Managing Knowledge Store projections

5 minutes 5 Questions

Knowledge Store projections in Azure Cognitive Search allow you to persist enriched data from your AI enrichment pipeline into durable storage for further analysis and downstream processing. Managing these projections effectively is crucial for building robust knowledge mining solutions. There are three types of projections you can configure: 1. **Table Projections**: Store structured data in Azure Table Storage. These are ideal for tabular data that you want to query using Power BI or other analytical tools. Each table projection maps enriched fields to table columns. 2. **Object Projections**: Save complex JSON objects to Azure Blob Storage. These preserve hierarchical data structures and are useful when you need to maintain nested relationships between enriched content. 3. **File Projections**: Store binary content like images or normalized documents in Blob Storage. This is particularly useful for preserving extracted images from documents. When managing projections, you define them within a skillset using the knowledgeStore property. Key configuration elements include: - **storageConnectionString**: The connection to your Azure Storage account - **projections**: An array containing your projection definitions - **generatedKeyName**: A unique identifier for tracking documents Best practices for managing projections include: - Use the Shaper skill to create well-structured data shapes before projecting - Group related projections together to maintain referential integrity - Plan your table schemas carefully to support your reporting needs - Consider data volume and storage costs when designing your projection strategy You can update projections by modifying the skillset definition and rerunning the indexer. The indexer will repopulate the knowledge store with the new projection configuration. Monitoring projection health involves checking indexer execution status, reviewing any warnings or errors, and validating that data appears correctly in your storage containers. Using Azure Portal or REST APIs, you can inspect and troubleshoot projection issues to ensure your knowledge mining solution functions as expected.

Managing Knowledge Store Projections

Why Is This Important?

Knowledge Store projections are a critical component of Azure Cognitive Search that allow you to persist enriched data from AI enrichment pipelines into durable storage. Understanding how to manage these projections is essential for the AI-102 exam because it demonstrates your ability to design end-to-end knowledge mining solutions that extract, transform, and store valuable insights from unstructured data.

What Are Knowledge Store Projections?

A Knowledge Store is a feature in Azure Cognitive Search that stores enriched content created by an AI enrichment pipeline. Projections define the shape and destination of this enriched data. They determine how your enriched documents are structured and where they are stored in Azure Storage.

There are three types of projections:

• Table Projections: Store data in Azure Table Storage as structured rows and columns. Ideal for analytical queries and reporting.

• Object Projections: Store data as JSON blobs in Azure Blob Storage. Useful for preserving complex nested structures.

• File Projections: Store binary data like images extracted during enrichment as files in Blob Storage.

How Do Knowledge Store Projections Work?

1. Define a Skillset: Create a skillset with cognitive skills that enrich your documents (entity recognition, key phrase extraction, OCR, etc.).

2. Configure the Knowledge Store: Within your skillset definition, add a knowledgeStore property that specifies the Azure Storage connection string.

3. Define Projections: Specify which enriched fields to project and in what format. Use the projections array to define table, object, or file projections.

4. Use Shaper Skills: The Shaper skill consolidates multiple enriched fields into a single complex type, making it easier to define projection shapes.

5. Run the Indexer: When the indexer processes documents, enriched content flows through the pipeline and is persisted to the Knowledge Store based on your projection definitions.

Key Configuration Elements:

• tableName: Specifies the name of the table for table projections
• container: Defines the blob container for object and file projections
• source: References the enriched field or Shaper output to project
• generatedKeyName: The column name for the generated key in tables

Exam Tips: Answering Questions on Managing Knowledge Store Projections

• Know the projection types: Be able to identify when to use table, object, or file projections based on the scenario requirements.

• Understand the Shaper skill: Questions often test your knowledge of using Shaper skills to consolidate fields before projection. Remember that Shaper helps create clean, organized structures.

• Connection strings matter: The Knowledge Store requires a valid Azure Storage connection string. Look for questions about configuration requirements.

• Recognize relationship scenarios: Table projections can maintain relationships between entities. Understand how generatedKeyName and referenceKeyName work together.

• File projections are specific: Only normalized images and binary content can be projected as files. Text content requires object or table projections.

• Read scenarios carefully: If a question mentions analytics or Power BI integration, table projections are typically the answer. For preserving document structure, object projections are preferred.

• Remember the storage types: Table projections go to Table Storage, while object and file projections go to Blob Storage.

• Practice JSON syntax: Be familiar with the JSON structure for defining projections within a skillset, including nested properties and arrays.
Test mode:
More Managing Knowledge Store projections questions
39 questions (total)