Last year, SEMANTiCS, leading conference in Semantic Web, Knowledge Graphs (KGs), and Artificial Intelligence (AI), marked its 20th anniversary and I was honoured to serve as Research and Innovation Track Chair. My esteemed co-chairs were Mehwish Alam, from Télécom Paris, Institut Polytechnique de Paris (FR), and Femke Ongenae, from IDLab research group – Ghent University – imec (BE).
The conference was located in the stunning city of Amsterdam at the brink of autumn, where we could still enjoy some sparks of summer but not so unpleasantly hot and humid. Amsterdam is renowned for being a hub of semantic technologies, thanks to the presence of several research groups across the University of Amsterdam (UvA) and VU University Amsterdam (VU), such as the Intelligent Data Engineering Lab (at UVA), Knowledge Representation and Reasoning (at VU), and Business Web and Media (at VU). Hence, organising SEMANTiCS in such a prestigious and blazoned city would only increase the overall quality of the conference.
The conference was indeed a great success. It had an intense and diverse program, which included workshops, tutorials, research paper talks, industry talks, sponsor talks, a poster session, a fishbowl group discussion, plenty of networking opportunities, and 5 keynote and invited speakers. This ensured its 350+ attendees were constantly engaged, leading to a rich and rewarding experience for all.
Another incredible success was its record-breaking number of submissions at the Research and Innovation (R&I) track. It reached a new high of 95, a significant increase from the 54 submissions in 2023. This record can be attributed to several factors, including the attractiveness of Amsterdam, the exceptional work of our publicity chair Blerina Spahiu, and most importantly, the growing community of practitioners involved in SEMANTiCS. The latter is potentially due to the relatively young team organising the conference, hence making it more accessible, and the increased recognition by the scientific community on the value of semantic technologies in AI, thus boosting participation.
Eventually, we accepted 26 papers for presentation at the R&I track, reaching a 27% acceptance rate. You can read the proceedings on IOS Press, here: https://doi.org/10.3233/SSW60
Let’s dive a bit on the main themes of these accepted papers.
SEMANTiCS 2024: A Retrospective
Six main themes emerged from the accepted papers, which are in line with other major venues of the field like ISWC and ESWC. These include Knowledge Engineering with Large Language Models, Embeddings and Machine Learning on Knowledge Graphs, Ontologies and Knowledge Graphs, Linked Data Management, Question Answering and Querying Systems, and Digital Humanities and Cultural Heritage.
The figure below shows the distribution of the 26 accepted papers SEMANTiCS 2024 according to the six themes.
Four submissions on the Knowledge Engineering with Large Language Models explore the increasing role of LLMs in tackling various challenges within the Semantic Web and information processing domains. These studies present novel approaches to improve ontology matching by leveraging graph search to enhance LLM prompt context, automate legal entity extraction from industrial documents using LLMs and semantic models to reduce reliance on manual training data, expand taxonomies through LLM-generated child taxons and alternative labels, and automate metadata enrichment by classifying dataset column headers with LLMs and controlled vocabularies. Ultimately, these works aim to enhance automation, reduce the need for extensive manual effort and domain expertise, and improve the interoperability and accessibility of semantic data through the strategic application of LLMs.
For Embeddings and Machine Learning on Knowledge Graphs, five papers focus on the integration of KGs with various AI techniques. These include enhancing image understanding via KGs, refining entity linking for out-of-KG cases, developing advanced link prediction models, enabling large-scale relational GCNs through sampling, and boosting interpretability in healthcare AI with semantic descriptions. These studies leverage structured KG knowledge to make AI systems perform better, scale effectively, and be more understandable.
Five papers for the Ontology and Knowledge Graph showcase the development of specialised ontologies for assessing the FAIRness of software repositories, standardising smart building data, mapping sustainability reporting standards, representing event information, and modelling deepfake phenomena. Furthermore, they demonstrate methodologies for leveraging these ontologies to automate tasks like data validation, score calculation, knowledge graph enrichment, and inferencing, emphasising the potential of semantic technologies to enhance data interoperability, understanding, and automation in complex domains.
Five papers for Linked Data Management that address the need for robust access control in federated data systems, explore cross-border multilingual search of legislative documents using semantic web technologies, and tackle the issue of funding knowledge graph maintenance through a delayed-answer auction model. Additionally, they propose a method for creating smaller, context-specific graphs to improve the efficiency of knowledge graph exploitation, and introduce a novel window-based streaming partitioning technique for managing increasingly large RDF graphs in distributed environments.
Three papers focus on Question Answering and Querying Systems. Firstly, ORAQL focuses on optimising federated queries by selectively choosing data sources based on overlap and reliability, improving efficiency and accuracy. Secondly, research explores the use of large language models to verbalise knowledge graph question-answering results, aiming to enhance user interaction in dialogue systems and voice assistants. Finally, CoT-Sparql proposes a chain-of-thought prompting approach to improve the translation of natural language questions into SPARQL queries, simplifying knowledge graph access for non-expert users.
Finally, four papers detail advancements in Digital Humanities and Cultural Heritage. The first paper describes a new data model and annotation client designed to capture multi-level semantic annotations of images, addressing the limitations of existing models, particularly for complex objects. The second paper introduces OperaSampo, a Linked Open Data service that provides access to and analysis of historical Finnish opera performance data. The third paper outlines the development of the Medieval Manuscript Data Integration Ontology (MMDIO), which aims to improve the organization and integration of heterogeneous medieval manuscript data across different platforms. Finally, the fourth paper explores the use of faceted search and visualizations to analyze relationships between entities in cultural heritage knowledge graphs, demonstrating how this approach can uncover patterns and facilitate comparative analysis.
The Semantic Web: Future Trajectories
Beside SEMANTiCS, in 2024, I had the privilege of attending ESWC in Crete and ISWC in Baltimore. On several occasions, especially keynote speakers have provided insights about the future of Semantic Web. In their view, Semantics’ technologies and approaches will increasingly feed into AI.
AI and Semantics have a long history together, dating back to the very beginning of AI in 1956. The relationship between these two fields is not new; rather, it’s a reunion of sorts. As shown in the timeline (figure below) of key milestones in Symbolic AI and Semantics, presented by Maria-Esther Vidal at SEMANTiCS 2024, the two fields have always been interconnected.
However, the open question is: in which way Semantics will feed AI?
First, Neural-symbolic AI, trying to combine the strengths of probabilistic AI (e.g., neural networks) and symbolic AI (e.g., semantics technologies). There are several approaches and strategies currently being researched, including the Semantic Web Machine Learning, as shown below, and discussed in the very same keynote speech from Maria-Esther Vidal.
Other fronts include GraphRAG, and agents. GraphRAG is a technology that supports extracting knowledge graphs from unstructured text sources, and then uses this graph to enhance RAG at inference stage. While the potential of this technology is significant, further research is needed to ensure the generation of high-quality knowledge graphs and to precisely identify the relevant graph components to feed it as context at inference time.
Ora Lassila, renowned researchers in Semantic Web and KGs, and author of the Scientific American article pictured in Maria-Esther slide above, gave himself a keynote to ISWC, in Baltimore. He argued that with LLMs, natural and flexible conversational interfaces have become a reality, allowing users to interact with agents in a truly intuitive manner. Furthermore, the integration of curated and audited knowledge graphs provides these agents with trusted, verifiable information, mitigating the risk of LLM hallucinations and ensuring reliable task execution. This convergence of conversational AI and grounded knowledge sources paves the way for agents that can not only understand our requests but also act upon them with accuracy and dependability, effectively bringing the Semantic Web’s vision to life. According to his perspective (see slide from his keynote below), we need i) Knowledge Representation and Reasoning, ii) Planning, and iii) ability to converse with agents, and the first two are already sufficiently developed.
Exciting times ahead!
SEMANTiCS 2025: Shaping the Path Forward
For the time being, I will not venture my own predictions here, as Niels Bohr said, “It is difficult to make predictions, especially about the future.” Instead, I will leave Blerina Spahiu and Mehdi Ali, current Research and Innovation Track chairs of this SEMANTiCS edition, to provide a more compelling overview of their expectations in their upcoming blog post.
On the other hand, I would like to highlight that SEMANTiCS sits at the intersection of AI, Semantic Web, and Natural Language Processing, and that the future Semantic Web trajectories mentioned above are all in its scope.
This year, Vienna will host SEMANTiCS, which is another major hub for Semantic Web’s research due to the presence of many local groups, including the Institute for Data, Process and Knowledge Management, the Institute for Complex Networks from WU Vien, the Research Group Knowledge Engineering at the University of Vienna, and the Semantic Web Company GmbH.
Additionally, SEMANTiCS is notoriously known as a collaborative forum for industry leaders and academics for knowledge exchange, rapid technology implementation, increasing adoption, and shaping the path forward.
With SEMANTiCS 2024’s success, the conference series’ growing momentum, the Semantic Web’s increased popularity in AI, and SEMANTiCS 2025 being held in another prestigious location, it is reasonable to conclude that SEMANTiCS 2025 will be even more successful.
Disclaimer
I wrote this post for SEMANTiCS 2025 blog (read here: https://2025-eu.semantics.cc/page/news?page=2025-04-22).
