Anatomy of a Privacy-Safe Large-Scale Information Extraction System Over Email

Abstract

Extracting structured data from emails can enable several assistive experiences, such as reminding the user when a bill payment is due, answering queries about the departure time of a booked flight, or proactively surfacing an emailed discount coupon while the user is at that store. This paper presents Juicer, a system for extracting information from email that is serving over a billion Gmail users daily. We describe how the design of the system was informed by three key principles: scaling to a planet-wide email service, isolating the complexity to provide a simple experience for the developer, and safeguarding the privacy of users (our team and the developers we support are not allowed to view any single email). We describe the design tradeoffs made in building this system, the challenges faced and the approaches used to tackle them. We present case studies of three extraction tasks implemented on this platform-bill reminders, commercial offers, and hotel reservations-to illustrate the effectiveness of the platform despite challenges unique to each task. Finally, we outline several areas of ongoing research in largescale machine-learned information extraction from email.

Online template induction for machine-generated emails

“…• The Google Assistant can answer queries such as "When is my next bill due?" using information mined from bill reminder emails [31].…”

“…• If a user walks by a store for which they have received a promotional offer email, the Google Assistant can proactively remind them of the offer using information mined from the email [31].…”

“…Rather, 90% of email traffic consists of machine-generated business-to-consumer (B2C) emails instantiated from a collection of a few million email templates, as illustrated in Figure 1. Existing email mining systems take advantage of this templatic nature of emails to extract information from emails using a two-step process [11,20,21,31].…”

“…Traditionally, email mining systems separate these two logical steps (i.e. template induction and email annotation) into two distinct computational operations [20,31]. Template induction is performed offline in batch, while email annotation is performed online as new emails arrive into the system.…”

“…In this paper, we present a fully online email mining system that greatly reduces this delay by performing both template induction and email annotation online. Our system, called Crusher, is a redesign of Juicer, an email mining system that was developed at Google and previously described in [31]. Whereas Juicer performs template induction offline and email annotation online, Crusher performs both online.…”