Yuri Filimonov, of Improve the Web, and I were discussing how Google handles spell correction suggestions, how those are triggered, and how they might influence search results.
One of Google’s patent applications describes a process that may answer some of those questions, and Yuri and I thought it might be good to make the discussion public. I believe that Yuri may be describing some of the business issues around spell corrections later this week.
I wrote about that Google patent application which may be involved, in Google’s Query Rank, and Query Revisions on Search Result Pages.
My earlier post looked at the bigger picture of how the search engine might consider offering query refinements – usually based upon looking at users’ search sessions, and watching the different queries that those searchers use when looking for specific information. People may remove some of the words in a query when they only get a few results. They may add words if the results may seem too broad. That kind of user behavior is recorded in Google’s log files.
Google may also be looking at whether or not people click on any of the links returned in the sets of results from those searches, and how long people might stay on those pages.
One of the changes that is often recorded is when a searcher makes a spelling change to a term that they’ve typed in after the results they received might not match what they thought that they were searching for, and they realize that they made a spelling mistake.
The patent application tells us that the display of a query refinement may be triggered by the absense of relevant pages in the search page results.
Here are the steps that it suggests in returning spelling corrections as query refinements:
1. A search resulting in no, or just a few relevant results may trigger a look at possible query refinements
2. One kind of potential query refinement is offering a spell correction, based upon a score for a query (the misspelling in this case), and it’s relationship to a highly-ranked query (the correct spelling, possibly).
3. If the search engine thinks that the spelling correction offered is something a lot of people choose, it may start showing some results for the correct spelling at the top of the results page, in addition to offering a link to a search for the refined query.
That last step is interesting in that in addition to returning results for the misspelled word and a query refinement suggestion, Google will sometimes also display some highly placed results for what it believes are the correct spellings of a word. Keep in mind that it doesn’t do this for most misspellings, but it does for some.
Testing when Google Might Merge Results from Correct and Incorrect Spellings
To see that in action, I took some words from the 100 most commonly misspelled words list at yourdictionary.com.
In a search for the incorrect spelling, the top ten results in Google for me showed three results that only had the correct spelling, and three more that had both spellings.
None of the top ten results of a search for the incorrect spelling returned included the misspelling on their pages.
The first four results for the lightening search only had the “lightning” spelling. The rest of the results included two pages with both spellings, two more with only the incorrect spelling, and two others on the topic of “skin lightening.”
I tried searches on 20 of the misspellings from that list, and only received these three where the results for the correct spelling seemed to be mixed in with the incorrect spellings. But it is interesting to see how Google might change results like this when it has a lot of confidence that a word is misspelled.
It’s not unusual to create a new word when deciding upon a name for a business. It’s possible that the new word might be seen by a search engine as a misspelling of a word that it thinks it recognizes. What kinds of implications might that have for a business with a name like that?