Automatic classification of display ads using ad images and landing pages

• US 8,732,014 B2
• Filed: 12/20/2010
• Issued: 05/20/2014
• Est. Priority Date: 12/20/2010
• Status: Expired due to Fees

• Current Assignee: R2 Solutions LLC (Acacia Research Corporation)
• Original Assignee: Yahoo! Inc. (Apollo Global Management, Inc.)
• Inventors: Kae, Andrew, Kan, Kin Fai, Narayanan, Vijay K.
• Primary Examiner(s): Sittner, Matthew T
• Application Number:
  US12/973,858
• Publication Number:
  US 20120158525A1
• Time in Patent Office: 1,247 Days
• Field of Search: None
• US Class Current: 705/14.4
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising
• CPC Class Codes: G06Q 30/08 Auctions
• CPC Class Codes: G06Q 40/00 Finance; Insurance; Tax strategies; Processing of corporate or income taxes
• CPC Class Codes: G06Q 40/04 Trading; Exchange, e.g. stocks, commodities, derivatives or currency exchange
• CPC Class Codes: G06Q 40/06 Asset management; Financial planning or analysis

Abstract

A system and method for automatically classifying ads into a taxonomy of categories, the method including: extracting text features from ad images using OCR (optical character recognition) techniques; identifying objects of interest from ad images using object detection and recognition techniques in computer vision; extracting text features from the web-page of the advertiser to which the user is re-directed when clicking the ad; training statistical models using the extracted features mentioned above as well as advertiser attributes from a historical dataset of ads labeled by human editors; and determining the relevant categories of unlabeled ads using the trained models.

Claims

1. A method for classifying display ads automatically into a taxonomy of categories, the method comprising:

using a processor device, causing a computer to perform steps of;

extracting first text features from an ad image of a display ad using optical character recognition (OCR) techniques;

identifying objects of interest from the ad image using object detection and recognition techniques in computer vision;

extracting second text features from a title, keywords, and content of a landing page (a web-page of an advertiser that a user is redirected to when clicking an ad) associated with the display ad;

generating bag-of-words ad features using the extracted first and second text features, as well as attributes of the advertiser;

using the generated bag-of-words ad features to categorize the display ad;

training statistical models using the generated bag-of-words ad features on a historical dataset of ads labeled by human editors; and

determining relevant categories of unlabeled ads using the trained statistical models to classify the display ads.

2. The method of claim 1 further comprising:

facilitating manual editorial categorization by suggesting categories;

generating training features and targets in use interests modeling, and generating attributes in match/rank algorithms used in ad selection algorithms.

3. A system of display ad categorization, the system comprising:

a processor device;

a storage device operably coupled with the processor device, said storage device comprising instructions that are executed by said processor device;

wherein the instructions cause a computer to perform a method comprising steps of;

reading an ad image for a display ad and a landing page associated with said display ad from the storage device;

extracting first text features from the ad image using optical character recognition (OCR) techniques;

executing object detection and recognition to identify objects of interest from the ad image;

parsing the landing page to extract second text features from a title, keywords, and content of said landing page;

generating bag-of-words ad features using the extracted first and second text features, as well as attributes of the advertiser;

using the generated bag-of-words ad features to categorize the display ad;

storing the extracted first and second text features from the ad image and the landing page in the storage device;

training statistical models using the generated bag of words ad features on a historical dataset of ads labeled by human editors; and

determining relevant categories of unlabeled ads using the trained statistical models to classify the display ads.