I rarely comment on videos, but I greatly appreciate short, focused introductions on methods before committing hours to diving in further to applications, methods and alternatives. Thank you!
@10:14 its slightly confusing. You mentions that there are two Dirichlet distributions. It seems that what your pointing at ( topic distributions over words ) is just a collection of multinomial distributions and not a distribution over distributions (Dirichlet)
13:11 did you mean to say "the next step is to now look at the 'word' distribution beta corresponding to whatever topic assignment we use.."? Beta is a distribution over words and not topics
I think we're both correct! \beta is often called itself a topic, so it probably would have been best to say "that topic's distribution over words" to avoid this confusion. Thanks for the opportunity to clarify that!
I try to stay away from cutting edge research because it changes so fast. But: 1) we are doing research in this area, and thanks to Corona all of our research talks become RU-vid videos, so we may have something 2) we do have some new topic models videos, including touching on deep learning soon.
@@JordanBoydGraber Thank you. That's great, I'll look out for it. It's difficult to find documentation on newer models apart from the research papers about them and sometimes it is difficult to understand everything reading those...or at least for me as a student/intermediate beginner.😅 Have a great day and please continue doing RU-vid, you've already helped me so much over the years!
Excellent point! These are just made up. There are some automatic approaches to label topics though: medium.com/datadriveninvestor/automatic-topic-labeling-in-2018-history-and-trends-29c128cec17
