Context Suggester

The context suggester is an extension to the suggest API of Elasticsearch. Namely the suggester system provides a very fast way of searching documents by handling these entirely in memory. But this special treatment does not allow the handling of traditional queries and filters, because those would have notable impact on the performance. So the context extension is designed to take so-called context information into account to specify a more accurate way of searching within the suggester system. Instead of using the traditional query and filter system a predefined ``context`` is configured to limit suggestions to a particular subset of suggestions. Such a context is defined by a set of context mappings which can either be a simple category or a geo location. The information used by the context suggester is configured in the type mapping with the context parameter, which lists all of the contexts that need to be specified in each document and in each suggestion request. For instance:

PUT services/_mapping/service
{
    "service": {
        "properties": {
            "name": {
                "type" : "string"
            },
            "tag": {
                "type" : "string"
            },
            "suggest_field": {
                "type": "completion",
                "context": {
                    "color": { 
                        "type": "category",
                        "path": "color_field",
                        "default": ["red", "green", "blue"]
                    },
                    "location": { 
                        "type": "geo",
                        "precision": "5m",
                        "neighbors": true,
                        "default": "u33"
                    }
                }
            }
        }
    }
}

However contexts are specified (as type category or geo, which are discussed below), each context value generates a new sub-set of documents which can be queried by the completion suggester. All three types accept a default parameter which provides a default value to use if the corresponding context value is absent.

The basic structure of this element is that each field forms a new context and the fieldname is used to reference this context information later on during indexing or querying. All context mappings have the default and the type option in common. The value of the default field is used, when ever no specific is provided for the certain context. Note that a context is defined by at least one value. The type option defines the kind of information hold by this context. These type will be explained further in the following sections.

The category context allows you to specify one or more categories in the document at index time. The document will be assigned to each named category, which can then be queried later. The category type also allows to specify a field to extract the categories from. The path parameter is used to specify this field of the documents that should be used. If the referenced field contains multiple values, all these values will be used as alternative categories.

The mapping for a category is simply defined by its default values. These can either be defined as list of default categories:

"context": {
    "color": {
        "type": "category",
        "default": ["red", "orange"]
    }
}

or as a single value

"context": {
    "color": {
        "type": "category",
        "default": "red"
    }
}

or as reference to another field within the documents indexed:

"context": {
    "color": {
        "type": "category",
        "default": "red",
        "path": "color_field"
    }
}

in this case the default categories will only be used, if the given field does not exist within the document. In the example above the categories are received from a field named color_field. If this field does not exist a category red is assumed for the context color.

Within a document the category is specified either as an array of values, a single value or null. A list of values is interpreted as alternative categories. So a document belongs to all the categories defined. If the category is null or remains unset the categories will be retrieved from the documents field addressed by the path parameter. If this value is not set or the field is missing, the default values of the mapping will be assigned to the context.

PUT services/service/1
{
    "name": "knapsack",
    "suggest_field": {
        "input": ["knacksack", "backpack", "daypack"],
        "context": {
            "color": ["red", "yellow"]
        }
    }
}

A query within a category works similar to the configuration. If the value is null the mappings default categories will be used. Otherwise the suggestion takes place for all documents that have at least one category in common with the query.

POST services/_suggest?pretty'
{
    "suggest" : {
        "text" : "m",
        "completion" : {
            "field" : "suggest_field",
            "size": 10,
            "context": {
                "color": "red"
            }
        }
    }
}

A geo context allows you to limit results to those that lie within a certain distance of a specified geolocation. At index time, a lat/long geo point is converted into a geohash of a certain precision, which provides the context.

The mapping for a geo context accepts four settings, only of which precision is required:

Since all locations of this mapping are translated into geohashes, each location matches a geohash cell. So some results that lie within the specified range but not in the same cell as the query location will not match. To avoid this the neighbors option allows a matching of cells that join the bordering regions of the documents location. This option is turned on by default. If a document or a query doesn’t define a location a value to use instead can defined by the default option. The value of this option supports all the ways a geo_point can be defined. The path refers to another field within the document to retrieve the location. If this field contains multiple values, the document will be linked to all these locations.

"context": {
    "location": {
        "type": "geo",
        "precision": ["1km", "5m"],
        "neighbors": true,
        "path": "pin",
        "default": {
            "lat": 0.0,
            "lon": 0.0
        }
    }
}

Within a document a geo location retrieved from the mapping definition can be overridden by another location. In this case the context mapped to a geo location supports all variants of defining a geo_point.

PUT services/service/1
{
    "name": "some hotel 1",
    "suggest_field": {
        "input": ["my hotel", "this hotel"],
        "context": {
            "location": {
                    "lat": 0,
                    "lon": 0
            }
        }
    }
}

Like in the configuration, querying with a geo location in context, the geo location query supports all representations of a geo_point to define the location. In this simple case all precision values defined in the mapping will be applied to the given location.

POST services/_suggest
{
    "suggest" : {
        "text" : "m",
        "completion" : {
            "field" : "suggest_field",
            "size": 10,
            "context": {
                "location": {
                    "lat": 0,
                    "lon": 0
                }
            }
        }
    }
}

But it also possible to set a subset of the precisions set in the mapping, by using the precision parameter. Like in the mapping, this parameter is allowed to be set to a single precision value or a list of these.

POST services/_suggest
{
    "suggest" : {
        "text" : "m",
        "completion" : {
            "field" : "suggest_field",
            "size": 10,
            "context": {
                "location": {
                    "value": {
                        "lat": 0,
                        "lon": 0
                    },
                    "precision": "1km"
                }
            }
        }
    }
}

A special form of the query is defined by an extension of the object representation of the geo_point. Using this representation allows to set the precision parameter within the location itself:

POST services/_suggest
{
    "suggest" : {
        "text" : "m",
        "completion" : {
            "field" : "suggest_field",
            "size": 10,
            "context": {
                "location": {
                        "lat": 0,
                        "lon": 0,
                        "precision": "1km"
                }
            }
        }
    }
}