Raphael Sommer - Research

Knowledge Sharing in Online Learning Communities: A Socio-Technical Systematic Review of Member-Level Success Factors Across Community Types

journal article in preparation · based on the 2nd project thesis, DHBW Stuttgart

A systematic literature review (PRISMA 2020) of 39 quality-gated studies on which member-level factors make knowledge sharing in online learning communities succeed. 76 individual factors are consolidated into 27 constructs and mapped onto four socio-technical dimensions (personal, relational, governance, technological) — across educational, professional and interest-based community types.

Top 3 results:

Governance (norms, facilitation, community design) and platform design consistently predict knowledge sharing positively — personal and relational factors show more mixed, context-dependent effects.
To the authors' knowledge the first review differentiating success factors across an explicit community typology; evidence is heavily concentrated on professional communities — educational and interest-based ones are under-researched.
Findings are translated into concrete design guidance for e-learning platforms (e.g. reputation systems, peer mentoring, rubric-based peer feedback).

Towards an LLM-based Information Extraction and Evaluation Pipeline for Scientific Articles in IS Research

bachelor's thesis · DHBW Stuttgart · submitted 05/2026 · grade: 1.5

Following a Design Science Research approach, the thesis designs and evaluates an LLM-based extraction and evaluation pipeline that automatically extracts methodology-related information from information-systems research articles — as a foundation for verifiable academic knowledge repositories. Multiple LLMs are tested against a curated ground-truth dataset and assessed with both deterministic metrics and LLM-as-a-Judge approaches.

Top 3 results:

Frontier LLMs reliably extract methodological passages, while reference extraction is far more heterogeneous — a paper's length and complexity predict difficulty better than the choice of model.
Current LLM-as-a-Judge metrics show ceiling effects and correlate too weakly with ground truth to replace human evaluation.
Automated construction of academic knowledge repositories is economically feasible (approx. $55 per 1,000 papers); five design principles for future pipelines are derived.

From Perception to Fallback Loop: A Structuration-Guided Case Study Analysis of GenAI in Higher Education

integration seminar · DHBW Stuttgart · 07/2026 · with Patoin Ousmane Fad Konfe, Martial Mone & Leander Uka

A comparative case study of how generative-AI governance emerges in higher education when formal rule-making stalls — examined at the heavily regulated DHBW Stuttgart and the largely unregulated Université Aube Nouvelle (Burkina Faso). It draws on thirteen interviews with students, lecturers and decision-makers, analyzed through Giddens' structuration theory across macro, meso and micro levels.

Top 3 results:

Despite opposite regulatory conditions, both institutions converge on the same fallback mechanism: the invigilated in-person exam.
The governance gap is a broken feedback loop between improvised everyday practice and official rule-making — governance patterns emerge rather than being chosen.
Recommendation: shift from GenAI detection toward competence-centered, adaptive enablement — including a transferable analytical framework based on Giddens.

Want to know more — full texts, abstracts or methodology? Feel free to reach out.