Working with Semantic Search
Semantic Search enables PeopleSoft Search Framework to return search results based on the meaning and context of a search query, rather than relying only on exact keyword matches. Semantic Search uses vector embedding to represent indexed text and user queries as numeric vectors.
OpenSearch compares these vectors to identify results that are semantically similar to the search text.
PeopleSoft uses Semantic Search with hybrid search. Hybrid search combines keyword search and vector search so that exact keyword matches continue to receive higher priority, while semantic matches improve result recall when users enter related terms or phrases.
For example, a keyword search may not return results when the indexed content does not contain the exact search phrase. With Semantic Search enabled, a query such as healthcare products can return related catalogue items such as bandages or gauze products, even when the exact phrase is not present in the indexed text.
In PeopleTools 8.63, Semantic Search is enabled by default for the Menu index, also known as PTPORTALREGISTRY. The delivered Menu index uses the title and description fields as vector fields and is enabled for Semantic Search. The delivered OpenSearch setup includes the default model setup required for the delivered Semantic Search configuration. You do not need to install a custom embedding model to use the delivered Semantic Search configuration.
Semantic Search can be enabled for other query and connected query search definitions as needed. Existing keyword search behavior remains available unless Semantic Search is configured for the search definition, the search definition is redeployed, and the index is rebuilt with vector embedding. If you do not want to use Semantic Search, you can disable it from the Semantic Search configuration.
Use Semantic Search for search definitions that contain meaningful descriptive text, such as menu labels, menu descriptions, catalogue item descriptions, product details, supplier descriptions, contract descriptions, or profile descriptions.
Semantic Search generates embedding during indexing. The selected vector fields are converted into vector embedding and stored in the OpenSearch index. During search, OpenSearch converts the user’s query text into an embedding and compares it with the indexed embedding.
Indexing is the most resource-intensive part of Semantic Search because embedding are generated and stored for indexed documents.
Semantic Search Configuration
Use the following configuration pages and related setup topics to configure and manage Semantic Search:
- General Settings: use this page to enable Semantic Search and define processing options, such as vector field count, batch size, vector scoring method, scoring value, and search pipeline. For more information, see Specifying Semantic Search General Settings.
- Embedding Models: use this page to view, import, deploy, and activate embedding models used to generate vector embedding for semantic-search-enabled indexes. For more information, see Managing Semantic Search Embedding Models.
- Search Attribute Mapping: use this page to select fields in a search definition that are used to generate vector embedding for Semantic Search. For more information, see Mapping Search Attributes.
- Advanced Index Settings: use this page to enable chunking for indexes that contain large text fields selected for vector embedding. For more information, see Working with Chunking.
- Dedicated ML Node Configuration: use this setup topic to configure a dedicated OpenSearch ML node for Semantic Search processing. For more information, see Configuring a Dedicated ML Node.
- Custom Embedding Model Installation: use this setup topic to register and deploy a custom embedding model in OpenSearch before importing and activating it in PeopleSoft. For more information, see Installing Custom Embedding Models.