Specifying Semantic Search General Settings

Use the General Settings page to configure Semantic Search processing options.

Access the General Settings page by selecting PeopleTools, and then Search Framework, and then Semantic Search, and then General Settings.

This example illustrates the fields and controls on the General Settings page. You can find definitions for the fields and controls later on this page.

Field or Control	Description
Use Semantic Search	Select to enable Semantic Search. Search definitions that use vector fields must be redeployed and fully indexed before semantic results are available.
Vector Field Count	Enter the maximum number of fields that can be selected as vector fields in a search definition. The default value is 2. Increase this value only after evaluating CPU, memory, and storage capacity in the OpenSearch nodes. Dedicated ML nodes are recommended to improve processing speed and avoid the impact to regular search operations to the Master/Data nodes.
Batch Size	Enter the number of documents sent to the search server in a batch when Semantic Search is enabled. The default value is 300. Smaller batches help reduce memory and CPU pressure during vector indexing.
Vector Scoring Method	Select the method used to retrieve semantic matches. Available values are: kNN Max Distance Minimum Score Only one scoring method can be selected at a time. For more information on Vector Scoring Method, see the Vector Scoring Method topic below.
Value	Enter the value for the selected vector scoring method: For kNN: enter the number of nearest results to retrieve. For Max Distance: enter the maximum vector distance from the query vector. For Minimum Score: enter the minimum similarity score required for results to be returned.
Search Pipeline	Enter the OpenSearch search pipeline used to normalize and combine keyword-search scores and semantic-search scores for hybrid search. PeopleTools deploys the ptsf-search-pipeline search pipeline during search definition deployment.

Vector Scoring Methods

Semantic Search supports the following vector scoring methods. PeopleSoft uses hybrid search, which combines and normalizes the scores from keyword and neural search queries. The specified scoring method applies only to the neural query. The normalization process adjusts the score after receiving the individual scores from both queries.

Scoring Method	Description
kNN	Returns the top k nearest vector matches for the specified search query. Use this method when Semantic Search must retrieve a fixed number of results based on vector proximity. With the kNN method, the search keyword is converted into a vector representation, and the search engine identifies the k closest indexed vectors according to the configured similarity metric. Because the results are selected based on nearest-neighbor distance rather than exact semantic equivalence, some returned items may have lower semantic relevance. However, higher-ranked results represent the closest vector matches. PeopleSoft uses kNN as the default retrieval method, with k set to 50 by default.
Max Distance	Returns all vector matches within the specified maximum distance from the search query vector. Use this method when Semantic Search must retrieve only results whose vector distance is within a defined threshold, rather than returning a fixed number of nearest matches. A lower max_distance value applies a stricter similarity threshold and returns fewer, more closely related results. A higher max_distance value broadens the search radius and may return more results, including results with lower semantic relevance. The appropriate value depends on the embedding model, vector normalization, vector dimension, and configured OpenSearch space type. The value must be greater than 0.
Minimum Score	Returns vector matches whose calculated similarity score is greater than or equal to the specified minimum score. Use this method when Semantic Search must return only results that meet a required relevance threshold, rather than a fixed number of nearest matches or all matches within a maximum distance. With the min_score method, the search keyword is converted into a vector representation, and OpenSearch compares the vector with indexed vectors using the configured vector space. OpenSearch returns only results with a calculated score greater than or equal to the configured min_score value. A higher min_score value applies a stricter relevance threshold and returns fewer, more strongly matching results. A lower min_score value broadens the result set and may return more results with weaker semantic relevance. The appropriate value depends on the embedding model, vector normalization, vector dimension, and configured OpenSearch space type.

Scoring Method

Description

kNN

Returns the top k nearest vector matches for the specified search query. Use this method when Semantic Search must retrieve a fixed number of results based on vector proximity. With the kNN method, the search keyword is converted into a vector representation, and the search engine identifies the k closest indexed vectors according to the configured similarity metric. Because the results are selected based on nearest-neighbor distance rather than exact semantic equivalence, some returned items may have lower semantic relevance. However, higher-ranked results represent the closest vector matches. PeopleSoft uses kNN as the default retrieval method, with k set to 50 by default.

Max Distance

Returns all vector matches within the specified maximum distance from the search query vector. Use this method when Semantic Search must retrieve only results whose vector distance is within a defined threshold, rather than returning a fixed number of nearest matches.

A lower max_distance value applies a stricter similarity threshold and returns fewer, more closely related results. A higher max_distance value broadens the search radius and may return more results, including results with lower semantic relevance. The appropriate value depends on the embedding model, vector normalization, vector dimension, and configured OpenSearch space type. The value must be greater than 0.

Minimum Score

Returns vector matches whose calculated similarity score is greater than or equal to the specified minimum score. Use this method when Semantic Search must return only results that meet a required relevance threshold, rather than a fixed number of nearest matches or all matches within a maximum distance.

With the min_score method, the search keyword is converted into a vector representation, and OpenSearch compares the vector with indexed vectors using the configured vector space. OpenSearch returns only results with a calculated score greater than or equal to the configured min_score value.

A higher min_score value applies a stricter relevance threshold and returns fewer, more strongly matching results. A lower min_score value broadens the result set and may return more results with weaker semantic relevance. The appropriate value depends on the embedding model, vector normalization, vector dimension, and configured OpenSearch space type.

Semantic Search can return broader results than keyword search because it matches meaning rather than exact terms. Tune the vector scoring method and value when semantic results are too broad or include weak matches.