John Wilkerson on "Changing Focus of the FOMC Through the Financial Crisis" by John Miller and Kathleen McCoy:

This paper applies LDA topic modeling to the FOMC transcripts to investigate changing topic attention across time. The authors are very explicit in terms of describing their methodology, which is appreciated.
In my view the main strength of the paper is not the method but the attention to the details of the method. As a non CS person who is familiar with LDA but has doubts about its value given the subjective nature of topic selection, I would love to have an in-depth conversation about the specific techniques used in this paper.

The results seem intuitively correct, more so than some other results I've seen regarding the changing topics of FOMC meetings.

So overall a solid contribution that demonstrates the researchers' competency.

This paper applies LDA topic modeling to the FOMC transcripts to investigate changing topic attention across time. The authors are very explicit in terms of describing their methodology, which is appreciated.
In my view the main strength of the paper is not the method but the attention to the details of the method. As a non CS person who is familiar with LDA but has doubts about its value given the subjective nature of topic selection, I would love to have an in-depth conversation about the specific techniques used in this paper.

The results seem intuitively correct, more so than some other results I've seen regarding the changing topics of FOMC meetings.

So overall a solid contribution that demonstrates the researchers' competency.

The one thing that would make this paper more accessible is a little more description of the different techniques used to sort through the topics. For example, instead of simply referencing coherence, perhaps explain what coherence is and why it is appropriate for selecting topics. And why use these techniques instead of others that have been proposed, such as perplexity?

Also, and this is sort of a side comment for a non-CS person, is convergence a concern? Do we care if the results are stable? Are they stable?