Audioburst Search uses OData filter expressions to apply additional criteria to a search query besides full-text search terms. This article describes the syntax of filters in detail.


Syntax

A filter in the OData language is a Boolean expression, which in turn can be one of several types of expression, as shown by the following EBNF (Extended Backus-Naur Form):

boolean_expression ::=
collection_filter_expression
| logical_expression
| comparison_expression
| boolean_literal
| boolean_function_call
| '(' boolean_expression ')'
| variable


/* This can be a range variable in the case of a lambda, or a field path. */
variable ::= identifier | field_path

The types of Boolean expressions include:

  • Collection filter expressions using any or all. These apply filter criteria to collection fields.
  • Logical expressions that combine other Boolean expressions using the operators and, or, and not.
  • Comparison expressions, which compare fields or range variables to constant values using the operators eq, ne, gt, lt, ge, and le. Comparison expressions are also used to compare distances between geo-spatial coordinates using the geo.distance function. For more information.
  • The Boolean literals true and false. These constants can be useful sometimes when programmatically generating filters, but otherwise don't tend to be used in practice.
  • Calls to Boolean functions, including:
    -- geo.intersects, which tests whether a given point is within a given polygon. For more information.
    -- search.in, which compares a field or range variable with each value in a list of values. For more information.
    -- search.ismatch and search.ismatchscoring, which execute full-text search operations in a filter context. For more information.
  • Field paths or range variables of type Edm.Boolean. For example, if your index has a Boolean field called IsEnabled and you want to return all documents where this field is true, your filter expression can just be the name IsEnabled.
  • Boolean expressions in parentheses. Using parentheses can help to explicitly determine the order of operations in a filter. For more information on the default precedence of the OData operators, see the next section.

Operator precedence in filters

If you write a filter expression with no parentheses around its sub-expressions, Azure Cognitive Search will evaluate it according to a set of operator precedence rules. These rules are based on which operators are used to combine sub-expressions. The following table lists groups of operators in order from highest to lowest precedence:

An operator that is higher in the above table will "bind more tightly" to its operands than other operators. For example, and is of higher precedence than or, and comparison operators are of higher precedence than either of them, so the following two expressions are equivalent:

Rating gt 0 and Rating lt 3 or Rating gt 7 and Rating lt 10
((Rating gt 0) and (Rating lt 3)) or ((Rating gt 7) and (Rating lt 10))

The not operator has the highest precedence of all -- even higher than the comparison operators. That's why if you try to write a filter like this:

not Rating gt 5

You'll get this error message:

Invalid expression: A unary operator with an incompatible type was detected. Found operand type 'Edm.Int32' for operator kind 'Not'.

This error happens because the operator is associated with just the Rating field, which is of type Edm.Int32, and not with the entire comparison expression. The fix is to put the operand of not in parentheses:

not (Rating gt 5)


Filter size limitations

There are limits to the size and complexity of filter expressions that you can send to Azure Cognitive Search. The limits are based roughly on the number of clauses in your filter expression. A good guideline is that if you have hundreds of clauses, you are at risk of exceeding the limit. We recommend designing your application in such a way that it doesn't generate filters of unbounded size.

Tip

Using the search.in function instead of long disjunctions of equality comparisons can help avoid the filter clause limit, since a function call counts as a single clause.

Examples

Find all hotels with at least one room with a base rate of less than $200 that are rated at or above 4:

filter=Rooms/any(room: room/BaseRate lt 200.0) and Rating ge 4

Find all hotels other than "Sea View Motel" that have been renovated since 2010:

filter=HotelName ne 'Sea View Motel' and LastRenovationDate ge 2010-01-01T00:00:00Z

Find all hotels that were renovated in 2010 or later. The datetime literal includes time zone information for Pacific Standard Time:

filter=LastRenovationDate ge 2010-01-01T00:00:00-08:00

Find all hotels that have parking included and where all rooms are non-smoking:

filter=ParkingIncluded and Rooms/all(room: not room/SmokingAllowed)

- OR -

filter=ParkingIncluded eq true and Rooms/all(room: room/SmokingAllowed eq false)

Find all hotels that are Luxury or include parking and have a rating of 5:

filter=(Category eq 'Luxury' or ParkingIncluded eq true) and Rating eq 5

Find all hotels with the tag "wifi" in at least one room (where each room has tags stored in a Collection(Edm.String) field):

filter=Rooms/any(room: room/Tags/any(tag: tag eq 'wifi'))

Find all hotels with any rooms:

filter=Rooms/any()

Find all hotels that don't have rooms:

filter=not Rooms/any()

Find all hotels within 10 kilometers of a given reference point (where Location is a field of type Edm.GeographyPoint):

filter=geo.distance(Location, geography'POINT(-122.131577 47.678581)') le 10

Find all hotels within a given viewport described as a polygon (where Location is a field of type Edm.GeographyPoint). The polygon must be closed, meaning the first and last point sets must be the same. Also, the points must be listed in counterclockwise order.

filter=geo.intersects(Location, geography'POLYGON((-122.031577 47.578581, -122.031577 47.678581, -122.131577 47.678581, -122.031577 47.578581))')

Find all hotels where the "Description" field is null. The field will be null if it was never set, or if it was explicitly set to null:

filter=Description eq null

Find all hotels with name equal to either 'Sea View motel' or 'Budget hotel'). These phrases contain spaces, and space is a default delimiter. You can specify an alternative delimiter in single quotes as the third string parameter:

filter=search.in(HotelName, 'Sea View motel,Budget hotel', ',')

Find all hotels with name equal to either 'Sea View motel' or 'Budget hotel' separated by '|'):

filter=search.in(HotelName, 'Sea View motel|Budget hotel', '|')

Find all hotels where all rooms have the tag 'wifi' or 'tub':

filter=Rooms/any(room: room/Tags/any(tag: search.in(tag, 'wifi, tub'))

Find a match on phrases within a collection, such as 'heated towel racks' or 'hairdryer included' in tags.

filter=Rooms/any(room: room/Tags/any(tag: search.in(tag, 'heated towel racks,hairdryer included', ','))

Find documents with the word "waterfront". This filter query is identical to a search request with search=waterfront.

filter=search.ismatchscoring('waterfront')

Find documents with the word "hostel" and rating greater or equal to 4, or documents with the word "motel" and rating equal to 5. This request couldn't be expressed without the search.ismatchscoring function since it combines full-text search with filter operations using or.

filter=search.ismatchscoring('hostel') and rating ge 4 or search.ismatchscoring('motel') and rating eq 5

Find documents without the word "luxury".

filter=not search.ismatch('luxury')

Find documents with the phrase "ocean view" or rating equal to 5. The search.ismatchscoring query will be executed only against fields HotelName and Description. Documents that matched only the second clause of the disjunction will be returned too -- hotels with Rating equal to 5. Those documents will be returned with score equal to zero to make it clear that they didn't match any of the scored parts of the expression.

filter=search.ismatchscoring('"ocean view"', 'Description,HotelName') or Rating eq 5

Find hotels where the terms "hotel" and "airport" are no more than five words apart in the description, and where all rooms are non-smoking. This query uses the full Lucene query language .

filter=search.ismatch('"hotel airport"~5', 'Description', 'full', 'any') and not Rooms/any(room: room/SmokingAllowed)

Did this answer your question?