Hi, I've been playing around with the excellent python topic modelling package Gensim these last weeks.
On one of my training sets I have run into the issue of LDA converging towards 1 topic. The training set in question is smallish (700 movies, features are bag-of-words of a subset of the terms).
I'm new to LDA and don't understand what this might imply about my data. Any thoughts appreciated.
- UPDATE: A colleague explained that LDA revolves more around the composition of different topics than say LSI does so in some cases it's perfectly normal that one topic is dominant, it's the other terms that carry most of the information.
[link][7 comments]