Fostering Knowledge Infrastructures in Science Communication and Aerospace Engineering

This doctoral work proposes a framework for fostering robust knowledge infrastructures in fragmented domains like science communication and aerospace engineering by implementing tool-supported workflows—including AI, knowledge graphs, and stakeholder-driven interfaces—while highlighting the need to address persistent societal and legal barriers to achieve full adoption.

The Gist

Imagine you have two massive libraries.

In the first library (Aerospace Engineering), the books are incredibly detailed blueprints for building rockets. But here's the catch: the books are locked in different rooms, written in different languages, and the librarians are afraid to let anyone borrow them because of strict security rules. Engineers spend hours searching for a solution that someone else already wrote down three years ago, but they can't find it.

In the second library (Science Communication), there are thousands of videos and podcasts explaining science to the public. These are scattered across YouTube, Spotify, and personal hard drives. While the content is great, there's no central map to find the best videos, no way to check if the facts are true, and no system to connect a video about "black holes" to the actual research paper that proved it.

Tim Wittenborg's research is like building a super-smart, universal "connect-the-dots" system to fix both libraries.

Here is the breakdown of his work in simple terms:

1. The Problem: The "Silo" Effect

Right now, knowledge is trapped in "silos" (isolated towers).

• In Engineering: Experts don't share because of legal fears or because their data formats don't match.
• In Science Media: Great videos exist, but they aren't tagged or organized in a way that computers can understand.
• The Result: We are wasting time and money reinventing the wheel, and the public can't easily find trustworthy science info.

2. The Solution: A "Smart Librarian" System

Tim is building a set of tools (a Knowledge Infrastructure) that acts like a super-efficient librarian who speaks every language and knows every rule. This system has three main superpowers:

• The Translator (AI & Automation):
  Imagine a robot that can read a 50-page engineering report or watch a 20-minute science video, extract the most important facts, and instantly translate them into a standard format that everyone understands. This is what Tim calls "Human-in-the-loop AI." It's not just a robot doing everything; it's a robot doing the heavy lifting, while a human expert double-checks the work to make sure it's accurate.

  • Analogy: Think of it like a spell-checker that doesn't just fix typos, but rewrites your whole essay to make sure the facts are correct, then asks a teacher to sign off on it.

• The Universal Map (Digital Libraries & Wikis):
  Instead of having one giant, messy warehouse, Tim is building a network of smaller, specialized "mini-libraries" (called Wikibase instances) that can talk to each other.

  • For Aerospace, he created a secure, structured map where engineers can find existing designs without breaking security rules.
  • For Science Communication, he built the SciCom Wiki, a central hub where videos and podcasts are tagged, organized, and linked to the real science behind them.

• The Truth Detector (Fact-Checking):
  One of the coolest tools Tim built is a "Fact-Checker." When a YouTuber says, "The moon is made of cheese," this tool automatically checks the statement against a trusted database of scientific facts and gives it a "Truth Score." It helps separate good science from fake news.

3. The Challenges: It's Not Just Tech

Tim found that while the technology works, the real hard part is people and rules.

• The "Fear" Factor: In aerospace, engineers are scared to share data because of patents and government regulations. The system has to be secure enough to make them feel safe.
• The "Adoption" Factor: Even if you build a perfect tool, people won't use it unless it's easy and gives them a reason to join. Getting everyone to agree on how to label their data is like getting a thousand people to agree on a single dance move.

4. The Big Picture

Tim's work is a blueprint for the future. He is showing us that we don't need to start from scratch. We can take existing tools (like Wikipedia's engine, but for specific jobs) and combine them with smart AI to build a world where:

• Engineers stop wasting time searching for old solutions.
• The public can easily find and trust science videos.
• Knowledge flows freely, safely, and efficiently.

In short: Tim is building the plumbing and the map for the world's knowledge. He's turning a bunch of disconnected puddles of information into a flowing, navigable river that everyone can use.

Technical Summary

Based on the provided paper, here is a detailed technical summary of the doctoral work by Tim Wittenborg.

1. Problem Statement

The research addresses the critical fragmentation of knowledge in two distinct domains: Science Communication (SciCom) and Aerospace Engineering.

• The Core Issue: Valuable knowledge in these fields is continuously produced but remains siloed across disparate formats, repositories, and organizational boundaries. This fragmentation severely undermines the FAIR Data principles (Findable, Accessible, Interoperable, Reusable).
• Domain-Specific Challenges:
  • Science Communication: High volumes of audiovisual content (videos, podcasts) exist with rich metadata and annotations, but this secondary data is often stored in isolated repositories disconnected from the media itself.
  • Aerospace Engineering: While formalized engineering processes (Knowledge-Based Engineering or KBE) exist, the reuse of heterogeneous knowledge (rules, best practices, results) is hindered by complex data formats, intellectual property concerns, and a culture of reluctance to share documents due to strict governance and regulatory environments.
• The Gap: While FAIR data management tools exist, their adoption is limited by technical heterogeneity, lack of structured support, and legal/social barriers. There is a need for robust Knowledge Infrastructures (KI) that connect people, artifacts, and institutions to generate and share specific knowledge.

2. Methodology

The research employs a mixed-methods approach combining systematic literature reviews (SLRs), requirements elicitation, and action-research-driven software development.

• Workflow Architecture: The core methodology involves developing a modular, data processing pipeline (depicted in Fig. 1 of the paper) capable of handling diverse data types:
  • Input: Scientific publications, datasets, complex models, and audiovisual media (videos/podcasts).
  • Processing: Utilizes Human-in-the-Loop (HITL) Artificial Intelligence. This involves using Large Language Models (LLMs) for information extraction and semantic modeling, validated by human experts to ensure accuracy.
  • Output: Transformation into FAIR knowledge representations (Linked Open Data, Knowledge Graphs) using standards like Wikibase, Wikidata, and the Open Research Knowledge Graph (ORKG).
• Evaluation Strategy:
  • Action Research: Development is iterative, involving domain experts for evaluation.
  • Data Collection: Surveys, interviews, and prototype testing with stakeholders in both aerospace and science communication sectors.
  • Systematic Literature Review (SLR): Used to map current practices and identify gaps, specifically applied to aerospace KBE.

3. Key Contributions & Technical Artifacts

The thesis proposes and implements several modular tools and frameworks to foster knowledge infrastructure:

• SWARM-SLR (Streamlined Workflow Automation for Machine-actionable Systematic Literature Reviews):
  • A modular, requirements-driven workflow formalizing 65 SLR requirements.
  • Implemented as four Jupyter notebooks guiding users from planning to reporting.
  • Integrates reference managers, NLP pipelines, and digital libraries with HITL validation to enable semi-automated, interoperable reviews.
• ExtracTable:
  • A refinement of the SWARM-SLR data extraction workflow.
  • Combines LLMs with human-defined schemas to extract structured data specifically for downstream integration into Knowledge Graphs (KGs).
• SciCom Wiki:
  • An open, Wikibase-based digital library designed for scientific videos and podcasts.
  • Built on stakeholder requirements to improve discovery, annotation, and FAIR metadata capture for non-textual media.
• Aerospace.Wikibase:
  • A specialized instance aimed at creating a common foundation (taxonomy and persistent identifiers) for the highly restricted aerospace domain.
• Computational Fact-Checking Prototype:
  • Pairs LLM-based statement extraction with symbolic verification against curated ground-truth KGs.
  • Generates interpretable scientific-accuracy scores for media content.

4. Preliminary Results

• Aerospace Engineering Analysis: Using SWARM-SLR, the author screened over 1,000 documents and analyzed 164 core articles.
  • Finding: A solid technical base exists (e.g., CPACS, CMDOWS formats), but adoption is fragmented.
  • Conclusion: Solutions must balance open exchange formats with the restrictive governance needs of the industry.
• SciCom Wiki Evaluation: Early testing (surveys, interviews) indicates the tool significantly aids findability and curation. However, broader adoption faces challenges regarding legal compliance, user onboarding, and scalability.
• Fact-Checking: The prototype successfully provides veracity indications but is currently limited by non-FAIR ground-truth representations and extraction inefficiencies.
• General Workflow: The modular framework successfully demonstrates that audiovisual content can be processed similarly to textual publications to achieve FAIR representation.

5. Significance and Future Directions

• Bridging Technical and Societal Gaps: The work demonstrates that while technical solutions (AI workflows, Knowledge Graphs) are promising, the primary barriers to robust Knowledge Infrastructures are often societal and legal (e.g., IP concerns, governance, incentives for collaboration) rather than purely technical.
• Domain Agnosticism: The proposed artifacts (SWARM-SLR, Wikibase instances) are designed to be adaptable. They offer a blueprint for other fragmented domains to move from monolithic systems to a federated ecosystem (similar to the "Wikiverse" concept).
• Human-in-the-Loop Necessity: The research validates that full automation is insufficient for complex domains; human validation remains critical for trust, legal compliance, and handling nuance.
• Future Work: The author identifies the need to move beyond technical problem-solving to address the "incentives against collaboration" and legal barriers. Future iterations will focus on refining the abstraction of AI tools and expanding the federated Linked Open Data (LOD) environment to fully unlock the potential of these domains.

In summary, this thesis provides a concrete roadmap for transforming fragmented knowledge domains into robust, collaborative infrastructures by integrating advanced semantic technologies with human expertise and addressing the specific governance constraints of high-stakes industries.