Watch: A Complete List of Google's Autocomplete Signals in Chrome
Google Chrome utilizes a machine learning model for address bar autocomplete. This article lists all signals for the auto-complete function.
Transcript
When you type into Google Chrome's address bar, a machine learning model instantly predicts what you are looking for. This system is a neural network called a multilayer perceptron, designed to rank autocomplete suggestions in real time.
The model makes its predictions by looking at twenty different signals. These include your personal browsing history, like how often you visit a site, how recently you typed it, and whether you have it bookmarked. It also measures website characteristics, such as the length of the address, and match characteristics, like how closely your typed letters fit the page title or domain name.
To keep Chrome fast and lightweight, the model's architecture is kept simple. It processes these twenty signals through a dense layer of sixty-four neurons, using quantized weights to save memory. A final layer translates these calculations into a score between zero and one. The higher the score, the more relevant the suggestion, allowing Chrome to instantly rank the best matches right at the top of your dropdown menu.