Generating query variants using a trained generative model

Topics: AI (Deep Learning), AI Mode, AIOverviews, LLMO / GEO, Probably in use, Query Fan Out, Retrieval Augmented Generation (RAG), Search Intent, Search Query Processing, User Signals

The Google patent discusses a system and method for generating variations of a submitted query utilizing a trained generative model, particularly a neural network model. This model actively produces query variants in real-time based on the original query’s tokens and additional features, allowing for better search result retrieval. It can create variants for both common and less frequently submitted queries, enhancing query processing by providing richer input for search systems. This capability aims to improve user experience by quickly identifying relevant results without requiring users to reformulate queries. This patent seems to me as much relevant as the “Thematic Search” patent according to the query fan out process, because it is published worldwide not only in the US. So the possibility being used in practise in search is very probable.