Demo2

M12 DEMO #2

• Custom Skill for Azure Search

This code is provided for demo purposes only for course AI-102.

Requirements

• Azure Subscription
• Visual Studio Code or VS 2019
• .Net Core

Building Custom Skill for Azure Search

1. From the Azure portal create a new Cognitive Service Translator and pick the endpoint address and keys.

2. Create a new Function on azure portal or use Visual Studio Code to create and deploy HTTP triggered functions. Refer to the following tutorial for details.

3. In folder TranslateFunction you can find a demo project for deployment. Update TranslateFunction.cs the following values with your Translator service.

      static string CognitiveServicesTokenUri = "https://<your-service-name>.cognitiveservices.azure.com/sts/v1.0/issuetoken";
      static string SubscriptionKey = "<your key>";

4. Publish and test your function. Make sure that translation is working as expected. You can use following JSON as body for test:

   {"values": [
         {
               "recordId": "a1",
               "data":
               {
               "text":  "Tiger, tiger burning bright in the darkness of the night.",
               "from": "en",
               "to": "fr"
      }
   }] }

   

5. Create or reuse a new storage account. Create container reviews and upload file HotelReviews_Free.csv in it.

6. Create or reuse existing Azure Cognitive Search. Register new Data Source from the file you uploaded in the storage account above.

7. Click on 'Import' on top to start importing wizard. Some details you can find in doc.

8. When the index build you can open Search explorer and run a few queries to check if review content can be retrieved.

   

9. Now you need to add your function to custom skills update skill's JSON as explained in tutorial. Example of the custom skill modification

   {
         "@odata.type": "#Microsoft.Skills.Custom.WebApiSkill",
         "name": "custom translate to italian",
         "description": "translate description on italian by use custom skill",
         "context": "/document",
         "uri": "https://funazuretranslate.azurewebsites.net/api/Translate?to=en&code=<YOUR CODE>",
         "httpMethod": "POST",
         "timeout": "PT30S",
         "batchSize": 1,
         "degreeOfParallelism": 20,
         "inputs": [
         {
            "name": "text",
            "source": "/document/reviews_text"
         },
         {
            "name": "from",
            "source": "/document/Language"
         }
         ],
         "outputs": [
         {
            "name": "text",
            "targetName": "reviews_text_en"
         }
         ],
         "httpHeaders": {}
      }

10. Now you need to create a new field with name reviews_text_en referenced in skill. First you need to update indexer definition JSON with new output like following:

    "outputFieldMappings": [
       {
          "sourceFieldName": "/document/reviews_text_en",
          "targetFieldName": "reviews_text_en"
       }
    ],

11. Second add new field in index definition JSON like following:

    fields [
       ...
       {
          "name": "reviews_text_en",
          "type": "Edm.String",
          "facetable": false,
          "filterable": false,
          "key": false,
          "retrievable": true,
          "searchable": false,
          "sortable": false,
          "analyzer": null,
          "indexAnalyzer": null,
          "searchAnalyzer": null,
          "synonymMaps": [],
          "fields": []
       },
       ...

12. Finally you need to reset and run again the indexer. It should be successful. It is important to copalite reset before start a new run

    

13. Next request in Search explorer should return you back the reviews with english translation.