Creating Targeted, Interpretable Topic Models with LLM-Generated Text Augmentation