Posterior Probability in the Text Rule Building Node

[WWD]

My question pertains to the last demo within the “Text Rule Builder” section (4.2) within the Text Miner Module.

Within this demo, the instructor looked at the first row within the output table.   The text contained the word “accident” and had a posterior probability of 100%.  The second row contained text and had a posterior probability of 100%.  Rows three and four contained the word “accident” but had posterior probabilities less than 100%.  

What is a difference between these four cases? and how is the posterior probability determined for these four cases?

Thank you.

[Cynthia_sas]

Hi:
Thanks for your question and specifying where in the class the question was. We've sent your question to the instructors who teach the Text Miner class and will post back when they respond.

Thanks, in advance, for your patience,
Cynthia

[TWoodfield · Accepted solution]

From a SAS R&D professional: "The Text Rule Builder node predicts posterior probabilities by computing the posterior probabilities for each target for each rule and then combining them using Naïve Bayes to get an overall posterior. Note that if you use a decision matrix in your EM flow these posterior probabilities are taken into account in deciding which target to select."

I could not confirm this information with any existing external documentation, but I believe this information is dependable.