Search engines, which have become much smarter than before, have now become more than just tools that match relevant keywords. For example, in the search engine “How tall is the tower in Paris?” When you finish, he tells you that the Eiffel Tower is an 81-storey building with a height of 324 meters. In other words, it is possible to reach this answer even if you do not type the name of the tower in the search box.
Search engines can do this using machine learning. Machine learning algorithms; It uses vectors to show input data regardless of whether there is text in web pages, images, audio, or videos. Bing captures billions of vectors for all different types of indexed media. Microsoft uses an algorithm it calls SPTAG (“Space Partition Tree and Graph”) to search for vectors. The search query is being converted to a vector. SPTAG uses the “approximate nearest neighbors” (ANN) feature to find search-related vectors.
In line with this feature; the result after the search is “How tall is the tower in Paris?” It is the page closest to the subject where the answer to the question can be found. This page is most likely related to the Eiffel Tower.
Microsoft has released the open source SPTAG algorithm on GitHub today. The code is used to answer questions on Bing, but developers can use this algorithm for their own search engines. A single machine can handle 250 million vectors and answer 1,000 queries per second.
Microsoft CEO Satya Nadella spoke about his desire to “democratize artificial intelligence”. He said the algorithm is available to everyone. He also stated that it is a specialized tool that requires significant expertise, but that various developers can use it as part of their toolkit that solves various problems.