Ever wonder how Bing retrieves the searched query results in a flash? Well, Microsoft wants developers to learn this technology and implement it for their users who have to search for the required item among a heap of data in their domains. Thus, Microsoft has announced that it has already open-sourced a crucial part of how its Bing Search Services are able to swiftly retrieve the search results.
The crucial part in question includes a Microsoft library created for utilizing the collected data and AI models created specifically for Bing.
Microsoft brought up how web search has evolved with time. Earlier, users were comfortable with typing a few words and searching through loads of data to get the desired result. Now, they would prefer getting the results without putting in effort for example, dropping a picture in the search box or asking an intelligent assistant to do the work for them or simply expecting to receive a to-the-point answer to their question instead of multiple results.
To pertain to the needs of users, Microsoft makes use of Space Partition Tree and Graph (SPTAG) algorithm that enjoys sitting at the core position of the open-sourced Python library. Thus, Microsoft traverses billions of information pieces in less than a second to get you your query results.
This in no way means that Vector search is a new discovery. It was already in action earlier. However, Microsoft can be credited for coupling it with deep learning models. This involves encoding a pre-trained model data into vectors with each vector being a representation of a single word or pixel.
Enter the SPTAG library; a vector index is then generated. With every incoming query, the queried text or image is translated into a vector and the library then looks up for the most relevant vector in the index.
Microsoft claimed that the above mentioned approach now impacts over 150 billion pieces of information. This is surely a step up from the outdated matching of keywords. Words, characters, web full queries…you name it! And Bing can scan the indexed vectors and provide you the best-matched results in a matter of milliseconds.
The open-sourced library is licensed under MIT and boasts all the tools required in the building and searching of the distributed vector indexes. You are encouraged to study it and learn the offered concepts.
Featured photo: Christophe Morin/IP3 / Getty Images
Read next: Goodbye Bing Ads, Hello Microsoft Advertising! Focus on AI and personalization leads to Major Re-branding!
The crucial part in question includes a Microsoft library created for utilizing the collected data and AI models created specifically for Bing.
Microsoft brought up how web search has evolved with time. Earlier, users were comfortable with typing a few words and searching through loads of data to get the desired result. Now, they would prefer getting the results without putting in effort for example, dropping a picture in the search box or asking an intelligent assistant to do the work for them or simply expecting to receive a to-the-point answer to their question instead of multiple results.
To pertain to the needs of users, Microsoft makes use of Space Partition Tree and Graph (SPTAG) algorithm that enjoys sitting at the core position of the open-sourced Python library. Thus, Microsoft traverses billions of information pieces in less than a second to get you your query results.
- Also rad: Microsoft's Recent Stats regarding LinkedIn Reveal Remarkable Improvements in the Platform!
This in no way means that Vector search is a new discovery. It was already in action earlier. However, Microsoft can be credited for coupling it with deep learning models. This involves encoding a pre-trained model data into vectors with each vector being a representation of a single word or pixel.
Enter the SPTAG library; a vector index is then generated. With every incoming query, the queried text or image is translated into a vector and the library then looks up for the most relevant vector in the index.
Microsoft claimed that the above mentioned approach now impacts over 150 billion pieces of information. This is surely a step up from the outdated matching of keywords. Words, characters, web full queries…you name it! And Bing can scan the indexed vectors and provide you the best-matched results in a matter of milliseconds.
The open-sourced library is licensed under MIT and boasts all the tools required in the building and searching of the distributed vector indexes. You are encouraged to study it and learn the offered concepts.
Featured photo: Christophe Morin/IP3 / Getty Images
Read next: Goodbye Bing Ads, Hello Microsoft Advertising! Focus on AI and personalization leads to Major Re-branding!
