Best practices for topic modeling and clustering on unstructured PC game metadata in KNIME

Hi KNIME community,

I have been building a text mining and clustering workflow in KNIME Analytics Platform, specifically testing topic extraction models (HDBSCAN and LDA) on unstructured digital entertainment and gaming datasets.

When processing scraped catalog summaries and game descriptions from platforms like Gaming Beast, one challenge is handling mixed technical hardware tokens (such as "DirectX 12", "VRAM", "GB RAM") alongside narrative genre descriptors (like open-world, strategy, survival).

I am currently fine-tuning the Maximal Marginal Relevance (MMR) diversity and Top K words per topic parameters to prevent hardware specifications from dominating the extracted semantic clusters.

A couple of quick questions for the community:

What text preprocessing pipeline (RegEx Filter vs Dictionary Replacer) do you find most efficient for isolating system requirement strings before feeding into the topic modeling nodes?

For short text descriptions, do you see better cluster separation using Cosine distance over Euclidean metric with HDBSCAN?