AI Builder Models

These models are broadly classified 4 types based on operation –

Documents Type

This contains all the models which deals with documents

Prebuild Models
Invoice Processing	Extract information from invoices
Text Recognition	Extract all the text in photos and PDF documents (OCR)
Receipt Processing	Extract information from receipts
Identity Document Reader	Extract information from identity documents
Business Card Reader	Extract information from Business Cards
Custom Models
Document Processing	Extract custom information from documents

Document Type Models

Text Type

This contains all the models which deals with text analysis

Prebuild Models
Sentiment Analysis	Detect Positive, Negative, or neutral sentiment in text data
Category classification	Classify customer feedback into predefined categories
Entity Extraction	Extract key elements from text, and classifies them into predefined categories
Key Phrase Extraction	Extract most relevant words and phrases from text
Language Detection	Detect the predominant language of a text document
Text translation	Detect and translate more than 90 supported languages
Azure Open AI Service	Create text, answer questions, summarize documents and more with GPT
Custom Models
Category Classification	Classify texts into custom categories
Entity Extraction	Extract custom entities from your text

Structured Data

This contains model which deals with structure of data

Custom Models
Prediction	Predict future outcomes from historical data

Images

This contains models which deals with images

Prebuild Models
Text recognition	Extract all the text in photos and PDF documents (OCR)
Custom Models
Object Detection	Detect custom objects in images

Each model will pass through below 4 phases in sequence –

• Build
• Train
• Manage
• Publish

Build Model

To build a model using AI builder has below prerequisite –

• AI builder requires Microsoft Dataverse storage to store and manage business data
• AI builder must be enabled for the environment

Train Model

Before using AI builder models, you need to train the models. More the training of the model with different samples the confidence or accuracy will be more. So, training is a vital activity in AI automation.

Training takes some time in AI builder based on the level of difficulty so once the model is trained first time, you have access to details page where you can manage the model and you can see performance score of the models.

On the details page, training results appear in the last trained version section.

Manage Model

Optimising an AI model is an iterative process whose results can vary depending on the configurations you set and the training data you provide. Updating these factors can affect the performance of your model.

After model is trained a performance score will appear for each trained version so accordingly you can improve the model by adjusting the factors affecting the model.

After evaluation of model, you can check whether your model is perfectly fit, Underfit or Overfit

Underfit – This occurs when your model is not able to perform which is less than the expectation, in this case you need to train your model with more relevant information

Overfit – This occurs when your model gives accurate prediction for training data but not for new data, in this case you need to adjust relevant parameters and retrain the model

If your model is already once published, you will see one Published version and one Last Trained version which is not published yet after retraining. You can choose which one you want to use as final version to use and publish accordingly.

Publish Model

After the model is trained successfully you can publish to make it available. All users in your current environment will be able to use your published model when you publish it.