Content Moderation is about refining and revisiting that data , post , image or video to make it acceptable , socially , psychologically ,legally and in short , right content to the right audience.

With the high volumes of data flowing in the manual moderation might be quite near to impossible.

Azure provides a rich set of APIs that help you to moderate and is available for the below categories.

  1. Text Moderation
  2. Image Moderation
  3. Video Moderation

Azure provides a rich set of APIs that help you to moderate and is available for the below categories.

Text Moderation

Azure APIs help in moderating the text , and can deal with all of the below categories.

  • Profanity
  • Classification
  • Personal Data Identification
  • Auto correction of text
  • Language detection
  • PII

To start with any moderation , we have to provision the azure resource. called Content Moderation

Azure Moderation service

If you need more details on how to provision this resource , the below video can be helpful.

EndPoint

Navigate to the newly provisioned resource and then to Keys and End Point section save either of the keys and the end point

The endpoint used is

 <EndPointURI>/contentmoderator/moderate/v1.0/ProcessText/Screen?autocorrect=<bool>&PII=<bool>&classify= <bool>
Keyword ​Details
Endpoint The url from the previous screenshot marked as 3
AutoCorrect Rectifies any misspellings in the text
PII The personal Identifiable information includes Email , PII etc
Classify categories the text
   

Azure EndPoint

PII

The following personal information can be detected on the scanned text

  1. Email
  2. IPA
  3. Phone
  4. Address
  5. SSN
  6. Classify

Azure automatically categorises information to 3 different categories

Category Details  
Category1 Presence of Adult or sexually suggestive content
Category2 Presence of content that is mature  
Category3 Presence of Offensive words  
ReviewRecommended AI model suggests to reexamine the text , before publishing  

Example

To demonstrate the example I have created a sample power automate , An Instant cloud flow with the following parameters

Azure Flow Inputs

  1. The body of the text that needs to be scanned
  2. If PII classification should happen
  3. If the categories are needed
  4. Automatically corrects the spellings and returns the output

Azure Flow Inputs

All the inputs gathered from the step 1 should be used at the respective parameters.

The Oct-Apim-Subscription-Key refers to the keys captured at this step

Azure Flow Inputs

I just noted that spell correct is given as text so given true as input and below is the screenshot of the output

Azure Flow Inputs 

{
  "OriginalText": "Is this a grabage or crappy , email abcdef@abcd.com, phone: 4255550111, IP: \n255.255.255.255, 1234 Main Boulevard, Panapolis WA 96555.",
  "NormalizedText": "   grabage  crappy , email abcdef@abcd.com, phone: 4255550111, IP: \n255.255.255.255, 1234 Main Boulevard, Panapolis WA 96555.",
  "AutoCorrectedText": "Is this a garbage or crappy , email abcdef@abcd.com, phone: 4255550111, IP: \n255.255.255.255, 1234 Main Boulevard, Pentapolis WA 96555.",
  "Misrepresentation": null,
  "PII": {
    "Email": [
      {
        "Detected": "abcdef@abcd.com",
        "SubType": "Regular",
        "Text": "abcdef@abcd.com",
        "Index": 36
      }
    ],
    "IPA": [
      {
        "SubType": "IPV4",
        "Text": "255.255.255.255",
        "Index": 77
      }
    ],
    "Phone": [
      {
        "CountryCode": "US",
        "Text": "4255550111",
        "Index": 60
      }
    ],
    "Address": [
      {
        "Text": "1234 Main Boulevard, Panapolis WA 96555",
        "Index": 94
      }
    ],
    "SSN": []
  },
  "Classification": {
    "ReviewRecommended": true,
    "Category1": {
      "Score": 0.0014615426771342754
    },
    "Category2": {
      "Score": 0.201115220785141
    },
    "Category3": {
      "Score": 0.9879999756813049
    }
  },
  "Language": "eng",
  "Terms": null,
  "Status": {
    "Code": 3000,
    "Description": "OK",
    "Exception": null
  },
  "TrackingId": "befb04b6-27b2-4bc6-9de9-34168fea88bb"
}

As noted the output would be in JSON format , which provides an easy access or integration with any systems

This article targets what can be addressed with Text content Moderation by default , if we need to screen the custom keywords , that is as well supported and will see that in further blogs