Her work reconfigures traditional academic boundaries and opens up new approaches to interpreting texts, data, and culture.

She enjoys poetry, all sorts of sensory things, as well as quiet, underwater worlds.

Featured Works

See all

Apps

Articles

Artificial Intelligence

Books

Congress

Course Materials

Cuba

Datasets

Digital Humanities

Government Data

Hansard

Knowledge Graphs

GAME-KG

Knowledge graphs (KGs) can augment large language models (LLMs) while also providing an explainable set of facts that can be inspected by a human. Explainability is valuable for fields that may otherwise avoid LLMs due to hallucinations, such as human trafficking analysis. Creating KGs poses challenges, however. KGs parsed from documents may include explicit connections (those directly stated in a document) but miss implicit connections (those evident to a human, but not directly stated). This research introduces GAME-KG, an approach to modifying explicit and implicit KG connections by crowdsourcing feedback through video games.

Dark Shadows

Dark Shadows is a film-noir style detective thriller that acts as a test bed for proof-of-concept and prototype system components, frameworks, and models that contribute to research in AI and machine learning. The gameplay focuses on social scenarios where players provide natural language input to progress the narrative. Dark Shadows includes PANGeA’s novel validation system, which leverages self-reflection to evoke a large language model's (LLM) intelligence when evaluating and responding to user input. Narrative and artwork are procedurally generated.

Agent-Driven, Game-Based Learning: Personalized CS Education for Diverse Students

AI agents can personalize education by identifying students’ strengths, weaknesses, and personalities to generate content tailored to them. This work presents "personalized education agents" deployed in an educational version of Minecraft. Agents bridge concepts from lessons to "big picture" thinking by creating connections between STEM and interdisciplinary topics, such as the Language Arts. Agents translate student progression and learning outcomes to teachers for their assessment of student progress.

The Hansard Viewer Demo App

Use an array of data-mining and statistical approaches to gain new insights into the evolution and nature of political language as it occurs in different time periods and in different contexts.

rOpengov Universe

Open government data enables citizens to see government activities and decision-making processes, empowering them to participate more fully in the democratic process. Researchers can use open government data for studies and projects, generating insights and contributing to evidence-based policy-making. To this end, this work introduces R packages hosted by the rOpengov Universe that are designed to make analyzing contemporary and historical open government data more accessible. Queried data is returned in a clean and analysis-ready dataframe.

Course Materials: Digital History

Computational methods are changing the way that we access information about history and society. These methods help us to detect change over time, to identify influential figures, and to name turning points. What happens when we apply these tools to the entire Hansard corpus or to a million congressional debates and tweets? This work provides an introduction to the analytic methodologies transforming the humanities and social sciences via a book, under contract at Cambridge University Press, and series of Jupyter Notebooks aimed at exploring questions like these.

The 19th-Century British Parliamentary Debates (Hansard Corpus)

ENTER

Syntactic Dependency Relationships and the Extraction of Grammatical Triples

This paper describes a method of triples extraction, posextract, which has been designed to meet the increasing need for high-accuracy triples outputs for the analysis of text.

Evaluating the Efficacy of LLMs to Emulate Human Personalities for Video Game Play

To improve the realism of affective Non-Player Characters (NPCs) in video games, this study investigates whether Large Language Models (LLMs) can emulate human personalities. Using the Big Five framework and over 50,000 responses from the International Personality Item Pool (IPIP), LLMs were prompted with self-assessment items corresponding to various personality profiles. Their outputs were then compared to human baseline responses to evaluate accuracy and consistency. Results showed that while some local models exhibited no alignment with human profiles, certain frontier models achieved high alignment. These findings suggest that LLMs can provide a method for designing NPCs with more realistic, personality-driven behavior.