Instead of relying solely on text matching within titles and descriptions, the system also uses embeddings, just like an AI large language model (LLM). This breaks content into small sections (or "chunks") that capture their meaning to provide relevant results even when the exact words aren’t matched.
For example, searching for "how to lead a team" might now surface content about "people management" or "leadership skills" even if those exact words weren’t used. This is called 'semantic similarity'.
What Search Looks For
Search will look for matches in:
- Learning experience titles
- Learning experience short and long descriptions
- Learning experience object titles and descriptions
- Supported resource content types (If Advanced Search is enabled. Available in Professional tier and above only)
Search queries will match on:
- Key words - Matches individual important words from the search query.
- Key phrases - Matches exact or near-exact multi-word phrases from the search query for more precise results.
- Contextual embeddings - Uses AI to understand the meaning behind the search and match it to semantically similar content.
- Fuzzy matching - Finds results even when there are minor typos or spelling variations in the search query.
Result visibility
Search will only display content that the learner has permission to access. When a search query is entered, the system checks access rules - such as enrolment, audience settings, or locked levels - to make sure learners only see relevant results.
How results are ranked
Search results are ranked, or ordered, using a combination of relevance scoring techniques:
-
Keyword Relevance: Matches keywords to the content and weighs terms based on how often they appear and how rare they are across all documents.
- For example: A rare word in the search query gets more weight than a common one, as it is considered more significant:
- When searching for something like 'Health and Safety in the Workplace', the words 'Health', 'Safety' and 'Workplace' get more weight than 'and', 'in' and 'the'.
- For example: A rare word in the search query gets more weight than a common one, as it is considered more significant:
-
Exact Matches vs. Fuzzy Matches: Exact matches are boosted over fuzzy matches.
- For instance, if you search for the word “run,” a document with an exact match for “run” scores higher than one with a similar word like “running,” since exact matches are more indicative of a strong query alignment.
-
Field Boosting: Certain fields within the document are considered more important and are thus given extra weight.
- For example:
- Learning Experience Name: This is boosted the most because it represents the primary identifier or title.
- Short Description: This is given a moderate boost as it provides a concise overview.
- Long Description and Content: These are scored as well, but with lower priority compared to the Learning Experience name and short description.
- For example:
Search Sync Timing
If new content is added or changes are made to existing learning, it may take up to 5 minutes for those updates to appear in search results. This short delay allows the system time to index the content and ensure accurate matching.