Google Autocorrect: Smart Example of Collaborative Filtering

RSS Archive

P2P By Design provides radical design thinking for brand builders and business leaders. It is dedicated to showing how design
can help businesses adapt to the peer-to-peer world rising from our networked culture. It takes a radical view, because, in this new world, the only path closed to us is the one where we turn back.

Preferred Posts P2P Examples Ask me anything

August 15, 2012

This is one of my favorite examples collaborative filtering. It comes from Google.

To understand it, it helps to know how Microsoft developed its solution to a misspelled search term. For the word “appropriate,” for example, Microsoft developed a long list of possible misspellings: apropriate, approriate, appropriate, aprpiate, etc. (I’m guessing they just used a algorithm to do so.) Whenever a user typed in one of those misspellings into Microsoft Search Engine, the system knew they were trying to search “appropriate” and suggested the proper spelling. It’s a logical approach. But when large dictionaries carry 400,000 words, creating such lists is a cumbersome task. Google’s approach was different. Its engineers noticed that when a person accidentally searches for a misspelled term their next move is to search the correctly spelled term. As such, Google recognized that it didn’t need to generate lists, it let the user base’s natural search behavior generate the list without them even knowing it. All Google did was present the next typical search term as a simple visual—”Did you mean: appropriate?” Google found a way to use the digital exhaust from its searches to create a lightweight solution to a user problem.