Text mining aims to turn written natural language into structured data that allow for various types of analysis which are hard or impossible on the text itself; machine learning aims to automate the process using a variety of adaptive methods, such as artificial neural nets which learn from training data. Typical goals of text mining are Classification, Sentiment Detection, and other types of Information Extraction, e.g. Named Entity Recognition: identify people, places, organizations; Relation Extraction, e.g. locations of organizations.

Connectionist methods and deep learning in particular have achieved much attention and success recently; these methods tend to work well on large training datasets which require ample computing power. Our institute has recently acquired high performance GPU units which are available for student use in thesis projects. It is highly recommended to use a framework such as PyTorch or Tensorflow/Keras for developing your deep learning application; the changes required to go from CPU to GPU computing will be minimal. This means that you can start developing using your PC or notebook, or the Jupyter notebook server of the department, with a small subset of the training data; when you later transition to the GPU server more performance will mean that larger datasets become feasible.

On text mining e.g.: Minqing Hu, Bing Liu: Mining and summarizing customer reviews. KDD '04: Proceedings of the tenth ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pp. 168-177, ACM, 2004

For a more recent work and overview e.g.: Percha B. Modern Clinical Text Mining: A Guide and Review. Annu Rev Biomed Data Sci. 2021 Jul 20;4:165-187. doi: 10.1146/annurev-biodatasci-030421-030931. Epub 2021 May 26. PMID: 34465177.

Datasets can be found e.g. at huggingface and kaggle.

See also: https://mitloehner.com/lehre/thesis-en.html

How can AR and VR be used to improve exploration of data? Developing new methods for exploring and analyzing data in virtual and augmented reality presents many opportunities and challenges, both in terms of software development and design inspiration. There are various hardware options, starting with Google Cardboard, to more sophisticated and expensive, such as Rift, Quest, and many others. Taking part in this challenge demands programming skills as well as creativity. A basic VR or AR application for exploring a specific type of (open) data will be developed by the student. The use of a platform-independent kit such as A-Frame is essential, as the application will be compared in a small user study to its non-VR version in order to identify advantages and disadvantages of the visualization method implemented. Details will be discussed with supervisor.

Some References:

• Butcher, Peter WS, and Panagiotis D. Ritsos. "Building Immersive Data Visualizations for the Web." Proceedings of International Conference on Cyberworlds (CW'17), Chester, UK. 2017.

• Teo, Theophilus, et al. "Data fragment: Virtual reality for viewing and querying large image sets." Virtual Reality (VR), 2017 IEEE. IEEE, 2017.

• Millais, Patrick, Simon L. Jones, and Ryan Kelly. "Exploring Data in Virtual Reality: Comparisons with 2D Data Visualizations." Extended Abstracts of the 2018 CHI Conference on Human Factors in Computing Systems. ACM, 2018.

• Yu Shu, Yen-Zhang Huang, Shu-Hsuan Chang, and Mu-Yen Chen (2019). Do virtual reality head-mounted displays make a difference? a comparison of presence and self-efficacy between head-mounted displays and desktop computer-facilitated virtual environments. Virtual Reality, 23(4):437-446.

• Korkut, E. H., and Surer, E. (2023). Visualization in virtual reality: a systematic review. Virtual Reality, 27(2), 1447-1480.

See also: https://mitloehner.com/lehre/thesis-en.html

Keywords: causal discovery, smart industrial systems, causal representation, sensor data, semantic representation

Context: Causal Discovery for time series data, such as sensor data is a growing field of research, where causal relations are derived from data, even without expert knowledge. There are various methods available, which aim to find a causal graph that is most fitting to the data. However, different methods produce different causal graphs based on their assumptions. To be able to further investigate discrepancies between various methods, an analysis of the assumptions guiding the discovery of a causal graph, as well as potential preprocessing steps conducted to fit the data into a model are crucial.

Problem: Currently, no method exists to consolidate such discrepancies in causal graphs from multiple sources. 

Goal/expected results of the thesis: This thesis will apply a set of causal discovery algorithms on existing data from an industrial use case, and compare their output. The goal is to identify assumptions and differences in the methods that could potentially explain the differences in generated causal graphs.

Research Questions: What are the main influences of causal discovery graph generation discrepancies?

• How different do causal graphs generated from different discovery methods look?

• What are differing assumptions/hyperparameter settings between applied methods?

• Which causal graph is “more trustworthy”, and why?

Methodology:

• Get familiar with existing causal discovery methods for timeseries data

• Analyse the use case data

• Apply a set of 3 causal discovery methods on the data, to create a causal graph

• Analyse differences between the generated causal graphs

• Analyse differences in assumptions, and metadata, between the applied causal discovery methods

• Discuss potential consolidation approaches

Required Skills:

• Experience with Python/R to implement the causal discovery methods

• Data analysis skills for processing generated outputs and evaluating performance.

• Motivation to work on causal discovery, as a growing field.

• Critical thinking and the ability to analyse discrepancies based on model assumptions

References

[1]      Göbler, K., Windisch, T., Drton, M., Pychynski, T., Roth, M., & Sonntag, S. (2024, March). causalAssembly: Generating Realistic Production Data for Benchmarking Causal Discovery. In Causal Learning and Reasoning (pp. 609-642). PMLR.

[2]      Assaad, C. K., Devijver, E., & Gaussier, E. (2022). Survey and evaluation of causal discovery methods for time series. Journal of Artificial Intelligence Research, 73, 767-819.

[3]      Pearl, J. (2010). An introduction to causal inference. The international journal of biostatistics, 6(2).

Introduction
Large Language Models (LLMs) are increasingly used as automatic evaluators in tasks such as summarization, translation, and dialogue, but most studies have focused on English, leaving open questions about their reliability in multilingual and cross-lingual settings. This gap is especially important given the global use of LLMs across languages with uneven resource availability. At the same time, evaluation protocols have expanded from direct scoring to pairwise comparisons, rubric-based checklists, and reasoning-augmented prompting such as chain-of-thought. Yet, it is unclear how well these methods transfer to other languages or which protocols provide the most consistent and trustworthy judgments.

Research Problem
How consistent and reliable are different evaluation protocols for LLM-as-a-judge in multilingual contexts, and do certain methods (e.g., pairwise comparisons, rubric-guided scoring, chain-of-thought prompting) perform better across languages with varying levels of resources?

Initial references

1. Li, H., Dong, Q., Chen, J., Su, H., Zhou, Y., Ai, Q., Ye, Z., & Liu, Y. (2024). LLMs-as-Judges: A Comprehensive Survey on LLM-based Evaluation Methods. ArXiv, abs/2412.05579.

2. Gu, J., Jiang, X., Shi, Z., Tan, H., Zhai, X., Xu, C., Li, W., Shen, Y., Ma, S., Liu, H., Wang, Y., & Guo, J. (2024). A Survey on LLM-as-a-Judge. ArXiv, abs/2411.15594.

3. Aman Singh Thakur, Kartik Choudhary, Venkat Srinik Ramayapally, Sankaran Vaidyanathan, and Dieuwke Hupkes. 2025. Judging the Judges: Evaluating Alignment and Vulnerabilities in LLMs-as-Judges.

4. Qin, Libo, et al. A survey of multilingual large language models. Patterns 6(1): 101118, 2025.

 

Keywords
LLM, multilingual evaluation, benchmarking, chain-of-thought

• Background: Retrieval-Augmented Generation (RAG) is a critical technique for grounding Large Language Models in factual, up-to-date information. The legal domain represents a high-stakes environment where accuracy, traceability, and grounding in specific legal texts are essential, making it an ideal but challenging application area for RAG.

• Research Question: What are the architectural patterns, evaluation methodologies, and key research challenges for state-of-the-art RAG systems applied to the legal domain?

• Papers

  • Martim, Hudson de. “An Ontology-Driven Graph RAG for Legal Norms: A Structural, Temporal, and Deterministic Approach.” arXiv:2505.00039, 2025.

  • Pipitone, Nicholas, Ghita Houir Alami. “LegalBench-RAG: A Benchmark for Retrieval-Augmented Generation in the Legal Domain.” arXiv:2408.10343, 2024.

  • R. S. M. Wahidur, S. Kim, H. Choi, D. S. Bhatti and H. -N. Lee, "Legal Query RAG," in IEEE Access, vol. 13, pp. 36978-36994, 2025

• Keywords: RAG, Legal Tech, Knowledge Graphs, Ontology, Survey

• Background: Graph-based RAG systems like Graph-CoT and GraphRunner empower LLMs to answer complex questions by traversing knowledge graphs. The model's primary task is to translate a user's natural language question into a correct sequence of traversal steps (operators).

• Research Question: What is a taxonomy for classifying natural language queries based on the graph traversal patterns (e.g., single-node lookup, path finding, intersection) required to answer them?

• Papers: 

  • Jin, Bowen, Chulin Xie, Jiawei Zhang, Kashob Kumar Roy, Yu Zhang, Zheng Li, Ruirui Li, Xianfeng Tang, Suhang Wang, Yu Meng, Jiawei Han. "Graph Chain-of-Thought: Augmenting Large Language Models by Reasoning on Graphs." arXiv:2404.07103, 2024.

  • Kashmira, Savini, Jayanaka L. Dantanarayana, Krisztián Flautner, Lingjia Tang, Jason Mars. "GraphRunner: A Multi-Stage Framework for Efficient and Accurate Graph-Based Retrieval." arXiv:2507.08945, 2025.

  • Dubey, M., Banerjee, D., Abdelkawi, A. & Lehmann, J. "LC-QuAD 2.0: A Large Dataset for Complex Question Answering over Wikidata and DBpedia". International Semantic Web Conference, 2019.

• Keywords: RAG, LLM, Taxonomy, Graph Traversal, Qualitative Analysis

• Background: Generative Retrieval (GR) is an emerging information retrieval paradigm where models directly generate document identifiers. This approach introduces unique failure modes not seen in traditional IR, such as hallucinating document identifier content or pruning relevant identifiers during constrained generation. Understanding these failures is essential for building more robust systems.

• Research Question: What is a classification of the failure modes observed in Generative Retrieval (GR) systems, and what are the most frequent errors observed in practice?

• Papers:

  • Zhang, Peitian, Zheng Liu, Yujia Zhou, Zhicheng Dou, Fangchao Liu, Zhao Cao. "Generative Retrieval via Term Set Generation." arXiv:2305.13859, 2023.

  • Zhang, Zhen, Xinyu Ma, Weiwei Sun, Pengjie Ren, Zhumin Chen, Shuaiqiang Wang, Dawei Yin, Maarten de Rijke, Zhaochun Ren. "Replication and Exploration of Generative Retrieval over Dynamic Corpora." arXiv:2504.17519, 2025.

  • Zeng, Hansi, Chen Luo, Hamed Zamani. "Planning Ahead in Generative Retrieval: Guiding Autoregressive Generation through Simultaneous Decoding." arXiv:2404.14600, 2024.

• Keywords: Generative Retrieval, Error Analysis, Model Robustness, Qualitative Analysis

Keywords: AI transparency, requirements engineering, visual modelling, market analysis, AI system engineering

Context: AI systems have become increasingly complex and difficult to understand, partially due to their nature as “black boxes” [1]. One approach to address this complexity and support AI system engineering is the Boxology Extended Annotation Model (BEAM). The original version of BEAM was introduced in a master thesis [2] and refined in [3]. BEAM enables simplified representations of AI systems through a dataflow-oriented abstraction, enhancing both comprehensibility and transparency. Using BEAM, an AI system can be depicted with various visual elements representing its components, such as data inputs and outputs, machine learning models, processes, and actors, along with the dataflow between them. Additionally, extensions for different perspectives, such as risk assessment or legal and ethical considerations, can be incorporated into the AI system representation. The visual representation can be automatically converted into a machine-readable format, suitable for further processing and analysis.

Problem: A prototype of the BEAM visual notation (see Github repository) exists as a library for draw.io, an open-source diagramming software. However, this prototype has several limitations. As draw.io is a general-purpose diagramming tool, it permits arbitrary drawings that deviate from the BEAM notation syntax, making automatic translation into a machine-readable format challenging. Moreover, support for various perspectives and levels of abstraction is limited, which would be beneficial to meet the needs of different stakeholders, such as engineers, project managers, or ethical experts.

Goal/expected results of the thesis:

• Collection of requirements for the visual tool for the BEAM visual notation

• Comparison of available tools (both open-source and commercial) against these requirements

Research Questions:

• Which techniques are suitable for eliciting requirements for visual modelling software?

• What are the specific requirements for visual modelling software tailored to the BEAM notation?

• Which tools are available that meet these requirements?

Methodology:

• Familiarize yourself with BEAM and AI system representation

• Identify suitable requirements elicitation techniques from literature (see [4] as a starting point)

• Conduct requirements elicitation using an appropriate method, e.g. involving a survey or interviews. Stakeholder groups and requirements regarding AI documentation were already identified in [2], this bachelor thesis shall extend this existing work with a focus on the visual modelling tool

• Research available tools and evaluate their suitability based on the identified requirements

Required Skills:

• Understanding of data science or machine learning workflows

• Interest in requirements engineering and/or visual modelling

References

[1] Königstorfer, F., & Thalmann, S. (2022). AI Documentation: A path to accountability. Journal of      Responsible Technology, 11, 100043. https://doi.org/10.1016/j.jrt.2022.100043.

[2] B. Kollmann, Towards a Workflow-based AI System Documentation, Master’s thesis, WU Wien, Vienna, AT, 2024. URL: https://semantic-systems.org/wp-content/uploads/2024/11/Kollmann.pdf.

[3] Ekaputra, F. J., Prock, A., & Kiesling, E. (2025). Towards Supporting AI System Engineering with an Extended Boxology Notation. Proceedings of the 2nd International Workshop on Knowledge Graphs for Responsible AI (KG-STAR 2025) co-located with the 22nd Extended Semantic Web Conference (ESWC 2025) (Vol. 4018). CEUR Workshop Proceedings.              http://ceur-ws.org/Vol-4018/

[4] Pacheco, C., García, I., & Reyes, M. (2018). Requirements elicitation techniques: a systematic literature review based on the maturity of the techniques. IET Software, 12(4), 365-378. https://doi.org/10.1049/iet-sen.2017.0144.

Abstract:

Post-quantum cryptography aims to develop systems secure against quantum attacks, which threaten cryptographic methods such as RSA and ECC [1]. Although ECC, used in services like eIDAS, is efficient today, both RSA and ECC are vulnerable to quantum algorithms like Shor's, raising concerns about future security [2]. Research on breaking RSA, particularly in Asia, has advanced significantly, and the region is also leading efforts to create quantum-safe alternatives like lattice-based cryptography [3, 4].

In this bachelor thesis, you will provide a systematic overview of the mechanisms behind conventional and post-quantum cryptography (= theoretical part). By means of a bibliometric analysis, you will provide an overview of the literature on postquantum cryptography with a particular focus on the contributing institutions’ places of origins (= empirical part). You will follow cross-disciplinary branches of research and identify current trends on these topics. In the second part of the thesis, you will discuss your findings and reflect on their impact for research and practice. 

Keywords:

post-quantum cryptography, bibliometric analysis

Initial References:

1. Mallouli, Fatma, et al. "A survey on cryptography: comparative study between RSA vs ECC algorithms, and RSA vs El-Gamal algorithms." 2019. 6th IEEE International Conference on Cyber Security and Cloud Computing (CSCloud) 2019 5th IEEE International Conference on Edge Computing and Scalable Cloud (EdgeCom). IEEE, 2019.
2. Bernstein, D. J., & Lange, T. (2017). Post-quantum cryptography. Nature, 549(7671), 188-194.
3. Oder, T., Pöppelmann, T., & Güneysu, T. (2014, June). Beyond ECDSA and RSA: Lattice-based digital signatures on constrained devices. In Proceedings of the 51st Annual Design Automation Conference (pp. 1-6).
4. Dam, D. T., Tran, T. H., Hoang, V. P., Pham, C. K., & Hoang, T. T. (2023). A survey of post-quantum cryptography: Start of a new race. Cryptography, 7(3), 40.

Abstract:

Bias is often associated with negative outcomes, and rightfully so in many contexts. As such, bias may lead to unfair outcomes, reinforcing inequalities and excluding marginalized groups [1]. In the context of developing a normative Semantic Web based on humanistic values, bias can play an effective role by guiding technology to integrate fairness, transparency, and democracy [2]. Instead of being harmful, an ethical bias may be used to create a more accountable Semantic Web, ensuring that humanistic values are embedded in technological decision-making, reflecting the foundations of the Digital Humanism initiative [3].

In the theoretical part of the thesis, you will conduct a literature review on the evolving role of bias in digital technology, and link this to humanistic values focusing specifically on inclusion and democracy. In the empirical part of your thesis, you will develop a conceptual model that shows to which extent Semantic Web (for example, Linked Data, Wikidata, Recommender Systems) may act upon an ethical bias that persists on humanistic values including inclusivity, fairness and democracy. In your conceptual prototype, you discuss features that address this ethical bias through processing algorithms, data selection, and data representation. The aim of this thesis is to demonstrate how ethical bias can create a more inclusive, transparent, and fair Semantic Web, by modeling empirical solutions to the pressing issues related to it.

Keywords:

Ethical Bias, Normative Technology, Semantic Web, Digital Humanism

Preliminary References:

1. Hanna, M., Pantanowitz, L., Jackson, B., Palmer, O., Visweswaran, S., Pantanowitz, J., ... & Rashidi, H. (2024). Ethical and Bias Considerations in Artificial Intelligence (AI)/Machine Learning. Modern Pathology, 100686.
2. Reyero Lobo, P., Daga, E., Alani, H., & Fernandez, M. (2023). Semantic Web technologies and bias in artificial intelligence: A systematic literature
review. Semantic Web, 14(4), 745-770.
3. Werthner, H., Ghezzi, C., Kramer, J., Nida-Rümelin, J., Nuseibeh, B., Prem, E., & Stanger, A. (2024). Introduction to Digital Humanism: A Textbook (p. 637). Springer Nature.

Further Reading: 

S. Tsaneva, S. Vasic, and M. Sabou, “LLM-driven Ontology Evaluation: Verifying Ontology Restrictions with ChatGPT,” in The Semantic Web: ESWC Satellite Events, 2024, 2024.
G. B. Herwanto, F. J. Ekaputra, G. Quirchmayr, and M. A. Tjoa, “Towards a Holistic Privacy Requirements Engineering Process: Insights from a Systematic Literature Review,” IEEE Access, 2024.

Thesis supervision starting February/March 2026! 

Keywords: ontology engineering, human-centric explanations, large language models

Context: Knowledge Engineering (KE) encompasses a variety of activities, including the acquisition of knowledge and its representation through semantic models such as ontologies. Traditionally, KE requires substantial manual effort to define, implement, and validate domain-specific requirements. Moreover, tool support for many KE tasks remains limited, increasing the likelihood of modeling errors, especially when ontology engineers lack advanced KE training or are working with complex logical constraints. Recently, to support ontology engineers, the potential of Large Language Models (LLMs) has been explored in the context of ontology verification, specifically for defect detection, classification, explanation, and correction. While initial studies demonstrate that LLMs can assist with these tasks, further experimentation is necessary to generalize and extend these findings.

Problem: Currently, there is a lack of tools and methods for the automated evaluation of AI-generated explanations within the context of ontology validation. 

Goal/expected results of the thesis: This thesis will investigate how LLMs can be utilised to annotate AI-generated explanations according to value-based requirements.

Research Questions: To what extent can LLMs annotate AI-generated explanations according to value-based requirements?

• How accurately do LLMs evaluate ontology defect explanations?

• Do different LLMs vary in their evaluation performance?

• How consistent are LLM-generated annotations across repeated evaluations of the same input?

Methodology:

• Get familiar with prior experiments on LLMs for ontology defect explanation and the produced explanations dataset.

• Design and implement scripts (e.g., in Python or other suitable languages) to prompt various LLMs for explanation evaluation tasks.

• Perform experiments across different LLMs and analyse the results.

Required Skills:

• Understanding of ontologies, ontology constraints and reasoning (SBWL K2 completed).

• Experience with Python (or other languages that support API access to LLMs).

• Data analysis skills for processing generated outputs and evaluating performance.

References

• Tsaneva, S., Herwanto, G. B., Llugiqi, M., & Sabou, M. Knowledge Engineering with Large Language Models: A Capability Assessment in Ontology Evaluation. https://www.semantic-web-journal.net/system/files/swj3852.pdf

• C.-H. Chiang, H.-y. Lee, Can large language models be an alternative to human evaluations?, in: Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics, 2023. 10.18653/v1/2023.acl-long.870

Keywords: computational social choice, recommender systems, digital democracy

Context: Online discussions are a crucial part of modern democratic deliberation. To reap the benefits of such debates, it needs to be possible to summarize them with statements that best represent the different discussion points. So far, most algorithms simply show majority opinions, which tend to neglect the full spectrum of beliefs. The research field of computational social choice offers algorithms for summaries with better representation guarantees. However, the sheer multitude of comments in online discussions makes it impossible for any single user to express their opinion on all of them. In this thesis project, we will explore whether we can use modern AI technologies to bridge this gap in information to further fair representation also in a digital democracy setting.

Problem: Currently, existing methods for selecting representative statements in online discussions struggle with the highly incomplete information on user opinions.   

Goal/expected results of the thesis: This thesis will investigate how machine learning techniques can best be used to predict user opinions on statements in online discussions.

Research Questions: To what extent can matrix completion methods accurately predict users’ approval of online discussion statements? Can these methods be combined with state-of-the-art algorithms to accurately represent users’ opinions?

Methodology:

• Get familiar with the field of computational social choice and machine learning-based matrix completion methods.

• Design and implement scripts (in Python) to use these completion methods on data from real-world online discussions.

• Perform experiments across different completion methods and analyse the results.

Required Skills:

• Good understanding of data analysis, ideally with python.

• Willingness to learn about mathematical measures of fairness.

References

• Piotr Faliszewski, Piotr Skowron, Arkadii Slinko, and Nimrod Talmon. Multiwinner Voting: A New Challenge for Social Choice Theory. In Ulle Endriss (editor), Trends in Computational Social Choice, chapter 2, pages 27–47. AI Access, 2017. https://archive.illc.uva.nl/COST-IC1205/BookDocs/Chapters/TrendsCOMSOC-02.pdf

Zhaoliang Chen, Shiping Wang. A review on matrix completion for recommender systems. https://doi.org/10.1007/s10115-021-01629-6

Keywords: Computational Social Choice, Fairness, Democracy, Voting, Simulations, Simulations in Python.

Context:

A growing number of novel, digital participation processes allow citizens to express their opinions and directly influence policy decisions on a wide range of topics, from the design of a new park in Vienna [1] to the shape of the new constitution in Chile [2] and Iceland [3]. One major challenge in such digital democracy processes is the fair representation of minority opinions. Computer scientists have, in recent years, developed novel tools and algorithms that can be used to make the different forms of digital participation fairer and more representative.

In this thesis, we will focus on a setting that has not received a lot of attention so far, namely on elections where voters can express which candidates or opinions they approve of and which they disapprove of - a model that captures in particular many large scale deliberation processes, hosted on platforms like Pol.is. Kraiczy et al. (2025) [4] recently proposed fair and proportional voting rules for this setting and showed that they perform well in theory. In this thesis, we will investigate in simulations on real-world and synthetic data whether the proposed rules also work well in practice.

Problem:

Research into voting with approvals and disapprovals is very new and only theoretical, therefore there is a lack of more practical studies and simulations.

Goal/expected results of the thesis:

This thesis will experimentally investigate how fair the outcomes of the different voting rules in this setting are and how much they are affected by small changes in the voting instance.  

Research Questions:

• How fair are the outcomes produced by the newly proposed voting rules?

• How much does the outcome change if we add new candidates?

• To what extent does the probability of approval or disapproval of an added candidate change the outcome?

Methodology:

• Get familiar with the setting and the intuition behind the proposed voting rules

• Develop Python scripts to run the simulations on different real-world and synthetic data-sets.

• Evaluate the results of the simulations.

Required Skills:

• Good understanding of data analysis, ideally with python.

• Willingness to learn about mathematical measures of fairness.

References:

• [1] https://mitgestalten.wien.gv.at/de-DE/projects/miep-gies-park

• [2] https://europeandemocracyhub.epd.eu/wp-content/uploads/2023/12/Case-Study-Chile-FINAL-v2.pdf

• [3] Hélène Landemore, When public participation matters: The 2010–2013 Icelandic constitutional process, International Journal of Constitutional Law, Volume 18, Issue 1, January 2020, Pages 179–205,https://doi.org/10.1093/icon/moaa004

• [4] Sections 1,2,3 of:

Kraiczy, Sonja, Georgios Papasotiropoulos, and Piotr Skowron. "Proportionality in Thumbs Up and Down Voting." arXiv preprint arXiv:2503.01985 (2025).

• Boehmer, Niclas, et al. "Guide to numerical experiments on elections in computational social choice." arXiv preprint arXiv:2402.11765 (2024).

Main idea: Populating the Football-CDF Ontology (into a Knowledge Graphs – KG) with selected datasets and developing an LLM-based Question Answering application on top of the KG.

Background: The applications of AI in sports, particularly in football (soccer), have been growing in the last few years. Example applications include player recruitment, performance monitoring, and player selection [1]. Despite this progress, practical barriers often hinder the effective use of football data for AI applications. To address these issues, the Football Common Data Format (CDF) has been proposed as a community-driven effort [2], and recently, an ontology serialization of it has been introduced [3]. The evaluation of the Football CDF ontology, however, has been limited, and it has not been utilised in user-facing applications.

Research Problem and Questions: Pertinent to the challenge mentioned above, the main research question for this thesis topic is: To what extent can we utilise the Football-CDF ontology for question answering mechanisms in the football domain?

Several (sub-) research questions can be explored as sub-research questions, as follows:

• How to populate the Football-CDF ontology with selected datasets into a KG?

• How to design and develop a GraphRAG-based solution for Football-CDF KG?

• How to evaluate the developed QA system of the Football domain?

Expected Tasks:

• Literature study on the methods for the KG population and GraphRAG-based QA

• KG population

• QA application development & evaluation

Prior-Knowledge and Skills:

• Understanding of ontologies (skill from course K2/K3 in the SBWL KM)

• Proficiency in at least one programming language (Java or Python preferred)

• The selected student would have a passion for the domain. Please contact me directly if you are really interested in working on this topic.

References:

[1] Decroos, T., Bransen, L., Van Haaren, J., & Davis, J. (2019, July). Actions speak louder than goals: Valuing player actions in soccer. In Proceedings of the 25th ACM SIGKDD international conference on knowledge discovery & data mining. https://dl.acm.org/doi/10.1145/3292500.3330758

[2] G. Anzer, K. Arnsmeyer, P. Bauer, J. Bekkers, U. Brefeld, J. Davis, N. Evans, M. Kempe, S. J. Robertson, J. W. Smith, J. V. Haaren, Common Data Format (CDF): A Standardized Format for Match-Data in Football, Technical Report arXiv:2505.15820, 2025. URL: http://arxiv.org/abs/2505.15820.

[3] FJ Ekaputra, G Käfer, M Kempe. An Ontology for the Common Data Format on Football Match Data. Proceedings of the ISWC 2025 Posters, Demos, and Industry Tracks (accepted for publication). URL: http://bit.ly/4mIZLXj

Keywords: Question Answering, KG population, Sport Science, Football, Large Language Model

Main idea: There is no established data format for election data. When countries publish election results, this is typically done in an ad-hoc fashion. This thesis should explore what requirements an election data format should have and how different types of election information should be recorded.

This question can be studied more broadly from the perspective of collective decision making, that is, group decisions also outside of a political context. For example, the Preflib.org website stores a large amount of preference data in various formats, which can be further used to test collective decision mechanisms. Apart from election data, there are data sets related to blockchain decisions, sport competitions, Kidney donations, and many other topics.

As a concrete first step, you should collect 10 data sets as diverse as possible and record their similarities and differences. The main task is then to propose a data model (as an ontology or relational database schema) to capture the broad range of elements/components of the election data. Furthermore, mapping between the collected datasets to the proposed data model needs to be established to show the feasibility of the proposed data model as part of the evaluation.

Example election and preference data:

• Nationalratswahl 2024: https://www.data.gv.at/datasets/e40e3b00-1a98-4338-acb7-42547e6fee55?locale=de

• Spotify: https://preflib.github.io/PrefLib-Jekyll/dataset/#00048

• Polkadot blockchain: https://preflib.github.io/PrefLib-Jekyll/dataset/#00060

• Bundestagswahl 2025: https://www.bundeswahlleiterin.de/bundestagswahlen/2025/ergebnisse/opendata.html

Possible Research Questions:

• What are the key elements of election data in the context of collective decision results?

• How to design a data model for preference data and collective decision results?

• In which way is the proposed data model a generalization/improvement over the (rather restrictive) Preflib.org file formats (see https://preflib.github.io/PrefLib-Jekyll/format).

Expected Tasks:

• Literature study on related topics.

• Technical

  • Collect (at least) 10 datasets of election data. Note that the diversity of these datasets is the most critical aspect.

  • Gain an understanding about these datasets and their underlying elements.

  • Record their similarities and differences. Analyze and categorize the components of the datasets.

  • Data model development (e.g., ontology (preferred), JSON Schema, or SQL schema)

Prior Knowledge and Skills:

• Data model building (e.g., ontology engineering, SQL schema design)

• Data analysis and evaluation skills.

References:

• Ontology development: Noy, Natalya F., and Deborah L. McGuinness. "Ontology development 101: A guide to creating your first ontology." (2001).

• Mattei, Nicholas, and Toby Walsh. "A preflib.org retrospective: Lessons learned and new directions." Trends in Computational Social Choice (2017): 289-309. https://archive.illc.uva.nl/COST-IC1205/BookDocs/TrendsCOMSOC.pdf

Keywords: Election data, data model development

This topic is about automating a reproducibility study:

In 2020 we published a study about Semantic Web research progress in the past 2 years:

Sabrina Kirrane, Marta Sabou, Javier D. Fernández, Francesco Osborne, Cécile Robin, Paul Buitelaar, Enrico Motta, and Axel Polleres. A decade of semantic web research through the lenses of a mixed methods approach. Semantic Web -- Interoperability, Usability, Applicability (SWJ), 11(6):979--1005, October 2020. [ DOI | http ]

One of the challenges in this study was to obtain and analyse fulltexts from Semantic Web conferences and venues and analyze them at scale with (back then) state of the art methods for topic analysis and clustering and other mixed methods and tools.

The goal of this thesis is to try to redo this study 5 years later, and devise a pipeleine to do such analyses for different fields and conferences and for different time frames. AI use is *allowed* in this exercise (vibe coding, text summarization via prompting, etc.) but the main goal is to compare the results with the original paper and approach, document your tool usage, and come up with a reusable pipeline.

Background

Knowledge graphs (KGs) are nowadays the main structured data representation model on the web, representing interconnected knowledge of different domains. There are several methods to model a KG. For instance, they can be extracted from semi-structured web data, like DBpedia, or edited collaboratively by a community, like Wikidata. Since there is no perfect method and knowledge about the world is constantly changing, regular updates in the KGs are required.
Knowledge graph refinement is the process of improving the quality and accuracy of a knowledge graph by adding, modifying, or deleting entities, relationships, or attributes based on new information or corrections. This process is crucial for ensuring that a knowledge graph reflects the current state of knowledge in a particular domain and that it can be used effectively for applications such as search, recommendation, and decision-making.
Wikidata has different constraint mechanisms to identify possible inconsistent data, however, it relies exclusively on its user community to fix inconsistencies.
Overall, knowledge graph refinement is an important and ongoing process that is essential for ensuring that knowledge graphs remain up-to-date, accurate, and useful for a range of applications. As new information becomes available and our understanding of the world evolves, it will be necessary to continue refining and improving knowledge graphs to ensure that they reflect the current state of knowledge in a particular domain.

The goal of the thesis

The goal of this thesis is to extend an already existent dataset of Wikidata historical repairs by including the user behind the repair and to analyze the user behavior. The student would have to: (1) work with the dataset and extract the users; (2) analyze the results towards the role of humans and bots, the specificities of different constraint types, and the domain knowledge.

Requirements

Pro-activity and self-organization. Programming skills.

Initial references

● To learn about RDF KGs: HOGAN, Aidan et al. Knowledge graphs. ACM Computing Surveys (CSUR), v. 54, n. 4, p. 1-37, 2021.
● To learn about Wikidata property constraints: Ferranti, N., De Souza, J. F., Ahmetaj, S., & Polleres, A. (2024). Formalizing and validating Wikidata’s property constraints using SHACL and SPARQL. Semantic Web, 15(6), 2333-2380.
● Data quality in Wikidata: Shenoy, K., Ilievski, F., Garijo, D., Schwabe, D., & Szekely, P. (2021). A Study of the Quality of Wikidata. arXiv preprint arXiv:2107.00156.

Background

Knowledge graphs (KGs) are nowadays the main structured data representation model on the web, representing interconnected knowledge of different domains. There are several methods to create a KG. For instance, they can be extracted from semi-structured web data, like DBpedia, or edited collaboratively by a community, like Wikidata. Since there is no perfect method and knowledge about the world is constantly changing, regular updates in the KGs are required.
Knowledge graph refinement is the process of improving the quality and accuracy of a knowledge graph by adding, modifying, or deleting entities, relationships, or attributes based on new information or corrections. This process is crucial for ensuring that a knowledge graph reflects the current state of knowledge in a particular domain and that it can be used effectively for applications such as search, recommendation, and decision-making.
Wikidata has different constraint mechanisms to identify possible inconsistent data; for instance, one property constraint can be added to describe that the property gender can only be used with the values male or female. As knowledge changes over time, this constraint might include non-binary and other types of genders.
Entity embeddings are approaches that take concepts and represent them in a vector space, respecting their meaning. Therefore, for instance, “male” should be more similar to “female” than to “airplane”.

The goal of the thesis

The goal of this thesis is to analyze historical changes in constraint definitions and check whether already existing embedding models can capture the semantic similarity between those concepts.
Steps would consist of:
● Take historical constraint changes (we already have them for 2019, 2023);
● Possibly extend the historical constraint dataset (to include 2024)
● Use different types of embeddings (graph, sentence, etc) and map the selected entities to this space;
● Assess what similarity score these embeddings would give for the already executed historical changes, and evaluate how good they would be in helping the refinement process.

Requirements

Pro-activity and self-organization. Programming skills.

Initial references

● To learn about RDF KGs: HOGAN, Aidan et al. Knowledge graphs. ACM Computing Surveys (CSUR), v. 54, n. 4, p. 1-37, 2021.
● To learn about Wikidata property constraints: Ferranti, N., De Souza, J. F., Ahmetaj, S., & Polleres, A. (2024). Formalizing and validating Wikidata’s property constraints using SHACL and SPARQL. Semantic Web, 15(6), 2333-2380.
● Data quality in Wikidata: Shenoy, K., Ilievski, F., Garijo, D., Schwabe, D., & Szekely, P. (2021). A Study of the Quality of Wikidata. arXiv preprint arXiv:2107.00156.

Travelling by train across Europe often involves navigating complex information systems. Different countries have different rail systems, and crossing borders can create challenges in the passenger experience. This is where innovative aggregator platforms come into play—such as Trainline, Omio , and even Uber —which have successfully integrated train information and ticketing services.

At the heart of these systems lies timetable data, which provides real-time departures andarrivals for trains across Europe. However, combining this data fromvarious sourcesis a challenging task due to several factors:

• Information retrieval: Accessing accurate and up-to-date timetable information
from multiple operators.
• Data quality control: Ensuring consistency and reliability in the data.
• Regular updates: Keeping the timetable data current to reflect any changes or
disruptions.
• Disruptions in service : Handling delays, cancellations, or track alterations.

These challenges can lead to an unpleasant passenger experience, causing travelers to opt for alternativemodes of transportation. We are seeking students with a passion for railways and cross-border train travel who want to "get their hands dirty" with GTFS (General Transit Feed Specification), one of the many data standards used in scheduling.

Example of the thesis can be developed like:

Connecting real time timetable updates with static timetable data.
This topic consists of an analysis of real time timetable data formats (GTFS-RT, SIRI, proprietary formats) andan implementation of an integrating adapter that allows tomap the gathered data onto static timetable information.

Integrating open timetable data
The EU MMTIS regulation obligated all member states to provide rail timetables on national access points. The data covers the rail connections in the country and sometimes (parts of) internationalconnections. This leads to duplicated trips thatmight overlap completely, partially or not at all.
The aim of this topic would be to analyse the different data variants, develop and test different methods tomatch the broken up journeys. This can involve classical heuristics or machine learning. Test datasets will be provided. The results will be utilized by OpenTimetable.eu

Meet Your Supervisor: ShahromSohi (shahrom.hosseinisohi@pv.oebb.at)

Shahrom Sohi, a transport engineer and digital transportation enthusiast working with ÖBB, will be your main point of contact throughout this research project. This thesis offers a unique opportunity to collaborate with ÖBB and other European Mobility Digital players, contributing to the development of more efficient and user-friendly rail travel experiences across Europe.

References:

van Overhagen, L. (2021) ‘A design vision towards seamless European train journeys: Making the train the default option to travel within Europe’. Available at:
https://repository.tudelft.nl/islandora/object/uuid%3A01a0e501-2e1a-469d-b1c3-03df7abae737 (Accessed: 13 May 2024).

CER Ticketing Roadmap (no date). Available at: https://www.cer.be/cer-eu-projects-initiatives/cer-ticketing-roadmap (Accessed: 13 May 2024).

Railways, E.U.A. for (2024) Analysis of distribution rules in TAP, OSDM, and recent competition cases | European Union Agency for Railways. Available at:
https://www.era.europa.eu/content/analysis-distribution-rules-tap-osdm-and-recent-competition-cases 

The European railway domain increasingly depends on cross-border operations, where train drivers and infrastructure controllers often speak different national languages. The Translate4Rail (https://translate4rail.eu/) project has shown that providing a standardized set of predefined messages and a translation tool can mitigate miscommunication in normal and emergency situations.

However, challenges remain in (1) ensuring safety and flexibility in exceptional circumstances, (2) extending the message set to cover rare but critical cases, and (3) adapting the system to real operational conditions (e.g. latency, ambiguity, regional dialects).

This thesis will:

1. Analyze the limitations of the current Translate4Rail prototype and associated language-tool approaches, focusing on weak points such as message coverage, ambiguity, error handling, latency, and safety assurance.
2. Propose improvements or extensions, e.g. via dynamicmessage composition, error correction, fallback strategies, or probabilistic translationmodels using LLMs, to supportmore robust driver–controller exchanges even in unforeseen scenarios.
3. Implement a testbed or simulation framework (or small-scale pilot) to validate the enhancements, measuringmetrics such as misunderstanding rate, response delay, coverage of scenario space, and safetymargins.
4. Evaluate interoperability and safety aspects, possibly in collaboration with rail actors or infrastructuremanagers, to assess feasibility in real corridor settings.

Supervisor contact:
Meet Your Supervisor: Shahrom Sohi (shahrom.hosseinisohi@pv.oebb.at)
Shahrom Sohi, a transport engineer and digital transportation enthusiast working with ÖBB,will be yourmainpoint of contact (feel free to contact him anytime for questions!). This thesis offers you a chance to work at the intersection of rail operations, safety, and AI/linguistic tools contributing to more seamless language interoperability in European rail.

References:

Atanasov, I., Pencheva, E., & Vatakov, V. (2023). An Approach to Designing Critical Railway Voice Communication. Electronics, 12(6), 1406. https://doi.org/10.3390/electronics12061406

Rosberg, T., Thorslund, B. Radio communication-basedmethod for analysis of train driving in an ERTMS signaling environment. Eur. Transp. Res. Rev. 14, 18 (2022). https://doi.org/10.1186/s12544-022-00542-5

The European Union Agency for Railways (ERA) is building Knowledge Graphs (KGs) such as the Railway Infrastructure Register (RINF KG), which describe the European rail network’s assets, topology, and compliance attributes. In parallel, structured accident and incident data (e.g., ERAIL reports, NIB reports) are increasingly being digitized. Yet, these two domains infrastructure and safety events remain largely disconnected.

This thesis will: 

1. Map accident data models to infrastructure entities (e.g., linking derailments totrack sections, collisions tooperationalpoints, oraccidents at level crossings to specic infrastructure features).
2. Build a semanticlayer that connects unstructured accident reports (using NLP/NER extraction pipelines) to the ERA infrastructure KG, using ontologies and vocabularies from ERA and the SemanticWeb community.
3. Develop a prototype Knowledge Graph integration, aligning accident entities (time, location, type, cause) with RINF infrastructure elements, enabling cross-querying (e.g., “Which track segments have had repeated accidents of type X in the past 10 years?”).
4. Evaluate potential applications in safety monitoring, predictive risk analysis, and regulatory reporting, in collaboration with ERA datasets.

Supervisor contact:

Meet Your Supervisor: Shahrom Sohi (shahrom.hosseinisohi@pv.oebb.at)
Shahrom Sohi, a transport engineer and digital transportation enthusiast working with ÖBB, will be your main point of contact throughout this project (feel free to contact him any time for questions!). The thesis provides a unique opportunity to contribute to the next generation of European railway safety intelligence, aligning data from infrastructure and accident domains.

References:

Rail Accident Investigations
https://www.era.europa.eu/domains/accident-incident/rail-accident-investigation_en

RINF register of infrastructure
https://data-interop.era.europa.eu/

Intermodal transport systems, which incorporate multiple modes of transportation within a single journey, are essential for sustainable urban mobility (Riley et al., 2010). Among such systems, Park and Ride (P&R) facilities reduce car congestion, improving public transport accessibility, and enhancing overall travel efficiency. These facilities, strategically positioned near railway stations or bus stops, serve as key nodes for commuters transitioning from private vehicles to public transport (Litman, 2011; Pitsiava-Latinopoulou & Iordanopoulos, 2012). ÖBB INFRA operates numerous P&R stations across Austria, facilitating seamlessmultimodal trips.

ResearchObjectives:

Usage Patterns: How do commuters utilize P&R facilities in Austria?
Accessibility and Travel Behavior: What are the key factors influencingmode choice at these locations?
Forecasting Commuter Flows: How can mobility data be leveraged to predict where people travel after using P&R facilities?
Optimization Strategies: How can ÖBB INFRA enhance the efficiency of P&R facilities based onmobility insights?

This research will could employ a mixed-methods approach combining:
Quantitative Data Analysis: Processing and visualizing mobility datasets from ÖBB INFRA to identify peak usage times, parking turnover rates, and travel flows.
Survey Data Collection: Conducting on-site commuter surveys to understand behavioral factors affecting P&R usage.
GIS & Spatial Analysis: Mapping station accessibility and evaluating the relationship between facility location and utilization rates.
Predictive Modeling: Utilizing historical data to develop forecasting models for commuter flows and demand prediction.
By integrating mobility data analytics with user insights, this thesis will contribute to improving ÖBBINFRA’s intermodal strategies. The findingswill support evidence-based decision-making for future infrastructure planning, ensuring more sustainable and efficient mobility solutions in Austria.

Meet Your Supervisor: ShahromSohi (shahrom.hosseinisohi@pv.oebb.at)
Shahrom Sohi, a transport engineer and digital transportation enthusiast working with ÖBB, will be your main point of contact throughout this research project. This thesis offers a unique opportunity to collaborate withÖBB and other European Mobility Digital players, contributing to the development of more efficient and user-friendly rail travel experiences across Europe.

References:

Sohi, S., Wutz, G., Hrivnák, R., Reiter, F., Pichler, D., Anjomshoaa, A. and Polleres, A., 2025. Enhancing rail transit accessibility: a data-centric approach to Park and Ride. Transportation Research Procedia, 86, pp.405-412. Sohi, S., Wutz, G., Hrivnák, R., Reiter, F., Pichler, D., Anjomshoaa, A. and Polleres, A., 2025. Enhancing rail transit accessibility: a data-centric approach to Park and Ride. Transportation Research Procedia, 86, pp.405-412.
https://www.sciencedirect.com/science/article/pii/S2352146525002984

Riley, P. etal. (2010) IntermodalPassengerTransport inEuropen passengerintermodality from A to Z A TO Z the european forum on intermodal passenger travel. Available at: https://www.academia.edu/5074766/P_Intermodal_Passenger_Transport_in_Europe_PASSENGER_INTERMODALITY_FROM_A_TO_Z_the_european_forum_on_intermodal_passenger_travel_Link_is_funded_by_the_European_Commissions_Directorate_General_for_Mobility_and_Transport_DG_MOVE

Litman, T. (2007)‘Evaluating rail transitbenefits: Acomment’, TransportPolicy, 14(1), pp. 94–97. Available at: https://doi.org/10.1016/j.tranpol.2006.09.003.

Pitsiava-Latinopoulou, M. and Iordanopoulos, P. (2012) ‘Intermodal Passengers Terminals: Design Standards forBetter Level of Service’, Procedia - SocialandBehavioral Sciences, 48, pp. 3297–3306. Available at: https://doi.org/10.1016/j.sbspro.2012.06.1295.

Keywords:
1. Large LanguageModels (LLMs)
2. Web scrapping
3. Pattern recognition
4. B2B Sales
5. Lead Generation

Abstract:
This thesis topic is a highly relevant and practically applicable tuning of an LLM: Training an LLM to Generate Sales Leads from Media News Through mediamonitoring, it is possible to generate sales leads. This works very wellmanually but is time-consuming and doesn't scale. We can provide a set of media articles and, from this, themanually identified subset of sales leads. An LLMshould be trained to recognize a pattern from this so that it can then independently filter the sales leads relevant to the RCG from the media articles."

The challenge is to teach the LLM the specific context to understand through pattern recognition, so that it will be able to select news articles that represent a sales lead from other news articles.

Example: Company A is a prospective target customer. On a given day there are a X number of media articles published with company A’s name in it. One of them states that company A won a new contract and will because of this soon start exports to a new country. These transport streams provide a business opportunity for Rail Cargo Group; hence we classify it as a sales lead. The trained LLM should be equipped with such a contextual understanding that it can
(pre-)select this one article from the entire set of X articles. 

A functioning model would be an additional source to continuously feed the B2B sales funnel with new sales leads.

You will be provided by Rail Cargo Group with historical data on news articles, sales leads selected from it, as well as on a limited subset the conversion rates of these sales leads. In addition, elaborated list of key words is available, and you will receive close coaching and input by Rail Cargo Group subject matter experts.

V. Kumar et al. (2024). “AI-poweredmarketing:What, where, and how?” Journal: IndustrialMarketing Management (Elsevier) https://www.sciencedirect.com/science/article/pii/S0268401224000318

D. Herhausen et al. (2025). “From words to insights: Text analysis in business research” Journal: Journal of Business Research
https://www.sciencedirect.com/science/article/pii/S0148296325003145