AI-900 정리노트 - Document intelligence, Knowledge Mining

• Document intelligence extends OCR by extracting, understanding, and organizing text data.
• Automates document processing (e.g., receipts, forms), reducing manual work and errors.
• Example: Scans a receipt, extracts merchant info, total, and tax, and maps them into a database.
• Azure AI Document Intelligence provides prebuilt and custom models for document analysis.

Azure AI Search

Azure AI Search Features

• Built on Apache Lucene: Programmable search engine with 99.9% uptime SLA for cloud and on-premises.
• Data from any source: Accepts JSON data; auto-crawling supported for selected Azure sources.
• Multiple search options: Supports vector search, full-text search, and hybrid search.
• AI enrichment: Built-in image and text analysis capabilities using Azure AI.
• Linguistic analysis: Supports 56 languages with phonetic matching and language-specific linguistics.
• Configurable user experience: Offers vector queries, text search, hybrid queries, fuzzy search, autocomplete, geo-search, and more.
• Azure scale, security, and integration: Works across data, machine learning, Azure AI services, and Azure OpenAI.

Azure AI Search Data Flow

1. Start with a data source

   • Original data artifacts like PDFs, videos, images, or database text (e.g., Azure Storage, Azure SQL Database, Azure Cosmos DB).

2. Indexer

   • Automates data movement from source through document cracking, enrichment, and into indexing.
   • Converts original file types to JSON (called JSON serialization).

3. Document cracking

   • The indexer opens files and extracts content for processing.

4. Enrichment

   • Applies AI skills (built-in or custom) to enrich extracted data.
   • Skillsets define operations like OCR, text translation, captioning images, or evaluating text sentiment.
   • Enriched content can be saved in a knowledge store (tables/blobs in Azure Storage).

5. Push to index

   • Serialized JSON is used to populate the search index.

6. Querying the index

   • Users search (e.g., "coffee") against the search index.
   • The index has a schema (like a table) with fields, data types (e.g., string), and attributes for filtering, sorting, and searching.

How to use

• Azure portal's import data wizard
• with the REST API
• SDK