New neural networks make it possible to better correct typing errors, to index texts on web pages, to structure a search in subdomains and to detect key moments in videos.
During the “Search On” virtual conference, Google presented a series of new features that will improve its search function. They all rely on the use of neural networks. Thus, Google is now using a new algorithm that will only take 3 milliseconds to detect typing or writing errors in a search and offer a corrected spelling. Google states that about 10% of all searches are poorly written.
In order to be able to respond to particularly sophisticated searches, ie around 7% of the overall volume, Google now indexes not only the pages, but also the paragraphs of text in these pages. Which allows him to find more easily “The needle in the haystack”. This update is available in all languages.
Finally, Google is now able to divide a search subject into subdomains, which will be presented as such in the results page. This will allow the user to obtain greater richness in the responses. This new feature will be rolled out by the end of the year.
Finally, Google will generalize the automatic semantic analysis of videos, in order to detect and tag key moments. These tagged moments will be described and indexed automatically, allowing users to directly access the right moment in a video.
Source : Google