How to interpret Topic Loading and Topic Score in Latent Semantic Analysis, SVD in Text Explorer

[#1]

Hi Community,

When you run SVD on Text Explorer it produces Topic Score and Topic Loading. I understand that these are the Singular vectors at document level and at term level respectively. But how do you actually interpret the values of a particular Topic Score or Topic Loading. There are both negative and positive numbers. Does a high positive number mean something as opposed to a high negative number. For example if  term1 has high positive value on a Topic 1 and high negative value in Topic 2. Does that mean term1 is high on topic 1 and less on Topic 2. Is the interpretation similar when we have document scores against topics?

[#2 · Accepted solution]

The scores are a new representation of the original weighted DTM and the loadings are the correlations of the topics with the original weighted DTM. The larger the magnitude of the scores, the more different documents are from each other and the mean. A large negative loading value shows an important term because it is highly correlated.

Text analysis often refers to terms with the adjectives "attractive" or "repulsive" based on the sign.

[#3]

Thank you very Mark Bailey for taking the time to answer my question. I am still reviewing the results of my data. If I have further data interpretation questions, I will post under this discussion.