KB User's Guide - Semantic Search FAQ
- How does the new hybrid semantic search function, and in what ways does it differ from the original search engine??
- What makes the new search engine a hybrid search engine?
- Are there certain types of searches where semantic search won't perform as well?
- How do I enable semantic search for my site?
- Can I disable semantic search for my site once it has been enabled?
- How can I tell if semantic search is enabled for a site I am searching?
- Can a user switch back to the basic search if semantic search is enabled on a site?
- Will changing the search mode dropdown have an effect on my results?
- How can I report a possible issue with the semantic search engine?
- Do I need to change my existing documents to make them work with semantic search?
- See Also
How does the new hybrid semantic search function, and in what ways does it differ from the original search engine??
For an overview of how the two search engines compare, please see KB User's Guide - Understanding the New Semantic Search Engine.
What makes the new search engine a hybrid search engine?
When semantic search is enabled, the first set of results are a combination of semantic results and keyword results. By default, this first set will include up to 20 results, though this limit can be adjusted site-by-site.
When the results of the two engines are combined, any duplicates are removed. From there, the combined set of results are sorted for relevance using the following criteria (in order):
- Search priority ("top" priority docs are returned first, followed by "high", "normal", and "low" priority)
- Device matching (if the client OS / browser sort option is enabled, results matching the users device are prioritized)
- Keyword similarity (results in the combined set whose keywords match the search terms are prioritized)
Any additional docs matching the site's default search mode that were not included in the initial set of results, e.g. those that have exact matches in the title and keywords fields, will then be appended.
Are there certain types of searches where semantic search won't perform as well?
In general, semantic search will outperform traditional keyword- and title-based search because it can understand the meaning and context behind a query rather than relying solely on exact word matches. However, there are some scenarios where its performance may be more limited.
One such case involves searches that use descriptive language for services, tools, products, or other specialized terms that are unique to an organization and not widely recognized in standard language models.
Example: a user searching for "online discussion" may not automatically receive results related to the Canvas learning management system. While the large language model (LLM) behind the semantic search engine understands the common meaning of the word canvas, it may not recognize that Canvas refers to a specific course platform within our environment.
In practice, the impact of this limitation is fairly small because the search engine still incorporates traditional keyword matching. If a user searches for "Canvas online discussion," documents containing the term Canvas in their titles or keywords will be ranked appropriately. Likewise, documents about Canvas that include words such as online and discussion—or related concepts—can still be surfaced by semantic search.
Acronyms present a similar challenge. Searches that contain acronyms generally depend on those acronyms appearing in document titles, keywords, or content. For this reason, it is considered a best practice to spell out acronyms at least once within a document.
Example: if a document discusses the Cloud Computing Infrastructure (CCI) service, the full service name should appear somewhere in the content rather than relying exclusively on the acronym CCI. Doing so helps ensure that users searching for related concepts, such as "cloud hosting," can still discover the document even if they are unfamiliar with the acronym.
To further improve search accuracy for these types of queries, we have introduced a Unique Terms table. This table maps specialized terminology, descriptive phrases, and expanded acronyms to the concepts they represent. By providing additional context to the semantic search engine, it helps bridge gaps between user queries and organization-specific language, reducing the likelihood that relevant results will be missed.
How do I enable semantic search for my site?
You can enable the semantic search engine for your internal and/or external site by following the steps outlined in KB User's Guide - Enabling Semantic Search for Your KB Site(s).
Can I disable semantic search for my site once it has been enabled?
Yes, if you decide that you want to switch back to the traditional keyword+title search, you can disable it at any time using the above instructions.
How can I tell if semantic search is enabled for a site I am searching?
To avoid confusion for end users, we do not have any explicit indicators that semantic search is enabled. However, you can generally tell by observing one or both of the following:
- A larger number of search results, with some results lacking exact search terms in the title and keywords
- A slightly different search success/failure message (does not apply if you have customized this message)
- Standard message: "Not finding what you are looking for? Suggest a new document be created, or try adjusting your search criteria."
- Semantic search message: "Not finding what you are looking for? Suggest a new document be created, or try searching again with multiple search terms."
Can a user switch back to the basic search if semantic search is enabled on a site?
At this time, end users cannot disable semantic search while searching.
Will changing the search mode dropdown have an effect on my results?
Currently, interacting with the "search mode" dropdown will only affect the second set of results, i.e. those that load after the initial set of semantic search results.
How can I report a possible issue with the semantic search engine?
If you are getting unexpected search results from the semantic search engine, or if you believe it is otherwise not functioning as expected, please contact us at kb-team@doit.wisc.edu.
Do I need to change my existing documents to make them work with semantic search?
No. Existing documents will continue to work with semantic search without requiring any changes.
However, following a few content-authoring best practices can help improve search relevance and make it easier for users to find the information they need:
- Use clear structure and formatting. Organize content with headings, subheadings, lists, and other formatting elements to make key information easier to identify and understand.
- Write in plain language. Whenever possible, avoid jargon, internal terminology, and overly technical language that may be unfamiliar to users.
- Spell out acronyms at least once. Include the full term before introducing an acronym (for example, Cloud Computing Infrastructure (CCI)) to help both users and the search engine understand the content.
- Keep documents focused and manageable. Rather than creating a single, lengthy document that covers many topics, consider breaking content into smaller, topic-specific documents when appropriate.
While these recommendations are not required, they can improve both the user experience and the effectiveness of semantic search.