SearchSpec.Builder

Builder

public Builder ()

Constructs a new builder for SearchSpec objects.

addEmbeddingParameters

public SearchSpec.Builder addEmbeddingParameters (EmbeddingVector... searchEmbeddings)

Adds an embedding search to SearchSpec Entry, which will be referred in the query expression and the ranking expression for embedding search.

addEmbeddingParameters

public SearchSpec.Builder addEmbeddingParameters (Collection<EmbeddingVector> searchEmbeddings)

Adds an embedding search to SearchSpec Entry, which will be referred in the query expression and the ranking expression for embedding search.

addFilterNamespaces

public SearchSpec.Builder addFilterNamespaces (String... namespaces)

Adds a namespace filter to SearchSpec Entry. Only search for documents that have the specified namespaces.

If unset, the query will search over all namespaces.

addFilterNamespaces

public SearchSpec.Builder addFilterNamespaces (Collection<String> namespaces)

Adds a namespace filter to SearchSpec Entry. Only search for documents that have the specified namespaces.

If unset, the query will search over all namespaces.

addFilterPackageNames

public SearchSpec.Builder addFilterPackageNames (String... packageNames)

Adds a package name filter to SearchSpec Entry. Only search for documents that were indexed from the specified packages.

If unset, the query will search over all packages that the caller has access to. If package names are specified which caller doesn't have access to, then those package names will be ignored.

addFilterPackageNames

public SearchSpec.Builder addFilterPackageNames (Collection<String> packageNames)

Adds a package name filter to SearchSpec Entry. Only search for documents that were indexed from the specified packages.

If unset, the query will search over all packages that the caller has access to. If package names are specified which caller doesn't have access to, then those package names will be ignored.

addFilterProperties

public SearchSpec.Builder addFilterProperties (String schema, 
                Collection<String> propertyPaths)

Adds property paths for the specified type to the property filter of SearchSpec Entry. Only returns documents that have matches under the specified properties. If property paths are added for a type, then only the properties referred to will be searched for results of that type.

If a property path that is specified isn't present in a result, it will be ignored for that result. Property paths cannot be null.

If no property paths are added for a particular type, then all properties of results of that type will be searched.

Example properties: 'body', 'sender.name', 'sender.emailaddress', etc.

If property paths are added for the SearchSpec.SCHEMA_TYPE_WILDCARD, then those property paths will apply to all results, excepting any types that have their own, specific property paths set.

addFilterPropertyPaths

public SearchSpec.Builder addFilterPropertyPaths (String schema, 
                Collection<PropertyPath> propertyPaths)

Adds property paths for the specified type to the property filter of SearchSpec Entry. Only returns documents that have matches under the specified properties. If property paths are added for a type, then only the properties referred to will be searched for results of that type.

addFilterSchemas

public SearchSpec.Builder addFilterSchemas (Collection<String> schemas)

Adds a Schema type filter to SearchSpec Entry. Only search for documents that have the specified schema types.

If unset, the query will search over all schema types.

addFilterSchemas

public SearchSpec.Builder addFilterSchemas (String... schemas)

Adds a Schema type filter to SearchSpec Entry. Only search for documents that have the specified schema types.

If unset, the query will search over all schema types.

addInformationalRankingExpressions

public SearchSpec.Builder addInformationalRankingExpressions (String... informationalRankingExpressions)

Adds informational ranking expressions to be evaluated for each document in the search result. The values of these expressions will be returned to the caller via SearchResult.getInformationalRankingSignals(). These expressions are purely for the caller to retrieve additional information about the result and have no effect on ranking.

The syntax is exactly the same as specified in setRankingStrategy(java.lang.String).

addInformationalRankingExpressions

public SearchSpec.Builder addInformationalRankingExpressions (Collection<String> informationalRankingExpressions)

Adds informational ranking expressions to be evaluated for each document in the search result. The values of these expressions will be returned to the caller via SearchResult.getInformationalRankingSignals(). These expressions are purely for the caller to retrieve additional information about the result and have no effect on ranking.

The syntax is exactly the same as specified in setRankingStrategy(java.lang.String).

addProjection

public SearchSpec.Builder addProjection (String schema, 
                Collection<String> propertyPaths)

Adds property paths for the specified type to be used for projection. If property paths are added for a type, then only the properties referred to will be retrieved for results of that type. If a property path that is specified isn't present in a result, it will be ignored for that result. Property paths cannot be null.

addProjectionPaths

public SearchSpec.Builder addProjectionPaths (String schema, 
                Collection<PropertyPath> propertyPaths)

Adds property paths for the specified type to be used for projection. If property paths are added for a type, then only the properties referred to will be retrieved for results of that type. If a property path that is specified isn't present in a result, it will be ignored for that result. Property paths cannot be null.

If no property paths are added for a particular type, then all properties of results of that type will be retrieved.

If property path is added for the SearchSpec.SCHEMA_TYPE_WILDCARD, then those property paths will apply to all results, excepting any types that have their own, specific property paths set.

Suppose the following document is in the index.

Email: Document {
   sender: Document {
     name: "Mr. Person"
     email: "mrperson123@google.com"
   }
   recipients: [
     Document {
       name: "John Doe"
       email: "johndoe123@google.com"
     }
     Document {
       name: "Jane Doe"
       email: "janedoe123@google.com"
     }
   ]
   subject: "IMPORTANT"
   body: "Limited time offer!"
 }
 

Then, suppose that a query for "important" is issued with the following projection type property paths:

{schema: "Email", ["subject", "sender.name", "recipients.name"]}
 

The above document will be returned as:

Email: Document {
   sender: Document {
     name: "Mr. Body"
   }
   recipients: [
     Document {
       name: "John Doe"
     }
     Document {
       name: "Jane Doe"
     }
   ]
   subject: "IMPORTANT"
 }
 

addSearchStringParameters

public SearchSpec.Builder addSearchStringParameters (String... searchStringParameters)

Adds Strings to the list of String parameters that can be referenced in the query through the "getSearchStringParameter({index})" function.

addSearchStringParameters

public SearchSpec.Builder addSearchStringParameters (List<String> searchStringParameters)

Adds Strings to the list of String parameters that can be referenced in the query through the "getSearchStringParameter({index})" function.

build

public SearchSpec build ()

Constructs a new SearchSpec from the contents of this builder.

setDefaultEmbeddingSearchMetricType

public SearchSpec.Builder setDefaultEmbeddingSearchMetricType (int defaultEmbeddingSearchMetricType)

Sets the default embedding metric type used for embedding search (see AppSearchSession.search(String, SearchSpec)) and ranking (see setRankingStrategy(java.lang.String)).

If this method is not called, the default embedding search metric type is SearchSpec.EMBEDDING_SEARCH_METRIC_TYPE_COSINE. Metrics specified within "semanticSearch" or "matchedSemanticScores" functions in search/ranking expressions will override this default.

setJoinSpec

public SearchSpec.Builder setJoinSpec (JoinSpec joinSpec)

Specifies which documents to join with, and how to join.

If the ranking strategy is SearchSpec.RANKING_STRATEGY_JOIN_AGGREGATE_SCORE, and the JoinSpec is null, build() will throw an AppSearchException.

setListFilterHasPropertyFunctionEnabled

public SearchSpec.Builder setListFilterHasPropertyFunctionEnabled (boolean enabled)

Sets the LIST_FILTER_HAS_PROPERTY_FUNCTION feature as enabled/disabled according to the enabled parameter.

setListFilterQueryLanguageEnabled

public SearchSpec.Builder setListFilterQueryLanguageEnabled (boolean enabled)

Sets the LIST_FILTER_QUERY_LANGUAGE feature as enabled/disabled according to the enabled parameter.

setMaxSnippetSize

public SearchSpec.Builder setMaxSnippetSize (int maxSnippetSize)

Sets maxSnippetSize, the maximum snippet size. Snippet windows start at maxSnippetSize/2 bytes before the middle of the matching token and end at maxSnippetSize/2 bytes after the middle of the matching token. It respects token boundaries, therefore the returned window may be smaller than requested.

Setting maxSnippetSize to 0 will disable windowing and an empty String will be returned. If matches enabled is also set to false, then snippeting is disabled.

For example, maxSnippetSize = 16. "foo bar baz bat rat" with a query of "baz" will return a window of "bar baz bat" which is only 11 bytes long.

setNumericSearchEnabled

public SearchSpec.Builder setNumericSearchEnabled (boolean enabled)

Sets the NUMERIC_SEARCH feature as enabled/disabled according to the enabled parameter.

setOrder

public SearchSpec.Builder setOrder (int order)

Sets the order of returned search results, the default is SearchSpec.ORDER_DESCENDING, meaning that results with higher scores come first.

This order field will be ignored if RankingStrategy = RANKING_STRATEGY_NONE.

setPropertyWeightPaths

public SearchSpec.Builder setPropertyWeightPaths (String schemaType, 
                Map<PropertyPath, Double> propertyPathWeights)

Sets property weights by schema type and property path.

Property weights are used to promote and demote query term matches within a GenericDocument property when applying scoring.

Property weights must be positive values (greater than 0). A property's weight is multiplied with that property's scoring contribution. This means weights set between 0.0 and 1.0 demote scoring contributions by a term match within the property. Weights set above 1.0 promote scoring contributions by a term match within the property.

Properties that exist in the AppSearchSchema, but do not have a weight explicitly set will be given a default weight of 1.0.

Weights set for property paths that do not exist in the AppSearchSchema will be discarded and not affect scoring.

NOTE: Property weights only affect scoring for query-dependent scoring strategies, such as SearchSpec.RANKING_STRATEGY_RELEVANCE_SCORE.

setPropertyWeights

public SearchSpec.Builder setPropertyWeights (String schemaType, 
                Map<String, Double> propertyPathWeights)

Sets property weights by schema type and property path.

Property weights are used to promote and demote query term matches within a GenericDocument property when applying scoring.

Property weights must be positive values (greater than 0). A property's weight is multiplied with that property's scoring contribution. This means weights set between 0.0 and 1.0 demote scoring contributions by a term match within the property. Weights set above 1.0 promote scoring contributions by a term match within the property.

Properties that exist in the AppSearchSchema, but do not have a weight explicitly set will be given a default weight of 1.0.

Weights set for property paths that do not exist in the AppSearchSchema will be discarded and not affect scoring.

NOTE: Property weights only affect scoring for query-dependent scoring strategies, such as SearchSpec.RANKING_STRATEGY_RELEVANCE_SCORE.

setRankingStrategy

public SearchSpec.Builder setRankingStrategy (String advancedRankingExpression)

Enables advanced ranking to score based on advancedRankingExpression.

This method will set RankingStrategy to SearchSpec.RANKING_STRATEGY_ADVANCED_RANKING_EXPRESSION.

The ranking expression is a mathematical expression that will be evaluated to a floating-point number of double type representing the score of each document.

Numeric literals, arithmetic operators, mathematical functions, and document-based functions are supported to build expressions.

The following are supported arithmetic operators:

• Addition(+)
• Subtraction(-)
• Multiplication(*)
• Floating Point Division(/)

Operator precedences are compliant with the Java Language, and parentheses are supported. For example, "2.2 + (3 - 4) / 2" evaluates to 1.7.

The following are supported basic mathematical functions:

• log(x) - the natural log of x
• log(x, y) - the log of y with base x
• pow(x, y) - x to the power of y
• sqrt(x)
• abs(x)
• sin(x), cos(x), tan(x)
• Example: "max(abs(-100), 10) + pow(2, 10)" will be evaluated to 1124

The following variadic mathematical functions are supported, with n > 0. They also accept list value parameters. For example, if V is a value of list type, we can call sum(V) to get the sum of all the values in V. List literals are not supported, so a value of list type can only be constructed as a return value of some particular document-based functions.

• max(v1, v2, ..., vn) or max(V)
• min(v1, v2, ..., vn) or min(V)
• len(v1, v2, ..., vn) or len(V)
• sum(v1, v2, ..., vn) or sum(V)
• avg(v1, v2, ..., vn) or avg(V)

Document-based functions must be called via "this", which represents the current document being scored. The following are supported document-based functions:

• this.documentScore()

  Get the app-provided document score of the current document. This is the same score that is returned for SearchSpec.RANKING_STRATEGY_DOCUMENT_SCORE.

• this.creationTimestamp()

  Get the creation timestamp of the current document. This is the same score that is returned for SearchSpec.RANKING_STRATEGY_CREATION_TIMESTAMP.

• this.relevanceScore()

  Get the BM25F relevance score of the current document in relation to the query string. This is the same score that is returned for SearchSpec.RANKING_STRATEGY_RELEVANCE_SCORE.

• this.usageCount(type) and this.usageLastUsedTimestamp(type)

  Get the number of usages or the timestamp of last usage by type for the current document, where type must be evaluated to an integer from 1 to 2. Type 1 refers to usages reported by AppSearchSession.reportUsage, and type 2 refers to usages reported by GlobalSearchSession.reportSystemUsage.

• this.childrenRankingSignals()

  Returns a list of children ranking signals calculated by scoring the joined documents using the ranking strategy specified in the nested SearchSpec. Currently, a document can only be a child of another document in the context of joins. If this function is called without the Join API enabled, a type error will be raised.

• this.propertyWeights()

  Returns a list of the normalized weights of the matched properties for the current document being scored. Property weights come from what's specified in SearchSpec. After normalizing, each provided weight will be divided by the maximum weight, so that each of them will be <= 1.

• this.matchedSemanticScores(getEmbeddingParameter({embedding_index}), {metric})

  Returns a list of the matched similarity scores from "semanticSearch" in the query expression (see also AppSearchSession.search) based on embedding_index and metric. If metric is omitted, it defaults to the metric specified in SearchSpec.Builder.setDefaultEmbeddingSearchMetricType(int). If no "semanticSearch" is called for embedding_index and metric in the query, this function will return an empty list. If multiple "semanticSearch"s are called for the same embedding_index and metric, this function will return a list of their merged scores.

  Example: `this.matchedSemanticScores(getEmbeddingParameter(0), "COSINE")` will return a list of matched scores within the range of [0.5, 1], if `semanticSearch(getEmbeddingParameter(0), 0.5, 1, "COSINE")` is called in the query expression.

Some errors may occur when using advanced ranking.

Syntax Error: the expression violates the syntax of the advanced ranking language. Below are some examples.

• "1 + " - missing operand
• "2 * (1 + 2))" - unbalanced parenthesis
• "2 ^ 3" - unknown operator

Type Error: the expression fails a static type check. Below are some examples.

• "sin(2, 3)" - wrong number of arguments for the sin function
• "this.childrenRankingSignals() + 1" - cannot add a list with a number
• "this.propertyWeights()" - the final type of the overall expression cannot be a list, which can be fixed by "max(this.propertyWeights())"
• "abs(this.propertyWeights())" - the abs function does not support list type arguments
• "print(2)" - unknown function

Evaluation Error: an error occurred while evaluating the value of the expression. Below are some examples.

• "1 / 0", "log(0)", "1 + sqrt(-1)" - getting a non-finite value in the middle of evaluation
• "this.usageCount(1 + 0.5)" - expect the argument to be an integer. Note that this is not a type error and "this.usageCount(1.5 + 1/2)" can succeed without any issues
• "this.documentScore()" - in case of an IO error, this will be an evaluation error

Syntax errors and type errors will fail the entire search and will cause SearchResults.getNextPage(Executor, Consumer) to throw an AppSearchException with the result code of AppSearchResult.RESULT_INVALID_ARGUMENT.

Evaluation errors will result in the offending documents receiving the default score. For SearchSpec.ORDER_DESCENDING, the default score will be 0, for SearchSpec.ORDER_ASCENDING the default score will be infinity.

setRankingStrategy

public SearchSpec.Builder setRankingStrategy (int rankingStrategy)

Sets ranking strategy for AppSearch results.

setResultCountPerPage

public SearchSpec.Builder setResultCountPerPage (int resultCountPerPage)

Sets the number of results per page in the returned object.

The default number of results per page is 10.

setResultGrouping

public SearchSpec.Builder setResultGrouping (int groupingTypeFlags, 
                int limit)

Sets the maximum number of results to return for each group, where groups are defined by grouping type.

Calling this method will override any previous calls. So calling setResultGrouping(GROUPING_TYPE_PER_PACKAGE, 7) and then calling setResultGrouping(GROUPING_TYPE_PER_PACKAGE, 2) will result in only the latter, a limit of two results per package, being applied. Or calling setResultGrouping (GROUPING_TYPE_PER_PACKAGE, 1) and then calling setResultGrouping (GROUPING_TYPE_PER_PACKAGE | GROUPING_PER_NAMESPACE, 5) will result in five results per package per namespace.

setSearchSourceLogTag

public SearchSpec.Builder setSearchSourceLogTag (String searchSourceLogTag)

Sets an optional log tag to indicate the source of this search.

Some AppSearch implementations may log a hash of this tag using statsd. This tag may be used for tracing performance issues and crashes to a component of an app.

Call this method and give a unique value if you want to distinguish this search scenario with other search scenarios during performance analysis.

Under no circumstances will AppSearch log the raw String value using statsd, but it will be provided as-is to custom AppSearchLogger implementations you have registered in your app.

setSnippetCount

public SearchSpec.Builder setSnippetCount (int snippetCount)

Sets the snippetCount such that the first snippetCount documents based on the ranking strategy will have snippet information provided.

The list returned from SearchResult.getMatchInfos will contain at most this many entries.

If set to 0 (default), snippeting is disabled and the list returned from SearchResult.getMatchInfos() will be empty.

setSnippetCountPerProperty

public SearchSpec.Builder setSnippetCountPerProperty (int snippetCountPerProperty)

Sets snippetCountPerProperty. Only the first snippetCountPerProperty snippets for each property of each GenericDocument will contain snippet information.

If set to 0, snippeting is disabled and the list returned from SearchResult.getMatchInfos() will be empty.

The default behavior is to snippet all matches a property contains, up to the maximum value of 10,000.

setTermMatch

public SearchSpec.Builder setTermMatch (int termMatchType)

Sets how the query terms should match TermMatchCode in the index.

If this method is not called, the default term match type is SearchSpec.TERM_MATCH_PREFIX.

setVerbatimSearchEnabled

public SearchSpec.Builder setVerbatimSearchEnabled (boolean enabled)

Sets the VERBATIM_SEARCH feature as enabled/disabled according to the enabled parameter.