Aligned Neural Topic Model (ANTM) for Exploring Evolving Topics: a dynamic neural topic model that uses document embeddings (data2vec) to compute clusters of semantically similar documents at different periods, and aligns document clusters to represent topic evolution.
Added a few changes to allow the user to provide pre-embedded data as well as umap embeddings to skip the contextual embedding step and the aligned dimensionality reduction step, which are both known to take a long time if the dataset is big.
This allows the user to re-use the contextually embedded documents and the alignedumap outputs to speed up the fitting process.
Added a few changes to allow the user to provide pre-embedded data as well as umap embeddings to skip the contextual embedding step and the aligned dimensionality reduction step, which are both known to take a long time if the dataset is big.
This allows the user to re-use the contextually embedded documents and the alignedumap outputs to speed up the fitting process.