AzureFeeds

When users search for content within large PDFs—such as contracts, manuals, or reports—they often need to know not just what was found, but where it was found. Associating search results with page numbers enables:

The AI search service must have Storage Blob Data Reader access to the container. If using RBAC, ensure the managed identity is properly assigned. 

Ref: 

• AI search Search-blob-indexer-role-based-access


• How to Index Azure Blobs

 

The first step in skillset execution is document cracking, which separates text and image content. A common use case for Text Merger is merging the textual representation of images—such as OCR output or image captions—into the content field of a document. This is especially useful for PDFs or Word documents that combine text with embedded images. This ensures that the final enriched document includes all relevant textual data, regardless of its original format, and improves the accuracy of downstream search and analysis.

an index projection specifies how parent-child content is mapped to fields in a search index for one-to-many indexing. 

{
“@odata.etag”: “”0x8DDD58F12B5D0B9″”,
“name”: “pagenumskillset”,
“description”: “Skillset to feed document to OCR skill and use Index Projection to split the content page wise”,
“skills”: [
{
“@odata.type”: “#Microsoft.Skills.Vision.OcrSkill”,
“name”: “#1”,
“context”: “/document/normalized_images/*”,
“lineEnding”: “Space”,
“defaultLanguageCode”: “en”,
“detectOrientation”: true,
“inputs”: [
{
“name”: “image”,
“source”: “/document/normalized_images/*”,
“inputs”: []
}
],
“outputs”: [
{
“name”: “text”,
“targetName”: “text”
},
{
“name”: “layoutText”,
“targetName”: “layoutText”
}
]
}
],
“cognitiveServices”: {
“@odata.type”: “#Microsoft.Azure.Search.DefaultCognitiveServices”
},
“indexProjections”: {
“selectors”: [
{
“targetIndexName”: “pagenumidx”,
“parentKeyFieldName”: “ParentKey”,
“sourceContext”: “/document/normalized_images/*”,
“mappings”: [
{
“name”: “DocText”,
“source”: “/document/normalized_images/*/text”,
“inputs”: []
},
{
“name”: “DocName”,
“source”: “/document/metadata_storage_name”,
“inputs”: []
},
{
“name”: “DocURL”,
“source”: “/document/metadata_storage_path”,
“inputs”: []
},
{
“name”: “PageNum”,
“source”: “/document/normalized_images/*/pageNumber”,
“inputs”: []
}
]
}
],
“parameters”: {
“projectionMode”: “skipIndexingParentDocuments”
}
}
}

 

An index is defined by a schema and stored within the search service, ensuring millisecond response times by decoupling from external data sources. Except for indexer-driven scenarios, the search service never queries the original data directly making it ideal for high-performance search applications.

{
“@odata.etag”: “”0x8DDD58E7F7CF595″”,
“name”: “pagenumidx”,
“fields”: [
{
“name”: “ID”,
“type”: “Edm.String”,
“searchable”: true,
“filterable”: false,
“retrievable”: true,
“stored”: true,
“sortable”: false,
“facetable”: false,
“key”: true,
“analyzer”: “keyword”,
“synonymMaps”: []
},
{
“name”: “DocText”,
“type”: “Edm.String”,
“searchable”: true,
“filterable”: false,
“retrievable”: true,
“stored”: true,
“sortable”: false,
“facetable”: false,
“key”: false,
“synonymMaps”: []
},
{
“name”: “DocName”,
“type”: “Edm.String”,
“searchable”: true,
“filterable”: true,
“retrievable”: true,
“stored”: true,
“sortable”: true,
“facetable”: true,
“key”: false,
“synonymMaps”: []
},
{
“name”: “PageNum”,
“type”: “Edm.Int32”,
“searchable”: false,
“filterable”: true,
“retrievable”: true,
“stored”: true,
“sortable”: true,
“facetable”: false,
“key”: false,
“synonymMaps”: []
},
{
“name”: “DocURL”,
“type”: “Edm.String”,
“searchable”: true,
“filterable”: true,
“retrievable”: true,
“stored”: true,
“sortable”: false,
“facetable”: false,
“key”: false,
“analyzer”: “standard.lucene”,
“synonymMaps”: []
},
{
“name”: “ParentKey”,
“type”: “Edm.String”,
“searchable”: true,
“filterable”: true,
“retrievable”: true,
“stored”: true,
“sortable”: false,
“facetable”: false,
“key”: false,
“analyzer”: “keyword”,
“synonymMaps”: []
}
],
“scoringProfiles”: [],
“suggesters”: [],
“analyzers”: [],
“normalizers”: [],
“tokenizers”: [],
“tokenFilters”: [],
“charFilters”: [],
“similarity”: {
“@odata.type”: “#Microsoft.Azure.Search.BM25Similarity”
},
“semantic”: {
“defaultConfiguration”: “my_semantic_cfg”,
“configurations”: [
{
“name”: “my_semantic_cfg”,
“flightingOptIn”: false,
“rankingOrder”: “BoostedRerankerScore”,
“prioritizedFields”: {
“titleField”: {
“fieldName”: “DocName”
},
“prioritizedContentFields”: [
{
“fieldName”: “DocText”
}
],
“prioritizedKeywordsFields”: [
{
“fieldName”: “DocName”
},
{
“fieldName”: “ID”
}
]
}
}
]
}
}

 

Azure AI Search allows you to extract page-level data by configuring the indexer with:

This setting renders each PDF page as a separate image, which can then be processed using the OcrSkill. The OCR output can be mapped to a collection field, where each item corresponds to a page’s text. This method enables you to infer page numbers based on the position of matched content in the collection.

{
“@odata.context”: “https://searchinstancename.search.windows.net/$metadata#indexers/$entity”,
“@odata.etag”: “”0x8DDD58F3760067D””,
“name”: “indexer-pagenum”,
“description”: null,
“dataSourceName”: “azureblob-1754401027271-datasource”,
“skillsetName”: “pagenumskillset”,
“targetIndexName”: “pagenumidx”,
“disabled”: null,
“schedule”: null,
“parameters”: {
“batchSize”: null,
“maxFailedItems”: null,
“maxFailedItemsPerBatch”: null,
“configuration”: {
“dataToExtract”: “contentAndMetadata”,
“parsingMode”: “default”,
“imageAction”: “generateNormalizedImagePerPage”,
“pdfTextRotationAlgorithm”: “none”
}
},
“fieldMappings”: [],
“outputFieldMappings”: [],
“cache”: null,
“encryptionKey”: null
}