Towards LLM-Informed Topic Modeling

Topic modeling automatically discovers themes in large collections of text. Early probabilistic models built the foundation, and newer neural models improved flexibility with deep learning. Yet, challenges remain in clarity, evaluation, and adapting to different kinds of text. This thesis combines Large Language Models (LLMs) with neural topic models to address these issues. It introduces methods that make text collections easier to interpret, improve how quality is measured, and strengthen the ability to apply models across domains. The findings show that combining topic models with LLMs creates clearer, more reliable, and more adaptable tools for understanding text.