How To Build a SharePoint Premium Model (with screenshots)
Tiffany Songvilay
MVP | Modern Workplace Guru | Copilot for M365 | SharePoint Premium | Intranet Migrations and Modernizations | Power Platform | Viva Suite | LGBTQ+ | @tiffsongvilay
This post picks up after the Content Center is created. If you need to know how to set up Syntex and create a Content Center, then this article does a fantastic job of taking you through the steps on how to do that:
If you are a Microsoft Partner and you’d like to follow along, you can create a demo tenant on cdx.transform.microsoft.com. Syntex is included in the Microsoft Viva environment. That 1-year environment build is eligible for extension.
To find your Content Center in Microsoft’s demo tenant, go to the SharePoint Admin Center and look at Active Sites. Filter by Template (Content Center) and go to Operational Content Center.
How to Create an Unstructured Document Processing Model
Formerly known as document understanding model, this model is used for unstructured document processing. For more information on this model type, click this link. Overview of unstructured document processing in Microsoft Syntex - Microsoft Syntex | Microsoft Learn
1.??????From the Content Center home page, click on Models at the top then Create a Model.
2.??????Click Train a custom model – Click Next.
3.??????Give it a name (this will be the name of the content type) and notice the Advanced Settings. This is where you can decide to have Syntex create a new content type (using the name you just gave it) or you can select an existing content type. You can also choose the Sensitivity and Retention Labels you want assigned to this content type (more information on how to create those in another blog post).
4.??????Click Create.
Train the model to recognize the Content Type
1.??????Click Add files.
2.??????Upload – Files.
3.??????Select five (5) files that are the correct content type and one (1) that is NOT the correct content type and click Add.
4.??????Click Train classifier.
A preview window will show the first document and at the top of the preview pane is the question, “Is this file an example of Contract Change (that’s the name of the content type I used in this example)?”
5.??????Click Yes for the five (5) files that are the correct content type and No for the one (1) file that is NOT the correct content type.
6.??????Click Train at the top.
7.??????Under Explanations, click New – Blank.
8.??????Name the explanation and choose Phrase list from the “Explanation type” drop down. Type the phrase you want the model to look for in the document that identifies it as the correct content type. This is most likely a heading or a title or a group of words that identify the document.
9.??????Click Save and train. Notice the Positive and Negative Labels and the Evaluation column that confirms that it was able to positively identify all the documents.
Test the Model
1.??????Click Test at the top.
2.??????Click Add example files.
3.??????Select five (5) new files that are the correct content type and one (1) new file that is NOT the correct content type and click Add.
Notice the Positive and Negative Labels. If you are satisfied with the results, click Exit Training. If more training is necessary to get the correct results, go back to Train and add additional Explanations.
领英推荐
Create and train extractors
1.??????Click Create extractor button.
2.??????Name the extractor (this will be the new column name). Notice the Advanced Settings. This is where you can decide to have Syntex create a new column (using the name you just gave it) or you can select an existing column name.
3.??????Click Create.
You’ll see all the files you’ve uploaded for training and testing on the left side. In the preview window on the right, highlight the piece of metadata that you want to extract from the file. If this file does not contain the information you are trying to extract, then click No label.
4.??????Click Next file and either highlight the metadata in all the documents or click No label.
5.??????Click Save.
6.??????Click Train at the top and click New-Blank under Explanations.
7.??????Type the Name of the piece of metadata (you can reuse the name of the column) and select Phrase list under “Explanation type.”
8.??????Type the phrase that appears before the piece of metadata you are extracting.
9.??????Click Save and train.
Notice the Prediction and Evaluation Columns that let you know if the model found a Match or a Mismatch. Even the document with No label can be a Match which confirms that piece of metadata does not exist in the document using your Explanation.
To improve accuracy, continue adding Explanations until you achieve 100% accuracy (all Evaluations should return with the word “Match”).
10.??Click Test at the top and either test the extractor by adding more files or click the Exit Training button.
Apply model to libraries
1.??????Click Add library.
2.??????Select a site and a library or paste in a URL.
3.??????Click Add.
4.??????Click Go to library.
Applying the model automates the process of creating new columns and content types and adding them to a site and a library. It even creates a custom view. To test it, drag and drop the same documents you used to train the model into the library. Wait a few minutes (it could take up to five minutes) and refresh the page.
?If you want to set up a demo in your own environment, here is Microsoft’s Syntex Adoption page. Near the bottom is the link to a 30-day trial.
Other articles in this series:
Part 1 - What Does Microsoft Syntex Do?
Part 2 - Microsoft Syntex Processing Models