DQL overview

The Discovery Query Language defines syntax you can use to filter, search, and analyze your data.

How to write a Discovery Query Language query

The Discovery Query Language leverages the structure of indexed documents. The following JSON snippet shows an indexed document from a collection where the Entities enrichment is applied. As a result of the enrichment, the JSON structure captures any mentions of known entities, such as city names, companies, or famous people.

In this example, the recognized entity is the company name IBM.

{
  "document": {
    "document_id": "f7f27ea30eb3e4c0ce21830618d9ee99",
    "enriched_text": [
      {
        "entities": [
          {
            "model_name": "natural_language_understanding",
            "mentions": [],
            "text":"IBM",
            "type":"Organization"
          }
        ]
      }
    ]
  }
}

To create a query that returns all of the documents in which the entity IBM is mentioned, use the following syntax:

The structure of the query enriched_text.entities.text:IBM, where text in enriched_text is the field where the enrichment is applied, and IBM is the term you are looking for in the enriched_text.entities.text subfield. — Example query structure

This basic query contains a nested path expression before the : operator. Each path element is the name of a field in the document separated by a period (.). The : operator indicates that the text that follows the operator must be included in the result.

The :: operator indicates that the text must be matched exactly in the result. For more information, see Query operators. You can see how the two operators are used in the following examples.

To return matching documents in order of relevance, pass the following data object in the POST request:
```
{
  "query":"enriched_text.entities.text:IBM"
}
```
To return matching documents in any order, pass the following data object in the POST request as the query body:
```
{
  "filter":"enriched_text.entities.text::IBM"
}
```

Using the filter and query parameters together

The filter parameter returns faster than the query parameter and its results are cached. If you submit queries that use the filter and query parameters separately on a small data set, each request returns similar (if not identical) results.

In large data sets, if you need results to be returned in order of relevance, combine the filter and query parameters. Using the parameters together improves performance because the filter parameter is applied first. It filters the documents and caches the results. The query parameter then ranks the cached results.

Filter example: Get a document by its ID

Query body:

{
  "filter": "document_id::b6d8c6e3-1097-421b-9e39-75717d2554aa"
}

If the document exists, the query returns 1 matching result. If it doesn't, the query returns no matching results.

Filter example: Find a document ID by its file name

If you don't know the document_id of a document, but you know the original filename of the document, you can use the filter and return parameters together to discover the document_id.

Query body:

{
  "filter": "extracted_metadata.filename::100674.txt",
  "return": [ "document_id", "extracted_metadata" ]
}

Response:

{
  "matching_results": 1,
  "results": [
    {
      "document_id": "b6d8c6e3-1097-421b-9e39-75717d2554aa",
      "extracted_metadata": {
        "sha1": "AD447F7592A17CDCBF0A589C4E6EC2087AF7H35F",
        "filename": "100674.txt",
        "file_type": "text"
      }
    }
  ]
}

Filter example: Find documents that mention an entity value

The query looks for documents that mention the entity Gilroy and finds 4 matching documents.

Query body:

{
  "filter": "enriched_text.entities.text::Gilroy"
}

Response:

{
  "matching_results": 4
}

Filtering nested values

You can nest one filter inside another to ensure that the documents that are returned match more than one condition.

In the documents used for these examples, the entity "Gilroy" appears as both a "Location" (a town in California) and as a "Person" (a surname) entity type. To find documents where "Gilroy" appears as a location, write a query that filters on two nested fields at the same time: the entity text must be "Gilroy" and the entity type must be "Location".

One way to write the query is as follows:

{
  "filter": "enriched_text.entities.text::Gilroy,enriched_text.entities.type::Location"
}

This query matches documents where some path enriched_text.entities.text is Gilroy and some path enriched_text.entities.type::Location is Location. However, there is no guarantee that those two paths will be under the same entities object. For example, the query matches documents that have Gilroy as a Person entity type and, at the same time, have some other Location entity type object.

To accurately capture the nested semantics of this query, nest the filter values by using the following syntax:

Query body:

{
  "filter": "enriched_text.entities:(text::Gilroy,type::Location)"
}

This stricter query matches only those documents in which there is an entities object with text equal to Gilroy and type equal to Location.

As another example, if you want to match documents that contain an entities object with text equal to Gilroy but type not equal to Location, you can use the not equal operator in the query, for example:

{
  "filter": "enriched_text.entities:(text::Gilroy,type::!Location)"
}

You can also use aggregations to do more sophistocated filtering of the results. For more information about the available aggregation types, see Query aggregations.

For more information about the Discovery Query Language, see the following topics: