Dynamic organization of search results
This Google patent describes a method for improving search engine results by using a generative AI model to break down a user’s initial search query into multiple, highly specific sub-queries. read more
Search-query suggestions using reinforcement learning
This Microsoft patent outlines a method for generating high-quality search query suggestions using machine learning and reinforcement learning. Initially, a base model is trained using real search queries previously entered read more
Bottleneck aware query alteration for search
This Microsoft patent describes a method and system used by search engines to improve search results by identifying and resolving “bottleneck terms” within a user’s search query. A bottleneck term read more
Thinking to Recall: How Reasoning Unlocks Parametric Knowledge in LLMs
This document explores how enabling reasoning in Large Language Models (LLMs) significantly improves their ability to recall simple factual knowledge, even for basic questions that do not require complex logical read more
Revisiting Text Ranking in Deep Research
This paper investigates how well-established text ranking methods from information retrieval (IR) perform in the context of “deep research” — a task where AI agents iteratively search the web and read more
GIST: Greedy Independent Set Thresholding for Max-Min Diversification with Submodular Utility
This Google paper introduces GIST (Greedy Independent Set Thresholding), an algorithm designed to solve a new optimization problem called Max-Min Diversification with Monotone Submodular Utility (MDMS). The core challenge is read more
Trust Me on This: A User Study of Trustworthiness for RAG Responses
This Google paper presents a controlled user study examining how different types of explanations affect user trust in responses generated by Retrieval-Augmented Generation (RAG) systems. The researchers tested three explanation read more
Determining a quality score for a content item
This Google patent describes a system for calculating a “quality score” or “engagement level” for content items, specifically within social networking or web-based streams. Instead of simply showing the newest read more
Small models, big results: Achieving superior intent extraction through decomposition
The document from Google researchers introduces a method to improve “intent extraction”—the ability of an AI to look at a series of user actions on a phone or website and read more
Page-based prediction of user intent
This Google patent describes a system that predicts a user’s intent based on the specific webpage they are currently visiting and the navigational path they took to get there. Once read more