Using Text Explorer Topic Model Formula to predict Topics for New TextI've spent some time trying to answer this for myself (even "Googled It" 
, with no success, so now turning to the experts.
I have a client who has text data and JMP Pro. He uses his existing data (training & validation sets) to come up with a reasonable Topic analysis model to predict a nominal response. For example, he is using the text input to predict whether a prospective customer might be a good credit risk, based on data that has been accumulated when the customer called a company rep on the phone.
Now he has a model, using topic vectors to predict credit risk. He now wants to input new text data into the existing model and predict credit risk for the new customer. (Make sense?) How do we do that?
I've tried saving the Topic Vector Formula, but that doesn't update calculate new topic scores for added text in new rows.
Can it be done this way? Or would it be better to revert to the larger DTM's instead of topics?
Seems like this should be simple, but I'm not seeing it.
Any/all help is appreciated!
Jerry
Re: Using Text Explorer Topic Model Formula to predict Topics for New TextClarification: Topic Vector Formula actually updates with new documents. However, the results do not match the results from Save Topic Singular Vectors. This seems to be caused by the bug that I referred to above.
Re: Using Text Explorer Topic Model Formula to predict Topics for New TextThere appears to be a bug causing the formula not working in JMP Pro 14.1. The bug was fixed in JMP Pro 13.2 and JMP Pro 14, but it has reappeared in 14.1. I've reported it the Development.
Additionally, two saving options, Save Topic Singular Vectors and Save Topic Vector Formula, yield identical results with the difference being that the former produces individual columns while the latter the vector-valued format in a formula.
Re: Using Text Explorer Topic Model Formula to predict Topics for New TextAgree with @markbailey. I would use the Topic Vectors as inputs into other modeling platforms. Some modeling platforms (like Fit Model), know what to do with vector effects, others do not know what to do with them. If you want to build several models using various techniques and compare them you can use the right-click New Formula Columns -> Vector -> Columns to break out the vectors into individual columns. Then they will play nice with Neural, Partition and the like.
Re: Using Text Explorer Topic Model Formula to predict Topics for New TextRiffing off Mark Bailey, you can find an example of his second idea in Chris and my 2017 discovery talk. That’s a sentiment model, but the workflow is the same as what Mark discussed.
M
Re: Using Text Explorer Topic Model Formula to predict Topics for New TextI have two immediate ideas but I am still thinking...
First, is topic analysis the best choice when the problem is classification? What about using the discriminant analysis in Text Explorer?
Second, what about using Text Explorer to save the DTM (probably using TF-IDF weighting) and using other predictive modeling techniques to select features that have promise for classification?
