Verification of LLM-Based Metadata Extraction from Scientific Publications
Master
Open
Description
Large language models (LLMs) can automatically extract
content-related metadata from scientific publications.
Examples include artifacts, research methods, and research
questions. However, LLMs can hallucinate, so the reliability
of these extractions needs to be assessed. A manually
annotated metadata dataset is already available for evaluation.
This thesis aims to develop a method for extracting metadata
from publications and verifying the extracted information.
Its accuracy will be systematically evaluated against the
existing dataset.
Evaluation of Large Language Models for Process Mining Analyses on Process Modells
Bachelor
Master
Open
Description
The focus is on developing an automatic evaluation pipeline to investigate how well local Large Language Models can handle various Process Mining tasks based on condensed process representations. For this purpose, different task types and difficulty levels are to be defined, for example regarding process structure, variants, or bottlenecks. The thesis encompasses the design of suitable abstracted representations, the development of a proof-of-concept, and the systematic analysis of the performance of local Large Language Models in this application context.
Fine-tuning Local LLMs for Process Models
Master
Available
Description
The focus is on processing natural language process descriptions and the further development of PM4PY.LLM. Particular attention is given to the influence of larger training data and the automated generation of process descriptions.
Fine-tuning Local LLMs for Simulation Parameters
Master
Available
Description
Development of LLM-based methods for extracting simulation parameters for Business Process Simulation.
Connecting Process Mining, Sensor Data and Predictive Maintenance
Bachelor
Master
Available
Description
The main goal is the development of a Python application for machine monitoring through analysis of sensor data. Special attention is given to anomaly detection through time series analysis and the identification of unusual patterns.
A Framework for Data Communication in Process Mining
Bachelor
Master
Available
Description
The focus is on developing a systematic approach for documenting and handling data quality problems in the process mining pipeline. This is particularly relevant since in many process mining projects, data is deleted or manipulated without sufficient documentation. The work includes developing a proof-of-concept as well as systematic analysis of solution approaches through literature research.
Process Mining/Distributed Systems with Focus on Programming
Bachelor
Master
Available
Description
Send me an email with your ideas, or we can develop a suitable topic together.
