Dismiss
InnovationQ will be updated on Sunday, Oct. 22, from 10am ET - noon. You may experience brief service interruptions during that time.
Browse Prior Art Database

System for detecting and blocking ads by using visual features

IP.com Disclosure Number: IPCOM000205926D
Publication Date: 2011-Apr-08
Document File: 6 page(s) / 52K

Publishing Venue

The IP.com Prior Art Database

Abstract

We describe a system that identifies whether some RIA asset (such as images, SWFs and HTML5 content) represent an advertisement or not. This is done by using both traditional HTML features (e.g. patterns in the URL) and a set of innovative visual features, which were not taken into account by now. There are programs, such as Ad Blocker, that can block ads before loading a HTML page. The approach used by such applications is to block URLs that are determined as advertisements using a set of handcrafted rules, without analyzing the content. We illustrate a different approach based on Machine Learning which focuses primarily on visual features, thus requiring minimal maintenance and also supporting next generation ads such as HTML5 and Flash ads. Furthermore, taking into account the fact that advertisements have similar visual aspects (as they have similar purposes) regardless of the effective technology used, the system is able to accept any type of composition whose visual content can be extracted.

This text was extracted from a Microsoft Word document.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 22% of the total text.

Title

System for detecting and blocking ads by using visual features

Inventor/Author

Dan Banica

Paul-Alexandru Chirita

Summary

We describe a system that identifies whether some RIA asset (such as images, SWFs and HTML5 content) represent an advertisement or not. This is done by using both traditional HTML features (e.g. patterns in the URL) and a set of innovative visual features, which were not taken into account by now.

There are programs, such as Ad Blocker, that can block ads before loading a HTML page. The approach used by such applications is to block URLs that are determined as advertisements using a set of handcrafted rules, without analyzing the content. We illustrate a different approach based on Machine Learning which focuses primarily on visual features, thus requiring minimal maintenance and also supporting next generation ads such as HTML5 and Flash ads. Furthermore, taking into account the fact that advertisements have similar visual aspects (as they have similar purposes) regardless of the effective technology used, the system is able to accept any type of composition whose visual content can be extracted.

Background

While surfing online, users are often annoyed by intrusive, irrelevant ads and therefore many of them choose to use some programs that block advertisements. Ad blockers are usually based on a set of handcrafted rules and therefore ads served from new ad domains are usually not caught. This scenario can be better handled by analyzing the visual content, as our system does.

Furthermore, by allowing users to explicitly specify when they want a specific ad to be blocked, this information can be used in order to learn exactly what kind of ads annoys the user most, and block that kind of ads only. Using visual features for this task is essential, because they are the ones that determine the user to consider an ad annoying or not.

Prior Art/Solutions

·         AdBlocker - https://addons.mozilla.org/en-US/mobile/addon/adblock-plus/

·         AdBlock Plus - https://addons.mozilla.org/en-US/mobile/addon/adblock-plus/

As previously mentioned, programs like AdBlocker and AdBlock Plus are based on a list of manually crafted rules and then URLs are blocked by analyzing this list. If the URL contains some patterns (like “/ads”, “doubleclick”, etc), then the system decides to block that URL.

Such systems are limited for several reasons. First, no machine learning is used, so new rules have to be inserted each time a new URL is not matched by existing conditions (for this reason the lists are always huge and incur high maintenance costs).

Second, a system that only takes URL and similar features into account is intrinsically limited. All content served from a location that satisfies a URL pattern is treated in the same way, while a given domain may serve different types of ads. By using visual features we can learn to block only those ads that annoy the user most. Also, more and more Flash/HTML5 ads will be served, therefore our system...