Re: Cut off for Topic Loadings in Text Explorer -> LSA SVD -> Rotated SVD ?The loadings are sorted in descending scores. The scores are a distance measure. This is documented in the Text Explorer chapter of the Basic Analysis guide book.
Cut off for Topic Loadings in Text Explorer -> LSA SVD -> Rotated SVD ?In Text Explorer, when we do Latent Sematic Analysis with SVD, then subsequently do the Topic Analysis with the rotated SVD, we are given a list of "Top Loadings by Topic". This loading is clearly the absolute value of the loading (because we can have negative values too). We can see under the Topic Loadings the conditional formatting that show the values of each term for the topic and what's below the cut off for each topic.
My question is what is the metric that JMP is using to decide what a "Top" term is? Is it some difference or distance in the loading scores within a topic, or distance from the running average of the top topics, or...?
Re: Cut off for Topic Loadings in Text Explorer -> LSA SVD -> Rotated SVD ?What is the metric that JMP is using to decide what a "Top" term is?
I asked the same question before. This is the answer that I got (Rephrased).
Once a loading is <5% of the cumulative sum from the full list, then no more loadings are displayed. Alternatively, once the cum percent of a set of loadings is>75% of the total sum of the loadings, then that set of loadings is selected to be displayed. Those terms who meet the cut off are displayed as the top terms in the descending order of absolute values.
Re: Cut off for Topic Loadings in Text Explorer -> LSA SVD -> Rotated SVD ?John Sall also just shared this with me:
5% of the cumulative absolute loadings, up to 20 but at least 2.
