Behind Search Improvements at Pinterest
February 13, 2015
As a Pinterest user myself, I know how important the site’s search function is. Now, as Gigaom informs us, “Pinterest Explains How It’s Making Its Search Work Better.” It sounds like an approach to semantic machine learning inspired by the crowdsourcing phenomenon. Writer Jonathan Vanian tells us:
“Dong Wang, the Pinterest software engineer who wrote the post, explained that even though a user may search for the word ‘turkey,’ it’s unclear what exactly that person may be looking for. Does he want to find turkey recipes, is he planning a trip to Turkey or is he just interested in poultry — it’s hard to say without some context.
“If that person decides to search for ‘turkey recipes’ as part of his next query, Pinterest takes that into account and can assume that the next person who may be searching for ‘turkey’ might also be craving some turkey recipes as well; maybe it’s holiday season and everyone’s hungry. Pinterest learned that ‘the information extracted from previous query log has shown to be effective in understanding the user’s search intent’ and this can be applied to other Pinterest users as well.”
Pinterest’s data-collection workflow is called QueryJoin, and engineers use it to draw conclusions like the one about turkey recipes, above. Factors analyzed also include data like pins’ image signatures and “engagement stats” like the number of clicks and re-pins it has received. For more information, see Dong Wang’s original post.
Cynthia Murrell, February 13, 2015
Sponsored by ArnoldIT.com, developer of Augmentext
Comments
-
- Subscribe to Beyond Search
Feature archive
News archive
-
