Experience Training | Yext Hitchhikers Platform
What You’ll Learn
In this section, you will learn:
- Experience Training Overview
- Experience Training for Featured Snippets
- Experience Training for Spell Checking
- Experience Training for NLP Filters
Experience Training is a great way to improve the quality of search results. It allows admins to train the Search Algorithm by providing it feedback on its predictions. With your input on a specific search term, you can fix that query instantly and help Search improve the relevancy of results over time. It is a critical tool for debugging backend issues and optimizing results for all search experiences. The more feedback that you provide or corrections that you apply here, the better the algorithm can perform in the future.
Experience Training can be found within the Search > All Search Experiences > View Experience of your Yext account, or via Configuration as Code (CaC). To learn more about the Config resource schema, you can find the documentation here .
Experience training allows you to train the Search algorithm in three areas:
- Featured Snippets
- Spell Checking
- NLP Filters
As you learned about in the Core Configuration - Direct Answers unit , Featured Snippets extract direct answers to users’ questions from unstructured data. The experience training tab contains the full, unique set of all Featured Snippets produced with Document Search.
Within the UI, you have the ability to approve, reject, or change the Featured Snippets in the experience. With any action you take on featured snippets, you are providing feedback so the algorithm can get smarter over time. In addition, it also corrects queries:
- If you reject a Featured Snippet, it will suppress it from the results set and never return that snippet back to the end user.
- If you approve a Featured Snippet, that result will then display on search results if you have snippets set to
- If you change a Featured Snippet, the modified snippet will show in future search results.
Below we walk through several key features of the Featured Snippets Experience Training table:
Deduplication: The Featured Snippets table dedupes on both search term and featured snippet to show one row per featured snippet per search term. There can be multiple Featured Snippets for the same search term for two reasons:
- The results set can change (through changes in the Knowledge Graph or in the configuration)
- The algorithm gets retrained as it gets more feedback.
Users can see all the different predictions the algorithm has made for a particular search term.
Entity Information: See the entity (hyperlinked) and field that the featured snippet came from, which is crucial context to determine whether or not a featured snippet is correct.
New vs. Completed: Toggle to see the featured snippets that have already been approved, rejected, or modified. This makes it easier for users to find completed training examples.
Number of Searches: See the number of searches for this search term in the past 30 days (not the number of times this particular snippet was displayed) to prioritize search terms that have been searched more often.
Last Searched: See the most recent search time for each search term. The table is sorted by this column by default, and now you’ll be able to see it explicitly.
Training Actions: Here you have the ability to approve (checkmark button), reject (x button), or change (wand button) featured snippets suggested by the algorithm.
Featured Snippets Modal
When you click on the change wand button, you will see a pop-up modal that will allow you to approve, reject, or change featured snippets. If the AI algorithm identifies the incorrect portion of text from your entity, you can select a different portion of text to be used as the featured snippet for the query going forward.
The modal will automatically expand the highlighted section to only include full words. The right side of the modal shows the original versus updated answers.
When changing a rich text snippet, you will select the text you want to appear in the results by individually clicking the paragraphs or sections in sequential order. For all other text snippets, you can click and drag your mouse across the text you want returned.
Here you can review any suggested spell checks and approve or reject the corrections. If you reject a correction, the algorithm will no longer surface that correction when that search term is run for that experience. You can also toggle to see spell check suggestions that have been approved or rejected.
Here you can review the NLP filters that were applied to any queries run. You can then choose to approve or change the filters.
Approving NLP filters one helps to train the algorithm to better understand your brand and preferences.
By choosing to change filters, you can add, remove, or select different filters to be applied for a specific query. Options include selecting filters by vertical, field, and value for any given search term.
For example, when a user searches “jobs in New York” the algorithm may apply the filters
Entity Type = Jobs and
Location = York. While the location of York may be implied since it is part of the query, users are actually looking for jobs in the New York area. You will be able to override this result by selecting the proper vertical (i.e., Jobs) in the modal and then editing the
Location filter to be near