Elastic Search

Search Guidelines

Content can be returned in the order that most closely matches the user’s search terms!  In order to start to understand exactly what we mean by a relevance, the information below will highlight those factors that contribute to this determination, as well as give some general recommendations on how to structure content to be easily searchable for your users.

What determines relevance?

• Content titles, descriptions and tags are used to match search terms.
  • This means that text within a piece of content, or the names of content within Lessons, Learning Paths, or Scenarios will not currently have any impact on the search results. 

The following aspects of those matches on those fields will help determine which are more relevant than the other. 

• Content where a search term shows up multiple times is considered more relevant.
  • The more search terms a document matches, and the more times those terms match, the more relevant a document is. For example, if you are searching for the term “LXP”, a document with the title “LXP Guide”, a description of “An introduction to LXPs (Learning Experience Platforms), and the tag “LXP” is likely to be considered relevant since it matches all three fields of the content.
• Content matching “rarer” words are considered more relevant.
  • For example, imagine a search with the terms “Maestro learning content.” In the system, there might be many more documents with the terms “learning” and “content” in them.  For this reason, “Maestro” would be considered a “rarer” term, meaning that content with Maestro in its title, description, or tags would be considered more relevant.
• If a term matches a shorter field, that match is considered more relevant.
  • We will use the example of the “Maestro learning content” search again.  Imagine there are two pieces of content, one titled “Maestro Learning Development Guide” and another piece of content with the description “This presentation instructs Maestro employees on best practices for developing learning content. We recommend that…” In this example, while the description of the second content matches all three terms, the first content benefits from the fact that its title is much shorter.  Content #1 matches 50% of all of the words in the title where content #2 matches only a small fraction of the all of the words in that description.  In this sense, when a search term matches closely with a shorter field, like a tag, that has a large impact on the relevancy of the content.

How should I structure/name my content in Loop?

• Take advantage of tags
• • If you want to have a piece of content show up with a specific search term, adding that term as a tag will ensure that content is in the search results.  In this way, you can improve search functionality without having to rename or re-describe existing content.
  • Tags are shorter fields, so they have a large impact on the relevancy of a document.  Having a tag for the name of the product a piece of content is about has the potential to have more impact than having that product name in the title or description, as those tend to be longer fields.
• Avoid long titles
  • As mentioned earlier, longer fields have less impact as search terms are less likely to match a large percentage of that field.  If a document is often searched using a specific set of search terms, the relevancy of the document can be improved by making sure all of those terms are in the title of a document and that extraneous words.  For example, the title “Basic information clients who will work with Maestro” is probably less helpful than “Maestro: Basic information for clients”.  Searching for “Maestro client information” would recommend the second document first.
• Avoid too short titles
  • Similarly, if I have a document simply called “Maestro”, when searching for “Maestro client information”, that document is likely to be considered very relevant as my terms match 100% of the title.  While it is advantageous to have specific and brief titles and descriptions, making titles or descriptions too generic can also lead to unhelpful results.
• Special characters are ignored
  • When searching text, special characters are treated as a space character.  For example, searching for “anti-coagulant” is the same as searching for “anti coagulant”.  For this reason, when tagging a content with the tag “anti-coagulant”, you may also consider tagging it “anticoagulant” so both spelling variants would match the content.  Additionally, because special characters are ignored, having trademark symbols at the ends of words should not impact the search-ability of content.
• Exact word matches are preferred over partial word matches
  • One specific example we were asked about was if the engine would recognize the difference between a search for “LEAD” vs. “leadership” vs. “ERLEADA”.  The short answer is yes.  If you search for “LEAD” documents with that acronym or the word “lead” will be considered more relevant than documents where lead matches part of the word such as “ERLEADA” or “leadership.”
• Spelling variations will not always be recognized by the search engine  (acute vs acutely or prescribed vs prescribing)
  • As Loop is a system used across multiple languages, before we can support that kind of searching, we need to collect better information on what languages documents are in, and what language(s) a user may be searching in.  This is something we are working on and is likely to be supported in future releases.

Closing Notes

As search is a new functionality for Loop, we recognize that our search algorithm is not going to be perfect on its first iteration.  To start, relevancy searching will only be available from the user side of the application (content browse/global search).  Admin use cases are more complicated and will be implemented over time.  As users begin to use this new search functionality, we hope to learn more about where the algorithm can be improved.  Tags tend to be shorter fields and it is common for a search term to match multiple tags, so it is possible that tags will end up being weighted too highly in our search results.  On the other end of the spectrum, we might find that descriptions do not impact the relevance of content enough.  This fine tuning will happen over time as we receive feedback from our customers.  Therefore, this documentation is a starting point for search recommendations.  While the core aspects used in determining relevance are unlikely to change (number of matches, rarity of a term, and the length of a field), we will be continuing to adjust the algorithm for search results.  As we roll out this feature, your feedback on searches that work well vs those that don’t will be critical in improving those results.