Misspelling Detection

The Misspelling Detection feature maps misspellings to their corrected spellings. When Fusion receives a query containing a known misspelling, it rewrites the query using the corrected spelling in order to return relevant results instead of an empty or irrelevant results set.

Spelling corrections are applied in the Text Tagger stage of the query pipeline.

The Token and Phrase Spell Correction job automatically creates spelling corrections based on your AI-generated data. When you navigate to Relevance > Query Rewriting > Misspelling Detection, you can review or edit the output from the job and manually add new spelling corrections. Your changes remain in the _query_rewrite_staging collection until you publish them.

Tip
When you manually add new spelling corrections, subsequent job runs use those documents as input for machine learning to improve the job’s output. Unlike job-generated documents, manually-added query rewriting documents are never overwritten by new job output.
Tip
Misspelled terms are completely replaced by their corrected terms. To instead expand the query to include all alternative terms, see the Synonym Detection feature and set your synonyms to be bi-directional.

Misspelling Detection screen

Reviewing auto-generated spelling corrections

Spelling corrections that are automatically generated by the Token and Phrase Spell Correction job are assigned one of these Status values:

  • Auto

    These results have a sufficiently high confidence level to automatically deploy them to the _query_rewrite collection.

    No action is required on these results, though you can edit them if you wish.

  • Pending

    The confidence level is ambiguous, and the result must be reviewed by a user before it can be deployed. It will only be moved from the query_rewrite_staging collection to the _query_rewrite collection when its status has changed to "Approved" _and it has been published.

    See below for instructions.

How to review a pending spelling correction result
  1. Navigate to Relevance > Query Rewriting > Misspelling Detection.

    Tip
    Notice the Status facet on the left. Click Pending to view only the items that need review.
  1. Click the Edit icon icon next to the spelling correction.

  2. In the Status column, select either "Approved" or "Denied".

    Optionally, you can also edit the spelling correction itself.

    Tip
    Although the Confidence field is also editable, changing its value makes no difference.
  3. Click the Close icon next to the updated spelling correction:

    Close a spelling correction

Note
Approving a spelling correction does not automatically deploy it to the _query_rewrite collection. When you have finished your review, you must click Publish to deploy your changes.

Adding new spelling corrections

In addition to the spelling corrections generated by the Token and Phrase Spell Correction job, you can manually add your own.

How to add a spelling correction
  1. Navigate to Relevance > Query Rewriting > Misspelling Detection.

  2. At the bottom of the rules list, click the Add icon icon.

    A new spelling correction appears at the top of the list:

    Add a spelling correction

  3. Enter the misspelled word or phrase.

  4. Enter one or more spelling corrections.

    Tip
    It’s not necessary to set a confidence value.
  5. Select the spelling correction’s status, depending on whether you want to deploy it the next time you publish your changes ("Approved") or save it for further review ("Pending").

  6. Click the check mark to save the new spelling correction:

    Save a spelling correction

Publishing your changes

How to publish updated spelling corrections
  1. In the Misspelling Detection screen, click the PUBLISH button.

    Fusion prompts you to confirm that you want to publish your changes.

  2. Click PUBLISH.

Tip
You can un-publish a query rewrite by changing its status to "pending" or "denied", then clicking PUBLISH.

Tuning the misspelling detection job

The default configuration for the Token and Phrase Spell Correction job is designed for high accuracy and works well with most signal datasets, depending on the volume and quality of the signals. If you are seeing too few results, or too many inaccurate results, then you can try tuning the job to achieve better results.

Tip
To modify job configurations, you must be a Fusion user with the admin or developer role, or custom permissions that include access to job configurations.