Metadata mapping

Incorporating metadata mapping helps improve how aiR Assist retrieves documents by allowing the AI model to use structured metadata as part of the retrieval process, in addition to semantic relevance. This leads to more accurate, context appropriate results when answering user questions.

• Without the metadata mapping and usage, aiR Assist often receives questions with constraints not clearly stated in the document, such as sender or recipient email information and date ranges.
• With metadata mapping enabled, aiR Assist can detect these elements in a natural language question and translate them into metadata filters at retrieval time. This ensures that documents considered for the answer align with the user’s intent, not just with similar wording.

aiR Assist uses a hybrid retrieval approach when metadata mapping is configured:

1. Metadata filters are applied first, based on detected entities such as emails and dates.
2. Semantic similarity search is then performed on the filtered set of documents.

By restricting semantic search to documents that already meet the relevant metadata criteria, aiR Assist helps to reduce noise and improves ranking quality. This approach is more precise than relying on semantic search alone. For more information on aiR Assist's usage of a Retrieval-Augmented Generation (RAG) architecture, see Understanding aiR Assist responses.

Without metadata filtering, semantically related but contextually incorrect documents (for example, the wrong time period or sender) may be included in retrieval. This can negatively impact answer quality. Metadata mapping helps ensure that aiR Assist reasons over documents that are valid for the question being asked, reducing the likelihood of incorrect or over generalized responses.

aiR Assist supports five pre-configured mapping fields: Primary Date/Time, Email From, Email To, Email CC, and Email BCC. New ones cannot be added.

Considerations

• Each workspace can have only one metadata mapping configuration and it applies to all indexes listed in Index Manager.
• Metadata mapping is only supported for indexes created using aiR Assist. It is not supported for the Case Home index, which is managed by aiR for Case Strategy.

Configuring metadata mapping

Workspace administrators can configure how standard metadata fields are mapped to their specific workspace-related fields, with field names and IDs uniquely assigned to each workspace.

1. Click the aiR Assist icon in the side bar to open its chat panel.
2. Select Metadata Mapping from the hamburger icon in the aiR Assist chat panel.
   
   The Metadata Mapping panel slides open next to the aiR Assist chat panel. To close the panel, click the X opposite the panel heading or click Metadata Mapping again from the hamburger icon.
3. Toggle Enable Metadata Mapping to set up and use metadata in the workspace. By default, the toggle is disabled.
   

   The Fields Mapped list is pre-configured.

4. For each relevant metadata field listed that you want to use, select the appropriate workspace-level field from the corresponding drop-down. The fields listed in the drop-down will vary between workspaces. The first three metadata fields must be set up: Primary Date/Time, Email to, and Email from. Maximum of five fields total can be configured. Labels appear next to each metadata field indicate the mapped or not mapped status.

   Field	Field type	Required field	Description
   Primary Date/Time	Date/Time	Required	Used to understand when document was created or sent in queries that require date/time filters.
   Email To	Long text	Required	Used to identify the email addresses to which the email was sent in queries pertaining to emails or specific participants.
   Email From	Fixed-length text	Required	Used to identify the email address that sent the email in queries that ask about emails or specific participants.
   Email CC	Long text	Optional	Used to identify the email addresses included as carbon copy (CC) recipients in queries pertaining to emails or specific participants.
   Email BCC	Long text	Optional	Used to identify the email addresses included as blind carbon copied (BCC) in queries that ask about emails or specific participants.

5. Click Save & update indexes.
6. Click Confirm to proceed with mapping metadata fields. This action automatically updates all indexes in the workspace.

The re-indexing process may take some time to complete. To view its progress, access the Index Manager. For more information, see Monitoring the index build process.

Audit logs capture this action. See Auditing user activity for more information.

Using metadata for prompting

The Use Metadata toggle displays in the prompting Ask a question box if Metadata Mapping is enabled. Each user can choose whether or not to leverage metadata for their inquiry by toggling Use Metadata on or off. See Prompting for details.