I am pleased to share with you the programme for the upcoming KGELL Technical Meeting, which will take place in Málaga on 24–25 September.
The meeting will bring together the KGELL Working Groups to review our current progress, discuss the status of project deliverables and upcoming activities, strengthen cross-WG collaboration, and agree on priorities and next steps. The programme will also feature three keynote talks addressing current challenges and opportunities at the intersection of Large Language Models and Knowledge Graphs, offering perspectives from both industry and academia. Further details on the keynote speakers and their talks are provided below.
Regarding logistical arrangements, you should have already received an email from our local organizer, Prof. Maria del Mar Roldán, with the relevant information. I would like to take this opportunity to thank her once again for her support and for kindly hosting us in Málaga.
Here is the schedule:
24 September
09:00–10:00 Keynote Speaker 1 (Mélida López from Kanzo Tech)
10:00–10:30 Coffee break
10:30–11:30 WG Status Updates — Part I: WG1, WG2, WG3
11:30–12:30 WG Status Updates — Part II: WG4, WG5, WG6
12:30–13:30 Deliverables Status — progress, responsibilities, deadlines and critical issues (also discuss future events summer schools)
13:30–14:30 Lunch
14:30–16:30 Parallel WG Working Sessions I — WG1–WG6
16:30–17:00 Coffee break
17:00–18:00 Parallel WG Working Sessions II — WG1–WG6
20:00 Networking Dinner
25 September
09:00–10:00 Keynote Speakers 2 and 3 (Maciej Ribinski from the University of Malaga, Alessio Antonini from Open University)
10:00–11:00 Management Committee Meeting (reserved to MC members)
11:00–11:30 Coffee break
11:30–13:00 Parallel WG Working Sessions III / Cross-WG collaboration
13:00–14:00 Lunch
14:00–14:45 WG Outcomes & Next Steps — short report from each WG
14:45–15:00 Conclusions and Closing
Mélida López
Title: Knowledge Graph Construction from Text: LLM-Based and Semantic NLP Approaches
Abstract:
Recent research on knowledge graph construction from text has progressively shifted from modular information extraction pipelines towards generative approaches based on Large Language Models. Although this shift expands the capacity to process complex documents, questions remain about how linguistic evidence can be transformed into stable formal assertions. A central challenge lies in the contextual and semantic disambiguation of natural-language text and in representing the resulting interpretations within a knowledge graph. This keynote addresses the problem through two complementary approaches applied to Spanish Public Procurement of Innovation documents. One examines the consistency of direct RDF generation across different models and prompting strategies, while the other uses an explicit NLP pipeline to trace the progression from textual mentions to semantic candidates, structured values and RDF statements. Their comparison raises broader questions about the role of ontologies in controlling generative variability, the function of intermediate representations, and the criteria required to evaluate LLM-assisted knowledge graph construction.
Maciej Ribinski
Title: From information extraction to knowledge extraction from scientific literature with LLMs: are we there yet?
Abstract:
In this short talk I will go over the close relationship between automated information extraction and curation of knowledge resources from scientific sources. My perspective is based on hands-on experience in building an LLM-based literature review assistant, so the presentation will include some of the highlights, lessons learned, and open challenges (spoiler alert: we’re not quite there yet).
Alessio Antonini
Title: Modelling as a Tool for Thought a Hypertext Take on Knowledge Graph-LLM Integration
Abstract:
More than a set of technical tools and methods, knowledge graphs sit within the broader tradition of conceptual modelling. Knowledge graphs and their underlying models are respectively extensive and intensive conceptual representations of reality and domains, but more generally of hypotheses, ideas, and visions. From this perspective, knowledge graphs can be more than fact repositories for LLMs, grounding their output into curated resources, rather KGs act as maps to traverse their hypertext-like structures. Guiding “interpretation”, knowledge graphs can provide LLMs with the missing human perspective necessary to produce new ephemeral knowledge artefacts.
