公开(公告)号：US10374995B2

公开(公告)日：2019-08-06

申请号：US14755518

申请日：2015-06-30

Applicant: OATH INC.

Inventor： Liane Lewin-Eytan ,  Guy Halawi ,  Dotan Di Castro ,  Zohar Karnin ,  Yoelle Maarek ,  Michael Albers

IPC: H04L12/58

Abstract: As is disclosed herein, user behavior in connection with a number of electronic messages, such as electronic mail (email) messages, can be used to automatically learn from, and predict, whether a message is wanted or unwanted by the user, where an unwanted message is referred to herein as gray spam. A gray spam predictor is personalized for a given user in vertical learning that uses the user's electronic message behavior and horizontal learning that uses other users' message behavior. The gray spam predictor can be used to predict whether a new message for the user is, or is not, gray spam. A confidence in a prediction may be used in determining the disposition of the message, such as and without limitation placing the message in a spam folder, a gray spam folder and/or requesting input from the user regarding the disposition of the message, for example.

公开(公告)号：US10558822B2

公开(公告)日：2020-02-11

申请号：US14969201

申请日：2015-12-15

Applicant: Oath Inc.

Inventor： Liane Lewin-Eytan ,  Dotan Di Castro ,  Eyal Zohar ,  Yoelle Maarek ,  Ran Wolff ,  Doug Sharp

IPC: G06F17/20 ,  G06F21/62 ,  G06F17/22 ,  H04L12/58

Abstract: Methods, systems, and computer-readable media for anonymizing electronic documents. In accordance with one or more embodiments, structurally-similar electronic documents can be identified among a group of electronic documents (e.g., e-mail messages, documents containing HTML formatting, etc.). A hash function can be specifically tailored to identify the similarly structured documents. The structurally-similar electronic documents can be grouped into a same equivalence class. Masked anonymized document samples can be generated from the structurally-similar electronic documents utilizing the same equivalence class, thereby ensuring that the anonymized document samples when viewed as a part of an audit remain anonymous. An online process is provided to guarantee k-anonymity of the users over the entire lifetime of the auditing process. An auditor's productivity can be measured based on the amount of content revealed to the auditor within the samples he is assigned. The auditor's productivity is maximized while ensuring anonymization over the lifetime of the audit.