Read more about the project and its goals (in German).

Der Beitrag Virtual Research Librarian: campus | inform reports on the LibraryAI project erschien zuerst auf Prof. Dr. Bela Gipp, University of Göttingen, GippLab.

(1) SPaRC: A Spatial Pathfinding Reasoning Challenge
by Lars Benedikt Kaesberg, Jan Philip Wahle, Terry Ruas, and Bela Gipp

TL;DR: SPaRC is a 2D pathfinding dataset that necessitates multi-step spatial and rule-based reasoning, underscoring a significant gap in spatial understanding between human and AI reasoning models.

Paper: https://aclanthology.org/2025.emnlp-main.526 or https://sparc.gipplab.org

Poster: https://www.slideshare.net/slideshow/emnlp-2025-poster-sparc-a-spatial-pathfinding-reasoning-challenge/284023981

Presentation: Wed. Nov. 5, 14:30-16:00, Hall C

(2) Contrastive Learning Using Graph Embeddings for Domain Adaptation of Language Models in the Process Industry
by Anastasia Zhukova*, Jonas Lührs*, Christian E. Lobmüller, and Bela Gipp (*equal contribution)

TL;DR: We explore the use of knowledge graphs for fine-tuning language models, specifically the text encoder, when the text domain data is low-resource.

Paper: https://aclanthology.org/2025.emnlp-industry.103/

Poster: https://www.slideshare.net/slideshow/contrastive-learning-using-graph-embeddings-for-domain-adaptation-of-language-models-in-the-process-industry/283913368

Presentation: Wed., Nov. 5, 14:30-16:00, Hall C, Industry Poster Session 1

(3) MALLM: Multi-Agent Large Language Models Framework
by Jonas Becker*, Lars Benedikt Kaesberg*, Niklas Bauer, Jan Philip Wahle, Terry Ruas, and Bela Gipp (*equal contribution)

TL;DR: A framework for multi-agent debate to conduct your research experiments. Evaluate agent personas, response generators, discussion paradigms, and decision protocols.

Paper: https://aclanthology.org/2025.emnlp-demos.29/ or https://mallm.gipplab.org

Poster: https://www.slideshare.net/slideshow/mallm-multi-agent-large-language-models-framework-emnlp-2025/284028707

Presentation: Thu., Nov. 6, 16:30-18:00, Hall C

(4) TrojanStego: Your Language Model can Secretly Be A Steganographic Privacy Leaking Agent
by Dominik Meier, Jan Philip Wähle, Paul Röttger, Terry Ruas, Bela Gipp

TL;DR: We explore a novel threat model where LLMs secretly embed information in their output.

Paper: https://aclanthology.org/2025.emnlp-main.1386/

Poster: https://pdfhost.io/v/WsBM3XwhdJ_TrojanStego-Dm-JW-2-1

Presentation: Fri., Nov. 7, 10:30-12:00, Hall C

(5) Re-FRAME the Meeting Summarization SCOPE: Fact-Based Summarization and Personalization via Questions
by Frederic Kirstein, Sonu Kumar, Terry Ruas, Bela Gipp

TL;DR: We introduce FRAME, a fact-centric pipeline that incorporates statement–context tuples, along with SCOPE, a reason-out-loud personalization protocol, and P-MESA, a reader-aligned metric. We reframe summarization as enrichment, cutting hallucination/omission, and improving relevance, goal alignment, and knowledge fit in meeting summaries.

Paper: https://aclanthology.org/2025.findings-emnlp.1094/

Poster: https://de.slideshare.net/slideshow/emnlp25-poster/284036994

Presentation: Fri., Nov. 7, 12:30-13:30, Hall C

Der Beitrag GippLab Presents Five Papers at EMNLP Conference in China erschien zuerst auf Prof. Dr. Bela Gipp, University of Göttingen, GippLab.

The collaborative project LibraryAI is supported by the German Research Foundation (DFG) with more than one million euros for three years, starting in December 2025.

The virtual research assistant will enable users to “interact” with the library via chat: asking questions, summarizing or translating content, and navigating through complex source collections. The special feature is that the assistant not only responds quickly and accurately, but also understands scientific terminology, thanks to customized AI models trained explicitly for the context of libraries and research.

“LibraryAI will not be a generic chatbot, but a reliable companion in everyday research – tailored to the needs of scientific users,” says principal investigator Prof. Dr. Bela Gipp, deputy director of SUB and head of our research group. The AI will be closely integrated with the existing library systems at SUB Göttingen, allowing it to access current catalog data, full texts, and subject-specific resources without inheriting the typical weaknesses of generative AI, such as outdated content or incorrect statements.

The system is based on a modular, open architecture and will be developed so that other scientific libraries can adopt it. “With our project, we aim to make an important contribution to the digital transformation of library services in Germany,” says Gipp.

Listen to AI-Podcast

Report in the Göttinger Tageblatt

Der Beitrag DFG Funds Development of a Virtual Assistant to Facilitate Research in Libraries erschien zuerst auf Prof. Dr. Bela Gipp, University of Göttingen, GippLab.

The public event in Bursfelde Monastery begins at 6:30 p.m. All interested parties are welcome.

Read more (in German).

Der Beitrag Talk on the Ethical Challenges of AI in Bursfelde erschien zuerst auf Prof. Dr. Bela Gipp, University of Göttingen, GippLab.