Adding document metadata in Amazon S3

To customize chat results for your end users, you can add metadata or document attributes to documents in an Amazon S3 bucket by using a metadata file. Metadata is additional information about a document, such as its title and the date and time it was created. To learn more about metadata in Amazon Q Business, see Document attributes in Amazon Q Business.

Document metadata location

In Amazon S3, each metadata file can be associated with an indexed document. Your metadata files must be stored in the same Amazon S3 bucket as your indexed files. You can specify a location within the Amazon S3 bucket for your metadata files by using the AWS Management Console. Or, you can use the metadataFilesPrefix field of the Amazon S3 configuration parameter using the JSON schema when you create an Amazon S3 data source using the CreateDataSource API.

If you don't specify an Amazon S3 prefix, your metadata files must be stored in the same location as your indexed documents. If you specify an Amazon S3 prefix for your metadata files, they must be in a directory structure parallel to your indexed documents. Amazon Q looks only in the specified directory for your metadata. If the metadata isn't read, check that the directory location matches the location of your metadata.

The following examples show how the indexed document location maps to the metadata file location. The document's Amazon S3 key is appended to the metadata's Amazon S3 prefix and then suffixed with .metadata.json to form the metadata file's Amazon S3 path.

Bucket name:
     s3://bucketName
Document path:
     documents
Metadata path:
     none
File mapping
     s3://bucketName/documents/file.txt -> 
        s3://bucketName/documents/file.txt.metadata.json

Bucket name:
     s3://bucketName
Document path:
     documents/legal
Metadata path:
     metadata
File mapping
     s3://bucketName/documents/legal/file.txt -> 
        s3://bucketName/metadata/documents/legal/file.txt.metadata.json

Document metadata structure

You define your document metadata itself in a JSON file. The file must be a UTF-8 text file without a BOM marker. The file name of the JSON file must be <document>.<extension>.metadata.json. In this example, document is the name of the document that the metadata applies to and extension is the file extension for the document. The document ID must be unique in <document>.<extension>.metadata.json.

The content of the JSON file uses the following template:

{
    "DocumentId": "document ID",
    "Attributes": {
        "_category": "document category",
        "_created_at": "ISO 8601 encoded string",
        "_last_updated_at": "ISO 8601 encoded string",
        "_source_uri": "document URI",
        "_version": "file version",
        "_view_count": number of times document has been viewed,
        "custom attribute key": "custom attribute value",
        additional custom attributes
    },
    "AccessControlList": [
         {
             "Name": "user name",
             "Type": "GROUP | USER",
             "Access": "ALLOW | DENY"
         }
    ],
    "Title": "document title",
    "ContentType": "PDF | HTML | MS_WORD | PLAIN_TEXT | PPT | RTF | XML | XSLT | MS_EXCEL | CSV | JSON | MD"
}

All of the attributes and fields are optional, so it's not necessary to include all attributes. However, you must provide a value for each attribute that you want to include; the value can't be empty. Note, however, that if you don't specify the _source_uri, the links returned by Amazon Q in the chat results point to the Amazon S3 bucket that contains the document, instead of the document itself.

The _created_at and _last_updated_at metadata fields are ISO 8601 encoded dates. For example, 2012-03-25T12:30:10+01:00 is the ISO 8601 date-time format for March 25, 2012, at 12:30PM (plus 10 seconds) in the Central European Time time zone.

You can add additional information to the Attributes field about a document that you use to filter queries or to group query responses.

You can use the AccessControlList field to filter the response from a query. This way, only certain users and groups have access to documents.