Parameter tuning for topic model creation #5
Description
Experiment whether this makes sense: the optimization target is the paper coverage of the relevant topics, i.e. all papers should be associated with a relevant topic. Garbage topics are the outlier topic (-1) and the ones selected by the user, typically topics generated from copyright notices.
What we would need to do is to train the topic model with some guestimate parameters so we can identify garbage topics. Then we start optimization and evaluate coverage.