topic-modeling-lit

What problem does it solve?

Scientific literature topic modeling turns large abstract collections into understandable themes, helping you perform faster and more reliable literature reviews than manual reading alone.

Core Features & Use Cases

• LDA with coherence-driven topic selection: preprocess abstracts, train LDA across multiple topic counts, and choose the best model using C_v coherence.
• BERTopic clustering for semantic topics: use sentence embeddings plus UMAP + HDBSCAN to discover fine-grained topic clusters from abstracts.
• Dynamic topic modeling and exportable outputs: visualize topic evolution over time (BERTopic topics-over-time), export human-readable topic summaries, and generate interactive PyLDAvis HTML for LDA results.
• Common use case: analyze 2015–2024 abstracts from a field (e.g., “deep learning neural network”) to identify dominant themes and how they shift year over year, with publication-ready topic summaries and interactive visualizations.

Quick Start

Use the topic-modeling-lit skill to take a list of scientific abstract strings, fit an LDA model with coherence-based topic selection, and generate an interactive PyLDAvis HTML visualization for the best model.