Search engine giant Bing has tapped into artificial intelligence to improve the accuracy of its image search results, imbuing the search engine with a number of techniques.
Techniques that allow the service to better handles searches for pictures that have specific context or attributes.
Microsoft outlined these improvements in a blog post, where the Bing Image Relevance team dived deep into how it is using AI and machine learning to measurably improve the accuracy of the results that are served up on the popular image search engine.
“Our image search is evolving further toward a more intelligent and more precise search engine through multi-granularity matches, improved understanding of user queries, images and webpages, as well as the relationships between them. Deep learning techniques are a set of very exciting and promising tools lending themselves very well to both text and image.”
One such tool goes by as vector matching.
It maps queries and documents to semantic spaces in order to help find more relevant results. An attention mechanism models relationships among words and embedded images and pages with an awareness of each other.
As a result, the aforementioned documents become stronger summarizations of the salient areas of the photos and pages.
Another approach recently applied to image search on Bing is attribute match, which extracts a select set of attributes from both query and candidate documents, and then uses these attributes for matching and serving the results.
You can read up on the full technical details at the link above, they are a fascinating read.