Microsoft Bing makes every feature binary (SEM)
Microsoft announcement their next AI model, a large-scale sparse model that complements our production Transformer models, they call MEB or â€œMake Every Feature Binaryâ€. Microsoft said this makes search results on Bing more relevant.
In fact, MEB works in production for 100% of Bing searches, in all regions and all languages.
Comment, well, Microsoft said “To make research more precise and dynamic, MEB better harnesses the power of Big Data and enables an input feature space with over 200 billion binary features that reflect the subtle relationships between search queries and documents “. Search Engine Land’s George Has An Easier To Understand Version Of It All this way.
Microsoft said that MEB attributes each fact to a characteristic, “so it can assign weights that distinguish the ability to fly, for example, a penguin and a puffin. It can do this for each of the characteristics that make up a bird – or any entity Instead of saying “birds can fly”, the SEM associated with Transformer models can take this to another level of classification, saying “birds can fly, except ostriches, penguins and these other birds “.
How smart is MEB? Well, he’s been through a lot of training. SEM is formed with over 500 billion query / document pairs from three years of Bing research. The input features space has more than 200 billion binary features, Microsoft said.
Microsoft said the MEB model also demonstrates interesting capabilities for learning beyond semantic relationships. By examining the main functionalities learned by MEB, Microsoft discovered that it could learn hidden intents between the query and the document.
It matters and improves research, Microsoft shared these data points:
- A nearly 2% increase in click-through rate (CTR) on early search results. These results found “above the fold” without the need to scroll down.
- Reduction of manual query reformulation of more than 1%. Users who need to manually rephrase queries mean they didn’t like the results they found with their original query.
- A reduction in clicks on the pagination of more than 1.5%. Users who have to click on the “next page” button mean that they did not find what they were looking for on the first page.
Want to have fun about it? Read the Microsoft blog post.
And of course, I still have to share my pun, even though it wasn’t picked up in the final cut. https://t.co/6UckEEeh3f
– FrÃ©dÃ©ric Dubut (@CoperniX) August 4, 2021
Discussion forum at Twitter.