Deep search using large language models
Topics: AI Mode, E-E-A-T, LLMO / GEO, Microsoft, Probably in use, Query Fan Out, Ranking, Retrieval Augmented Generation (RAG), Scoring
This Microsoft patent describes a “deep search” system that uses large language models (LLMs) to provide more relevant search results. The system first analyzes a user’s query to generate several possible “intents” or meanings behind the search. After selecting the most likely intent, it creates alternative queries to explore that topic more thoroughly. The system then gathers results from a standard search utility for all these queries and uses an LLM to score each result for relevance before ranking and displaying the best ones to the user.
