Thesis Opportunities
DTALab offers Bachelor’s and Master’s thesis opportunities in data management, machine learning, natural language processing, large language models, explainable AI and data-driven decision support.
Thesis projects are typically connected to ongoing research activities and may involve the development of new methods, experimental evaluation, benchmark design or the application of AI techniques to real-world problems.
Open thesis opportunities
Master’s Thesis
OPEN
Applying Explainability Methods to Portfolio Allocation Tasks
Investigate how input features influence financial forecasting models and how prediction instability propagates to portfolio allocation decisions. The thesis will compare explainability methods across different models to identify the main drivers of their predictions and assess the stability of resulting investment strategies.
Master’s Thesis
OPEN
Studying Influential Training Records in Entity Matching Models
Investigate which training examples most influence the predictions of Entity Matching models by applying techniques such as influence functions and representer points. The analysis will help explain model behaviour, identify problematic or noisy training records, and understand how training data shapes matching decisions.
Master’s Thesis
OPEN
A Multilingual Toolkit of Propositionizers for Retrieval-Augmented Generation
Develop lightweight neural models that decompose documents into self-contained propositions to improve retrieval in RAG systems. The thesis will extend proposition-based retrieval beyond English by training and evaluating language-specific models across both high- and low-resource languages.
Master’s Thesis
OPEN
Learning to Optimize Constrained Paraphrasers for Attacking LLM Watermarkers
Develop adversarial paraphrasing models that test the robustness of LLM watermarking systems while respecting a predefined budget on how much of the original text can be modified. The resulting framework will measure how watermark reliability changes under increasingly strong but constrained paraphrasing attacks.
Master’s Thesis
OPEN
Investigating Data Contamination in Entity Matching Models
Investigate whether publicly available Entity Matching benchmarks have been encountered or memorized during language-model pre-training and whether this affects reported performance. The thesis will evaluate contamination-detection methods and compare potentially exposed benchmarks with newly created or modified records to distinguish memorization from genuine matching capabilities.
Master’s Thesis
OPEN
Automatic Extraction of Fund Governance Data from SEC Filings
Develop a pipeline for automatically retrieving and extracting structured information about trustees, officers, and portfolio managers from SEC fund filings. The thesis will combine document retrieval, LLM-based information extraction, record linking, and normalization to transform heterogeneous textual and tabular content into structured governance data.
Master’s Thesis
OPEN
Reconstruction and Quality Assessment of News Articles from GDELT Web NGrams
Develop and compare methods for reconstructing financial news articles when their original webpages are no longer fully accessible, using fragments available through GDELT Web NGrams. The thesis will explore lexical, graph-based, semantic, and neural reconstruction strategies and systematically evaluate the completeness and quality of the recovered articles.
Interested in a thesis?
If you are interested in one of the available topics, contact prof. Francesco Guerra, indicating the selected topic.
Students may also propose thesis topics related to DTALab research areas. Furthermore, topics can be adapted depending on the student’s background and may evolve according to ongoing research activities.