AI-Assisted Curation of Conference Scholarship: Compiling, Structuring, and Analyzing Two Decades of Presentations at the Society for Social Work and Research

This study utilizes AI-assisted curation to compile and analyze a comprehensive database of 23,793 presentations from the Society for Social Work and Research Annual Conference (2005–2026), revealing significant growth in participation, collaboration, and international engagement alongside a continued predominance of quantitative research methods.

The Gist

Imagine the world of social work research as a massive, bustling Grand Bazaar. Every year, thousands of scholars, students, and practitioners gather to set up their stalls, display their latest discoveries, and trade ideas. This is the Society for Social Work and Research (SSWR) Annual Conference.

For over 20 years, this bazaar has been growing. But until now, nobody had a map. The "stalls" (presentations) were listed in a giant, messy notebook where the handwriting changed every year, the ink faded, and the descriptions were written in different languages. It was impossible to see the big picture: Who is selling what? How has the market changed? Are people working alone or in teams?

This paper is the story of how three researchers built a super-powered, AI-assisted map of this entire bazaar, covering everything from 2005 to 2026.

Here is the breakdown of their journey, using some everyday analogies:

1. The Problem: A Mountain of Unorganized Paper

Think of the conference archives as a library where every book is written on a different type of paper, in a different font, and glued together in a random order.

• The Challenge: Researchers wanted to study how social work research has changed over two decades, but the data was "unstructured." It was just text on a website, not a neat spreadsheet.
• The Old Way: To fix this, humans would have to sit down and read thousands of abstracts one by one, typing everything into a computer. It would take a lifetime.

2. The Solution: The "Smart Librarian" Robot

Instead of hiring an army of humans, the authors used a Small Language Model (SLM).

• The Analogy: Imagine hiring a super-smart, tireless librarian robot. Unlike the giant, expensive "super-computers" that eat up a city's worth of electricity, this robot is small enough to run on a regular laptop. It's like a Swiss Army Knife compared to a bulldozer.
• What it did: The robot read every single abstract, figured out who the authors were, where they worked, and what kind of research they did (like sorting apples from oranges). It even cleaned up messy names (turning "J. Smith, U. Mich" into "John Smith, University of Michigan").
• The Safety Net: The human researchers didn't just let the robot run wild. They acted as quality control inspectors, checking the robot's work at every step to make sure it didn't make silly mistakes.

3. The Map They Drew (The Results)

Once the robot organized the data, the researchers could finally see the patterns. Here is what their map revealed:

• The Bazaar is Exploding:
  The conference has grown like a dandelion in spring. In 2005, there were about 423 presentations. By 2026, that number jumped to nearly 2,000. The field is doubling in size roughly every ten years.

• The "Team Sport" Shift:
  In the old days, researchers mostly worked alone (like a solo artist painting a picture). Now, it's more like a football team. The average number of authors on a single presentation has gone from 2.2 to 3.3. People are collaborating more, bringing different skills together to solve complex problems.

• The Methodology Mix:
  For a long time, the bazaar was dominated by Quantitative Research (numbers, charts, and statistics)—think of it as the "hard science" section. It still holds the biggest spot (about 61%). However, the Qualitative section (stories, interviews, and human experiences) has grown significantly, almost doubling its share. It's as if the bazaar is realizing that while numbers tell you how many people are hungry, stories tell you why they are hungry and what they need.

• The Global Crowd:
  The bazaar is becoming more international. In 2005, almost everyone was from the US. Now, about 1 in 7 participants is from another country. However, the authors noticed a recent dip in international first-time presenters, likely due to travel barriers (like visa issues or high costs), which acts like a "gate" that is getting harder for some people to cross.

• Who is Presenting?
  The bazaar is fueled by the next generation. Doctoral students and Assistant Professors are the most frequent sellers, making up nearly 40% of the crowd. This shows that the conference is a vital training ground for young scholars.

4. Why This Matters

Why did the authors go through all this trouble?

• To See the Future: Conferences are where ideas are born before they become published books. By mapping this data, we can see which topics are trending before they hit the mainstream.
• To Democratize Science: By using a small, efficient AI model, they showed that you don't need a billion-dollar budget to analyze huge amounts of data. You can do it with a laptop and a smart algorithm.
• To Build a Foundation: They didn't just write a report; they built a digital infrastructure. They are making this massive, organized database available to other researchers so they can ask new questions, like "How has research on poverty changed?" or "Which universities are leading the way?"

The Bottom Line

This paper is about turning a chaotic, messy pile of paper into a clear, navigable map. It shows us that social work research is growing fast, becoming more collaborative, and valuing both hard numbers and human stories. And it proves that with a little help from a smart, efficient robot, we can understand the heartbeat of a whole scientific field.

Technical Summary

Here is a detailed technical summary of the paper "AI-Assisted Curation of Conference Scholarship: Compiling, Structuring, and Analyzing Two Decades of Presentations at the Society for Social Work and Research (2005–2026)."

1. Problem Statement

The Society for Social Work and Research (SSWR) Annual Conference represents a massive repository of scholarly activity in the field of social work, yet this data remains largely unanalyzed due to its format.

• Unstructured Data: Unlike journal publications which have standardized metadata, SSWR conference abstracts exist as unstructured text within the Confex conference management system.
• Analytic Barrier: The lack of structured metadata (e.g., standardized author affiliations, methodology tags, institutional data) prevents systematic bibliometric or scientometric analysis.
• Resource Constraints: Traditional methods of curating such data require extensive manual annotation, which is infeasible for a dataset spanning two decades and nearly 24,000 records. Furthermore, reliance on large proprietary AI models raises concerns regarding cost, data privacy, and environmental sustainability (carbon footprint).

2. Methodology

The authors developed a multi-stage, AI-assisted pipeline to transform unstructured web data into a structured, analyzable database.

Data Collection & Preprocessing:

• Source: The SSWR Confex archive (2005–2026), covering 22 years of conferences.
• Web Scraping: Custom Python scripts using Crawl4AI and BeautifulSoup4 (with lxml parser) were used to extract titles, abstracts, and author strings.
• Cleaning: Records with abstracts under 50 characters were excluded. The pipeline handled evolving HTML structures across different conference years.

AI-Driven Extraction (Small Language Models):

• Model Selection: The authors utilized gpt-oss:20b, a Small Language Model (SLM) with 20 billion parameters. This model was chosen for its balance of performance and efficiency, allowing for local deployment (via llama.cpp) without reliance on proprietary cloud APIs.
• Affiliation Parsing: The SLM was prompted to extract structured fields (Name, Degree, Position, Institution, City, State, Country) from heterogeneous author strings, outputting JSON.
• Methodology Classification: The same SLM classified abstracts into five categories: Quantitative, Qualitative, Mixed Methods, Review, and Theoretical/Other.
• Validation: Human review was conducted at every major stage. Inter-rater agreement (Cohen's kappa) between human and model for methodology classification was 0.83.

Entity Resolution & Normalization:

• Author Deduplication: A graph-based entity resolution system was implemented using RapidFuzz and NetworkX.
  • Blocking: Names were grouped by first initial and last name to reduce computational complexity.
  • Similarity Scoring: Token sort-ratio was used for general matching. Special logic was applied to common Asian surnames (requiring exact first-name matches) to prevent false positives.
  • Threshold: A similarity score of 90+ connected name variants to a single canonical identity.
• Institution Normalization: A mapping table consolidated variants (e.g., "U. Michigan" vs. "University of Michigan") using a combination of exact, substring, and pattern matching.
• Position Classification: Positions were normalized into a standard taxonomy (e.g., Doctoral Student, Assistant Professor, Practitioner) using regex and keyword matching.

3. Key Contributions

1. Infrastructure Creation: The study produced the first comprehensive, structured database of SSWR presentations, containing 23,793 presentations and 69,924 author records representing 20,779 unique researchers.
2. Methodological Innovation: Demonstrated the efficacy of Small Language Models (SLMs) for large-scale academic curation. The study proved that locally hosted SLMs can achieve high accuracy in entity extraction and classification while offering benefits in data privacy, cost reduction, and lower environmental impact compared to frontier models.
3. Longitudinal Dataset: Provided a 22-year (2005–2026) longitudinal view of the discipline's research output, filling a gap where previous analyses were limited to small subsets or journal publications.

4. Key Results

• Growth: Conference presentations grew from 423 in 2005 to 1,935 in 2026, a Compound Annual Growth Rate (CAGR) of 7.5%. A significant inflection point occurred in 2015 (46% increase in one year).
• Methodology Trends:
  • Quantitative methods remain dominant but are declining in share (from ~71% in early years to ~51% in 2025–2026).
  • Qualitative methods are rising significantly (from 15% to 30%).
  • Mixed Methods and Reviews have shown modest but steady growth.
• Collaboration:
  • The mean number of authors per presentation increased from 2.22 (2005) to 3.31 (2026).
  • Single-authored presentations dropped from 38% to 21%, while presentations with four or more authors rose from 16% to 38%.
• Authorship Demographics:
  • Assistant Professors (19.6%) and Doctoral Students (19.0%) are the most frequent contributors.
  • There is a shift in faculty representation, with Associate Professors increasing their share while Assistant Professors' share declines slightly in recent years.
• Geographic Reach:
  • US institutions dominate (89.1%), but international participation (non-US affiliations) grew from 4.5% (2005) to 13.7% (2026).
  • Notable growth in participation from Canada, South Korea, and Israel.

5. Significance

• Research Infrastructure: The database serves as a foundational tool for testing hypotheses regarding disciplinary evolution, research priorities, and the trajectory of social work scholarship. It allows researchers to track trends before they appear in formal journal publications.
• Sustainable AI in Academia: The paper provides a blueprint for using efficient, locally hosted SLMs to democratize access to large-scale text analysis, making advanced computational methods accessible to disciplines with fewer resources than computer science.
• Disciplinary Insights: The findings reveal a maturing discipline characterized by increased collaboration, a diversification of methodological approaches (moving beyond pure quantitative dominance), and growing global engagement, despite recent geopolitical and logistical challenges affecting international travel.
• Policy Implications: The data highlights the need for conference organizers to support early-career researchers and international scholars, as the scale of the conference creates barriers to access and networking that require targeted mentoring and structural support.