NGramdata and Pinterest Image Search

Pinterest allows users to visually share- and discover new interests by posting images or videos.

“…there are more than 3 billion boards with 175 billion Pins…” (Hui Xu, Head of the Pinterest Discovery team)

Searching for images on Pinterest poses interesting semantic problems:

A search for “turkey filled with a duck stuffed with chicken for thanksgiving dinner” does not return a single result… but “turducken thanksgiving dinner” results in multiple returns.

Quality of matched search returns falls drastically with an increase in the number of keywords.

With NGramdata’s mix of word combinations, long input queries using different grammar structures with equivalent semantic meaning can be translated into powerful searches to yield result-optimised search returns:

“turkey filled with a duck stuffed with chicken”, is powerfully truncated to semantically equivalent “turducken”.

NGramdata Pinterest Turducken

NGramdata punches the stuffing out of efficient online search