~ Victor

I am an Associate Professor (UHD) at the Vrije Universiteit Amsterdam (VU), where I co-lead the User-Centric Data Science group of the Computer Science department. I am also a co-director of the Cultural AI Lab. In my research, I combine (Semantic) Web technologies with Human-Computer Interaction, Knowledge Representation and Information Extraction to tackle research challenges in various domains. These include Cultural Heritage, Digital Humanities and ICT for Development (ICT4D). I am currently involved in the following research projects:

For other and older research projects, see the “research” tab.

Arcuno: a 3D-printable Knowledge Graph construction toolkit

~ Victor ~ Leave a comment

[This post is partially based on Filippos Taprantzis Creative Technologies bachelor thesis and his amazing blog post]

As part of the VU-UTwente collaboration, VU Amsterdam now offers a bachelor program in Creative Technologies. This year, the first batch of students graduated with super nice and creative projects. I was happy to supervise one of these projects, by Filippos Taprantzis. Filippos developed Arcuno, a fully 3D-printable toolkit that can be used to arrange graph-like structures on a table. In particular, we focused on conceptual modeling (e.g. knowledge graphs, ontologies). One of the intentions of ArcUno is to enable a LEGO-building-like experience, which in turn can make the modeling task more intuitive and approachable.

Arcuno was accepted as a demo-paper at the 2026 SEMANTiCS conference in Ghent. In his blog post, Filippos details Arcuno, how it was presented at the conference, and why having 3-D printed components can actually save you when you forget to bring your actual demonstration material to the conference 🙂

From his blog:

“When Victor de Boer, my Graduation Project supervisor, initially proposed the idea of submitting the project to a conference, I could never have predicted that five months later, my friends Stijn, Nicky, Michela, and I would need to pull an all-nighter to remake the entire thing in under seven hours.

Before diving into the story, I would like to briefly describe what ArcUno, the project developed during my Creative Technology thesis, is and what problem it attempts to solve. “

read more at https://ftapr.com/rebuilding-arcuno-overnight

Arcuno github: https://github.com/robo-monk/ArcUno

Share This:

Can AI Understand Buddhist Concepts?

~ Victor ~ Leave a comment

[This post is based on Arjuna Hellingman’s Bachelor AI Thesis]

Religious texts are often rich in metaphor, symbolism, and implicit meaning, making them challenging for modern AI systems. While recent advances in natural language processing have improved concept extraction in domains such as news and scientific literature, much less is known about how these techniques perform on religious texts.

In his Bachelor’s thesis, Arjuna Hellingman investigated whether modern semantic AI models can identify key Buddhist concepts in the Dhammapada, one of the most influential texts in Theravada Buddhism. The study compared a traditional rule-based lexical approach with two embedding-based semantic models, BGE-base and BGE-m3.

To create an evaluation benchmark, three Buddhist monks annotated 120 verses from the Dhammapada using a predefined set of 32 core Buddhist concepts, including suffering, craving, enlightenment, wisdom, and compassion. These annotations were combined into two gold standards: a liberal version that included concepts identified by at least one monk, and a stricter majority version requiring agreement from at least two monks.

The results revealed an interesting trade-off. On the liberal gold standard, the semantic models performed best, identifying a broader range of concepts and achieving higher overall F1 scores. On the stricter majority gold standard, however, the simple rule-based method achieved the highest performance due to its greater precision. Semantic models generally found more relevant concepts but also produced more false positives.

One particularly surprising finding was that the English-focused BGE-base model outperformed the larger multilingual BGE-m3 model on several abstract Buddhist concepts, including craving, wisdom, rebirth, and enlightenment. This suggests that multilingual models do not necessarily have an advantage when working with English translations of religious texts.

Beyond the specific results, the thesis demonstrates the importance of evaluation design in digital humanities research. The choice of gold standard significantly influenced how the models were judged, highlighting that benchmark construction is as important as model selection when applying AI to low-resource cultural and religious domains.

Reference: Arjuna Hellingman. Concept Extraction under Domain Shift: A Comparative Study of Lexical and Semantic Methods in the Dhammapada. Bachelor’s Thesis, Vrije Universiteit Amsterdam, 2026

All data and supplementary materials can be found here: https://zenodo.org/records/20794173

Concept_Extraction_under_Domain_Shift__a_Comparative_Study_of_Lexical_and_Semantic_Methods_in_the_Dhammapada___Download

Share This:

Making Oral WW2 History More Searchable with Knowledge Graphs and AI

~ Victor ~ Leave a comment

[This post is based on Feruza Bakhtiyorova‘s Bachelor AI thesis]

Oral history interviews are rich sources of personal experiences, but they are often difficult to search. A single interview segment may refer to specific wartime locations, events, organizations, or people, while also touching on broader themes such as persecution, migration, or survival. Making these interviews accessible therefore requires high-quality metadata and meaningful concept annotations.

In her Bachelor’s thesis, Feruza Bakhtiyorova investigated how knowledge graphs can help improve the automatic annotation of oral history interviews from the WO2Net project. The existing workflow uses large language models (LLMs) to link interview segments to concepts from the WO2 Thesaurus, a controlled vocabulary of World War II-related concepts. Earlier work had shown that these links were often too broad or failed to capture important specific concepts such as camps, events, or organizations.

To address this problem, the thesis explored a knowledge graph retrieval-augmented generation (KG-RAG) approach. Instead of asking the language model to choose only from existing candidate concepts, the system first retrieves additional concepts from the WO2 Thesaurus and then asks the model to select the concepts best supported by the interview segment.

The study compared three approaches: baseline prompting, refined prompting, and KG-RAG prompting. Five real oral history segments were evaluated by domain experts, who judged both the correctness and specificity of the assigned concepts.

The results showed a nuanced picture. KG-RAG did not outperform the other methods in overall correctness, but it consistently produced more specific concepts. In several cases it successfully recovered important missing concepts such as Birkenau, Reichenbach, Kristallnacht, and Bergen-Belsen, which were unavailable to the prompting-only approaches. At the same time, retrieving more candidate concepts also increased the risk of introducing less relevant suggestions.

The thesis demonstrates that knowledge graphs can complement large language models by providing richer domain knowledge. While retrieval alone does not guarantee more accurate annotations, it can help generate more precise and historically meaningful concept links, improving access to large oral history collections and supporting future research into cultural heritage archives.

The project’s code can be found at https://github.com/ferufera/wo2-segmentation-optimization

Feruza Bakhtiyorova. Improving Concept Specificity in LLM-Based Concept Linking for Oral History Segments Using the WO2 Thesaurus. Bachelor’s Thesis, Vrije Universiteit Amsterdam, 2026.

Feruza_Bakhtiyorova_2740961Bachelor_ThesisDownload

Share This:

Enriching Smart Meter Data with Knowledge Graphs

~ Victor ~ Leave a comment

[This post is based on Gintaré Rutkuté’s Bachelor AI thesis]

As smart meters become increasingly common, large volumes of energy data are being collected in knowledge graphs. However, these datasets often lack important contextual information, such as which building a meter belongs to or where it is physically located. This makes building-level analysis and interpretation difficult.

In her Bachelor’s thesis, Gintarė Rutkutė investigated how a smart meter knowledge graph from the HEDGE-IoT project at Arnhems Buiten could be enriched to support location-aware energy analysis. The original graph contained meter observations, but lacked explicit links between meters, buildings, and locations.

Using semantic web technologies and the SAREF ontology, the thesis added building and meter location information, linked meters to the buildings they measure, and integrated building data from the Kadaster Knowledge Graph. The resulting enriched knowledge graph enabled energy consumption to be analysed at the building level rather than only at the level of individual meters.

The enriched graph was evaluated using eight competency questions developed together with domain experts. These included questions such as which buildings are measured by a meter, which buildings lack meter coverage, and how energy consumption differs between buildings over time. Interactive maps and visualisations showed that the added semantic and spatial context greatly improved the interpretability of the data.

Spatial representation of buildings and their aggregated energy usage.

Power usage figure for two separate buildings on a specific day (details withheld for anonimity)

The thesis demonstrates how knowledge graph enrichment can transform raw IoT measurements into a richer and more useful resource for analysis. By connecting energy data to its physical context, the approach supports more meaningful exploration of building energy use and creates a stronger foundation for future smart energy and energy-sharing applications.

Github: https://github.com/gintare-ru/Knowledge-Graph-Enrichment-for-Building-Oriented-Smart-Meter-Data-Analysis

Reference: Gintarė Rutkutė. Knowledge Graph Enrichment for Building-Oriented Smart Meter Data Analysis. Bachelor’s Thesis, Vrije Universiteit Amsterdam, 2026.

Knowledge-Graph-Enrichment-for-Building-Oriented-Smart-Meter-Data-AnalysisDownload

Share This:

Finding Near Duplicates in Digital Heritage Collections

~ Victor ~ Leave a comment

[This post is based on Senna Lindner’s Information Science Master thesis]

Digital libraries contain millions of documents, scans, newspapers, and books. As these collections continue to grow, so does a less visible problem: near duplicates. These are files that are not exact copies, but are almost identical, for example because the same document has been scanned twice, or because a thesis has been downloaded again with only a different cover page. Detecting such duplicates is difficult, but important for maintaining large digital collections.

Example of near duplicate scans

In his Master’s thesis, Senna Lindner investigated how near duplicate documents can be detected in the collections of the Koninklijke Bibliotheek (KB), the Dutch National Library. The research compared four existing duplicate detection techniques: the text-based methods SimHash and MinHash, and the image-based methods pHash and dHash. Rather than evaluating these methods individually, the study also explored combinations of methods.

The results showed that there is no single best method for all types of documents. For text-rich PDF theses, MinHash alone achieved perfect detection performance. For scanned documents, however, the best results came from combining text and image-based approaches, achieving an F1 score of over 93%. These hybrid approaches were more accurate, although they also required more computing time.

To understand how such a system could be used in practice, interviews were conducted with staff members of the KB. While interviewees acknowledged that near duplicates exist, they emphasized that usability and workflow integration are as important as detection performance. Participants preferred a human-in-the-loop approach, where the system suggests possible duplicates and staff members verify the results before any action is taken. High recall, finding as many duplicates as possible, was generally considered more important than perfect precision.

The thesis demonstrates that combining multiple detection methods can significantly improve the identification of near duplicates in digital heritage collections. At the same time, successful adoption depends not only on algorithms, but also on designing tools that are transparent, efficient, and fit existing archival practices.

The Github link to the code and data used in Senna’s thesis: https://github.com/sennalindner/Near-duplicate-detection

Senna Lindner. Near Duplicate Detection in Digital Libraries. Master’s Thesis, Vrije Universiteit Amsterdam, in collaboration with the Koninklijke Bibliotheek, 2026.

Master_IS_Research_Thesis_Senna_Lindner-7Download

Share This:

Can AI Agents Become the Interface?

~ Victor ~ Leave a comment

[This post is based on Marit Pasterkamp‘s Information Science Master thesis]

Recent advances in AI have led many software companies to add AI features to their products. But what happens when AI is no longer just a feature and becomes the primary interface? In her Master Information Science thesis at VU Amsterdam, Marit Pasterkamp explored this question in the context of marketing automation. Working with Spotler, she designed and evaluated an AI Agent that allows users to create marketing customer journeys through natural language instead of manually building workflows.

Using a user-centered design approach, she developed a functional prototype and tested it with both experienced and inexperienced users. The findings showed that both groups saw value in the AI Agent, but for different reasons. Experienced users appreciated it most for creating complex journeys quickly, while novice users benefited from the lower learning curve and intuitive conversational interface.

Screenshot of the AI Agent prototype showing the two-column editor, with on the left the conversational window and on the right the visual flow canvas.

A key finding was that trust differs between user groups. Experienced users tended to verify AI-generated output before accepting it, while inexperienced users often trusted the AI and instead doubted their own knowledge.

The research concludes that AI Agents can function as effective interfaces for journey creation, but work best in a hybrid approach that combines AI-generated output with opportunities for manual editing and validation. Transparency, usability, and user control remain essential for successful adoption.

Reference

Pasterkamp, M. (2026). Maintaining Usability and Trust: The Effective Design of an AI Agent-driven Interface for Customer Journey Automation in the Marketing Technology Domain. Master’s Thesis, Vrije Universiteit Amsterdam

MaritPasterkamp_2678404_MasterThesis_AIAgent-1Download

Share This:

Audience Poster Award at ICT Open

~ Victor ~ Leave a comment

Really nice news from ICT Open 2026. Our poster on Knowledge-Driven Data Exchange and Validation for Smart Buildings, which shows the TNO-VU collaboration in the HEDGE-IoT project around the Arnhem’s Buiten Electricity Campus won the Audience Poster Award. Xander Wilcke did an amazing job making the poster and presenting it successfully!

posterDownload

Share This:

HUMAIN project secures Horizon Europe funding

~ Victor ~ Leave a comment

We’re happy to share that the project HUMAIN has been successfully funded under Horizon Europe. HUMAIN focuses on the ethical and human-centric adoption of Artificial Intelligence in mission-driven sectors that are vital to society: Culture, Education and Open Science. The project responds to the rapid uptake of AI—from generative tools in the creative industries to AI-supported teaching and scholarly publishing—and addresses the need for data-informed guidelines and regulatory frameworks that align AI innovation with public values and societal missions.

The User Centric Data Science group at VU will contribute to the project by analysing the use and impact of AI in the cultural heritage and creative sector, helping ensure that AI applications support inclusive, responsible, and human-centered practices

Share This:

Matching B Corporations via a Hybrid Information Retrieval Framework

~ Victor ~ Leave a comment

[This post is based on Yitong Tang‘s Master Information Science thesis, conducted in an internship project with 2CoolMonkeys B.V]

In her master’s thesis, Yitong Tang presents the design, implementation, and evaluation of a hybrid information retrieval system that integrates structured filtering with LLMs to enhance partner matching within the B Corporations ecosystem. B Corporations are companies that have undergone verification to meet high standards of social and environmental performance, transparency, and accountability. As value-driven business networks continue to grow, organizations increasingly face the challenge of identifying suitable strategic partners within vast, unstructured datasets.

This thesis introduces a partner-matching system built on a hybrid information retrieval framework. Following a Design Science Research methodology, she iteratively developed and refined the system across four prototypes. The final architecture combines the efficiency of a traditional search engine (Elasticsearch) for structured data filtering with the deep semantic capabilities of an advanced LLM used to interpret unstructured text, such as company descriptions and mission statements.

Screenshot of the recommender system (see demonstration on youtube).

To evaluate the system, she employed a mixed-methods approach that included quantitative performance metrics—precision, recall, and response time—as well as qualitative interviews with B Corp stakeholders. The findings show that the final hybrid model significantly outperforms baseline approaches, offering more accurate, contextually relevant, and well-reasoned recommendations. Its strengths are particularly evident in interpreting complex natural-language queries and incorporating geographical constraints.

Overall, this research contributes a validated and transferable blueprint for developing intelligent partner-matching systems within private or domain-specific knowledge bases. It demonstrates that combining traditional search technologies with modern LLMs provides a powerful approach for transforming raw organizational data into meaningful and actionable partnership opportunities—an approach with broad potential for mission-driven and complex business networks.

The thesis can be found below. A demonstration video can be found at https://www.youtube.com/watch?v=tcowqxA41p4

Master_IS_Research_Thesis__Yitong-TangDownload

Share This:

Capturing Polyvocality of Cultural Heritage Events Through Crowdsourcing

~ Victor ~ Leave a comment

[This post is based on Mohamad Fernanda‘s Master Information Science thesis]

Cultural heritage event annotation often lacks diverse perspectives, resulting in incomplete or biased historical records. Master Information Science student Mohamad Fernanda’s research addresses that gap by examining how crowdsourcing can support polyvocality—bringing a broader range of viewpoints into the annotation process. His central research question is: How can annotations for cultural heritage events be sourced effectively and ethically to achieve polyvocality?

To explore this, he conducted a study in which he:

  1. gathered qualitative survey responses from 22 participants across three groups with different cultural backgrounds: a) native Dutch individuals, b) native Indonesians, and c) people of Dutch-Indonesian heritage;
  2. investigated how Large Language Models can recognize and synthesize polyvocal data.

The findings show that crowdsourcing can successfully capture multiple perspectives, resulting in a richer and more nuanced historical narrative. While LLMs offer promising support for analyzing such data, their use demands careful oversight and ethical consideration. Overall, this study demonstrates that a collaborative, ethically informed approach—combining crowdsourcing with LLM assistance—can help produce more balanced and representative accounts of cultural history.

The figure below shows the results of the coding done by the LLM of choice (Gemini 2.0 Flash). It recognizes five themes in the participants responses on questions about the representation of colonialism in Dutch museums:

Figure from Fernanda (2025)

The thesis, including the exact surveys and prompts used can be found below

Master-IS-Research-Thesis_Mohamad-FernandaDownload

Share This: