Ankura is an implementation of the anchor words topic modeling algorithm described in Arora et al., 2013. My ultimate goal is to use this as a basis to experiment with fine-grained topic models and improving token level topic assignments. Fair warning though: this is research code and is subject to frequent breaking changes and highly experimental code. In general I cannot provide any support.
Note on Fork
You may notice that this repo was forked from the byu-aml-lab. Please note that this fork has diverged in several breaking ways. This was done solely to make my life easier, without any intention to reintegrate with the main repo.
Important Tags and Branches
- master - token level topic assignment experiments
- ankura2 - scalable topic based classification
- acl17 - interactive topic modeling with tandem anchoring
- xref - generates topic-based cross-references