Finding the ideal number of topics when using LDA Seq model

[Jacob Celestine]

Hi,

I am trying to use the Gensim LDA Seq model to identify topics that change over time. I have 8 time slices, each having data from a decade. I tried using the NPMI score to identify the  ideal number of topics to use, but I got split results :

1. Two time-slices found 10 topics as the best.
2. Three time-slices found 14 topics as the best.
3. Three time-slices found 17 topics as the best.

In this scenario, how should I proceed? Should I take a look at another metric, or look at the quality of the topics themselves? Or is there another approach that I can take?