Reproduce “Keyword Assisted Embedded Topic Model” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Bahareh Harandizadeh, J. Hunter Priniski, Fred Morstatter
The Keyword Assisted Embedded Topic Model (KeyETM) combines the neural Embedded Topic Model (ETM) with user-defined keyword guidance. This approach produces more coherent, diverse, and human-interpretable topics, particularly effective at discovering niche themes in imbalanced datasets, while also enhancing document representations for classification.
Key insights
- Combining the semantic richness of word embeddings with practical keyword-based weak supervision enables the discovery of targeted, interpretable topics.
- Integrating semantically similar words to user-provided seed words into the prior knowledge matrix (via cosine similarity) reduces the burden of exhaustive seed word selection.
- Regularizing both variational (document-topic) and model (topic-word) parameters ensures learned topics align with user guidance and improve document representations for downstream tasks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.