Patent Application
20150371281
In recent years, online advertisers have attempted to improve the relevancy of advertisements (or “ads”) shown to users by profiling users' online interests and delivering ads relevant to those interests. Online trackers have become ubiquitous and cover a large fraction of a user's browsing behavior, enabling them to build comprehensive profiles of their online interests. This widespread tracking of users and the subsequent personalization of ads have received a great deal of negative feedback primarily because they lack insight into how their data is being collected and used. For example, consider a user that repeatedly receives ads about cures for a particularly private ailment. The user currently lacks a way to determine how advertisement targeting mechanisms are profiling her. Is it because the user's online interest profile matches the profile of users the advertiser is seeking to target? Is it because the websites that the user visits are contextually relevant to the ad and draw users that the advertiser is interested in targeting? Or, is it because the user actually tried to buy the particular medication online previously and the advertiser is re-marketing the product? Providing transparency into how users are profiled for target advertisements would lead to a new class of ad control mechanisms that enable end-users to exert fine-grained control over targeted advertising. Specifically, end-users would be able to block tracking along the actions related to specific ads, or indicate their ad preferences at a granularity that is not feasible via existing tools.
The primary challenge in providing transparency is to design mechanisms that account for the inherent complexity involved in ad delivery, including advertisers selecting from a variety of targeting mechanisms and multiple ad campaigns co-existing. Consequently, at any point, a webpage could contain ads from multiple campaigns that are targeting different aspects of the user's online interests. Furthermore, the ad selection process is based on a real-time auction, whose outcome also depends on financial parameters of the ad campaign like the cost per mille/thousand impressions (CPM) and desired click through rate (CTR).
Existing work seeking to address some of the transparency properties falls short in several ways. A common approach pushed forward by the industry is the AdChoices initiative and Google's ad preferences dashboard. These approaches provide the end-user visibility into their advertising profile and allow one to opt-out of certain “categories” across a few online trackers and ad networks. However, even with the limited participating entities, the mechanisms are not evenly implemented and often hard to use.
Various browser tools, such as Ghostery, AdBlock, NoScript and Collusion provide users visibility into the presence of third-party trackers on websites. However, these tools cannot determine the specific targeting mechanisms employed and consequently, only provide a very coarse grained control, by either turning off or on all ads and tracking. Policy proposals like Do Not Track provide a regulatory framework over the tracking and profiling of user data. However, in the present form, there is no legal mandate for the ad-networks to comply with this directive and this might require government intervention for universal enforcement.
Finally, a number of privacy preserving targeted advertising solutions have been proposed that aim to minimize the exposure of user data. Privad and Adnostic rely on local caching of ads and generation of the user profile, ObliviAd relies on a new secure processing hardware, and RePriv provides browser specific tools for third-parties to extract user profiles from the browser. However, these proposals require re-factoring large parts of the ad targeting ecosystem, hence making them difficult to deploy.
Therefore, there is a need for a mechanism that provides transparency into the targeted advertising ecosystem.
To address the challenges of targeted advertisement transparency, a method of identifying targeted advertisement is used. In one embodiment, the method includes a computing system receiving a set of content items having at least one advertisement element embedded within each of the content items and extracting the one or more advertisement element embedded within each of the content items. The method further includes determining a second content item associated with the advertisement element and a semantic category for the first content item and the second content item. The semantic categories of the first content item and the second content item are matched to determine if the advertisement is contextually related with the first content item.
In an embodiment, the method of identifying targeted advertisement is used to identify a domain that caused the advertisement element to be embedded within the content item. The method further comprises extracting the one or more advertisement element, wherein the advertisement element is a remarketing tag, and matching a domain associated with the second content item and the remarketing tag with a log of domains browsed by the user having embedded remarketing code in the web page source. The matching domain causing the advertisement to be embedded within the first content item can be displayed to the user.
In another embodiment, the method of identifying targeted advertisement is used to identify contextually targeted advertisements. The method comprises receiving by a computer system, a set of web pages having at least one advertisement element embedded within the set of web pages and extracting the one or more advertisement element. The method further comprises: determining a landing page associated with each of the extracted advertisement elements; determining at least one web page category for each of the web pages of the set of web pages; generating a model operable to relate the advertisement element with the associated set of web pages using a set of binary classifiers; and computing a targeting score corresponding to the advertisement associated with the set of web pages, the targeting score comprising the set of binary classifiers generated from the model, the targeting score determining if the advertisements are contextually associated with the set of web pages.
Described herein is a practical measurement and analysis framework that relies on end-user measurements to provide transparency into how online display ads (flash and image based ads) target users. In one embodiment, the measurement tool is a browser-based extension that provides detailed measurements of online display ads. The ads are characterized by three distinct targeting techniques: contextual targeting, behavioral targeting, and remarketed targeting. The analysis component uses a novel contextual model to predict the ad categories expected on a webpage in the absence of tracking to determine contextually targeted ads. A metric can be applied to quantify the extent to which the user is being behaviorally targeted. For remarketing based targeting, exact actions of the user's clickstream that led to the ad being targeted can be determined
Contextual targeting, as illustrated in
Behavioral Targeting, as illustrated in
Re-marketed targeting (or retargeting), as illustrated in
The insurance company (via the ad-network) can place re-marketing ads—e.g., insurance discounts—into other websites the user visits which can be unrelated to cars or insurance to lure the user back to finish the purchase. Here, the advertiser exploits a very narrow and explicit signal from the user to target ads.
While in one embodiment, contextual targeting, behavioral targeting and remarketed targeting are only used in characterizing ads, there are several other attributes and mechanisms that can be used for ad selection. For example, attributes, such as user geolocation, inferred demographics of website visitors, and browser/device identification can be used to characterize ads.
The client 202 can comprise a memory 204, a browser 206, a processor 208, and a document object model 224. The memory 204 provides storage for one or more a web pages, an exemplary one of which is illustrated at 214. The web page 214 can includes an identification tag. As described above, the identification tag allows the web page 214 to identify the type of advertisement or advertisements that are to be part of the web page 214. The client 202 also includes a browser 206. The browser 206 renders the web page 214, including an advertisement 216 according to a document object model 224. The document object model 224 defines the manner in which the web page 214 is rendered. The web page 214 also includes content 222, which can be, for example, text, images, or any other page content. As will be described more fully below, in an embodiment, the memory 204 can also include JavaScript software, a JavaScript Object Notation (JSON) object: 286 and a JavaScript library 220, which defines methods to insert ad creative into a document object model 224. The document object model 224 defines the web page 214. As will be more fully described below, the JavaScript software makes calls to the JavaScript library 212 that interprets the JSON object 286 to insert an advertisement in the page 214 that is rendered by the browser 206.
Detecting Contextually Targeted Ads
At step 300, a set of content items having at least one advertisement element embedded within each of the content items is received. In one embodiment, the content item can be a web page. At step 302, one or more advertisement elements are extracted. In one embodiment, this module parses the (possibly complex) DOM structure of the webpage and extracts specific attributes of display ads that reveal the landing page for each of the ads. The landing page refers to the website that would be visited by clicking on the ad. Extracting the advertisement elements is complicated by the fact that display ads are often embedded in nested iFrame tags spanning multiple levels. Furthermore, the same origin policy enforced by modem web browsers permits an outer iFrame to inspect and communicate with its immediate inner iFrame only if the two iFrames are from the same domain. To address this, custom JavaScript code is recursively injected into all iFrames on the webpage and a dedicated background page is set up as a communication bridge between nested iFrames. This code reads the <href> or <flashvars> attributes for image or flash ads, and aggregates information at the background page running within the context of the browser plugin. This module also logs ad tracker elements, such as re-marketing scripts and cookies, on the webpage. Re-marketing scripts are detected by searching for the unique ad tracker JavaScript code. Ad tracking cookies are detected by monitoring outgoing HTTP requests and comparing against the publicly available patterns provided by services such as the Ghostery tracker database.
At step 304, a second content item associated with the advertisement element is determined. In one embodiment, a second content item can be a landing page. For each identified ad element, the landing page is inferred by parsing the value of the attributes extracted by the DOM parser module and searching for specific patterns in the URL, such as “adurl=” and “redirect url=”. At step 306, semantic categories of the first content item and the second content item are determined For example, the webpage “www.nfl.com” will be associated with the following semantic categories: sports, American and football, and the webpage “www.webmd.com/cancer” will be associated with the following semantic categories: health, medical conditions, cancer. A single page or ad landing page can be associated with more than one semantic category. In one embodiment, the semantic categories of the web page and the landing pages of all the display ad elements is determined by querying a content analysis API. The content analysis API analyzes content and identifies concepts or key words/phrases within the content. Additionally, the content analysis API can tag or categorize the content. The content analysis API can be publicly available, such as Yahoo! Content Analysis API.
At step 308, a determination is made as to whether the advertisement element is contextually related with the first content item by matching the semantic category of the first content item and the second content item. In one embodiment, a contextual model is used to determine if an advertisement is contextually related to the webpage. For a given page that matches a set of web page categories, the model outputs a for 0 for each ad-category, which is a prediction on whether an ad of that category should appear on that web page. If the prediction of an ad appearance holds (output=1), then the ad-categories conform to the trained model, and the targeting is contextual. In another embodiment, the contextual model used to determine a contextual association between advertisement and a webpage can be a machine learning model. In a specific embodiment, the model learns by logistic regression with L1 regularization. The model learns a set of coefficients which weigh the relevance of each (input) page category to the (output) ad category. The L1 regularization enforces a sparse model, wherein only a few coefficients will be non-zero as most webpages are mapped to only a few categories Importantly, the learned classifier model also outputs a confidence score over the classification results, and can utilize this information to account for noise inherent in the data being modeled.
The models can be trained on either a tracking dataset or a no-tracking dataset. The tracking dataset uses an empty user profile wherein all cookies are cleared and visits the webpages using default browser settings. The user's cookies, cache pages, and track histories are taken into consideration. This allows the model to take into account ads that target the interest profile of the user over the pages visited previously. The no-tracking dataset uses a blank user profile, wherein cookies, cache pages and track histories are not taken into consideration. The no-tracking dataset offers a view of what kinds of ads would have been selected if the ad-network had no information about the user. This allows the model the functionality to predict the ad categories that can appear on a webpage. This model can be applied on users in the tracking dataset which provides a reason about how the user is being tracked by comparing predictions against the ads being loaded. The contextual model trained on a no-tracking dataset has some inherent noise, such as differences across ad campaigns for the same category and inherent dynamic of ad auctions and campaigns. These serve to weaken the association between webpage categories and the predicted contextual ad category. The noise reduces the confidence of the classifier for the two output classes, resulting in the distribution of confidence overlapping. Thus, in one embodiment, only samples whose classification confidence is above a certain threshold are considered.
To characterize the model generated using the above described approach, an Area Under the Curve (AUC) score is computed and the model parameters are inspected. The AUC score can range from a 0.0 representing a random ad placement to a 1.0 representing a perfect precision and recall advertisement placement. AUC scores typically range from 0.5 (random) to 1 (perfect precision and recall). In an example across 81 ad categories, the median AUC score is 0.71. 10% of the ad categories had an AUC score above 0.85 (e.g. American Football, Travel Transportation and Disease & Medical Conditions) and 9% of the ad categories had an AUC score below 0.6 (e.g. Credit, Gaming and Lottery).
Detecting Behaviorally Targeted Ads
Applying the learned contextual model to a user's web trace (tracking dataset), two cases are considered in each webpage instance: (i) the true-positive case (TP), which validates the classifier prediction and indicates that the ad was selected purely based on the page context, and (ii) the false negative case (FN), where the where the prediction is in-correct, indicating that the ad was selected based on factors beyond the context of the page (i.e., which the model completely accounts for). The other two cases, true negatives and false positives, are not strong indicators of the ad selection being contextual. Putting these together, a false negative rate (FNR) is denoted and computed for a set of pages. In an embodiment, the FNR=FN/(FN+TP) and can be used as the targeting score. When FNR is close to 0, ads placed on the page indicates the ads are contextually targeted, and values close to 1 indicate that the ads are behaviorally targeted.
Detecting Re-Marketing Ads
Re-marketing ad campaigns require advertisers to tag different pages on their sites with specific JavaScript code generated by the ad platform. This allows the advertiser to distinguish users that reach different parts of their site, and customize the advertising strategies accordingly. For example, a re-marketing ad can display ads for travel tickets to a specific destination based on the fare search the user performed. Hence, re-marketing campaigns ignore the user profile and “follow” the user on the web, re-marketing the product to convince the user to come back to the advertiser's webpage. In one embodiment, the method for identifying targeted advertisement monitors and logs all domains visited by a user that have embed JavaScript remarketing code in the page source. Subsequently, for every ad, the domain of the ad landing page is matched against the set of domains containing the remarketing scripts. When the two match, the exact pages in the user's clickstream that caused the specific ad to be targeted can be determined. This enables the user to learn about how their particular actions in the past resulted in the current ad being displayed.
Targeting Ad Analysis Display
In the case of a remarketed ad, the display can provide information relating to the tracker, page category, ad landing page, ad type, domain of the landing page, and related browsing history, including the date and time the website was visited. The representation can further disseminate information regarding how much of the user's profile is being used to serve targeted ads, denoted as a tracking or privacy risk.
According to the inventive principles as disclosed in connection with the preferred embodiment and other embodiments, the invention and the inventive principles are not limited to any particular kind of user device, but can be used with any general purpose computing device having networking capabilities, as would be known to one of ordinary skill in the art, arranged to perform the functions described and the method steps described.
In an embodiment of the present invention, some or all of the method components are implemented as a computer executable code. Such a computer executable code contains a plurality of computer instructions that when performed in a predefined order result with the execution of the tasks disclosed herein. Such computer executable code can be available as source code or in object code, and can be further comprised as part of, for example, a portable memory device or downloaded from the Internet, or embodied on a program storage unit or computer readable medium. The principles of the present invention can be implemented as a combination of hardware and software and because some of the constituent system components and methods depicted in the accompanying drawings can be implemented in software, the actual connections between the system components or the process function blocks can differ depending upon the manner in which the present invention is programmed.
The computer executable code can be uploaded to, and executed by, a machine comprising any suitable architecture. Preferably, the machine is implemented on a computer platform having hardware such as one or more central processing units (“CPU”), a random access memory (“RAM”), and input/output interfaces. The computer platform can also include an operating system and microinstruction code. The various processes and functions described herein can be either part of the microinstruction code or part of the button program, or any combination thereof, which can be executed by a CPU, whether or not such computer or processor is explicitly shown. In addition, various other peripheral units can be connected to the computer platform such as an additional data storage unit and a printing unit.
The functions of the various elements shown in the figures can be provided through the use of dedicated hardware as well as hardware capable of executing appropriate software. When provided by a processor, the functions can be provided by a single dedicated processor, by a single shared processor, or by a plurality of individual processors, some of which can be shared. Explicit use of the term “processor” or “controller” should not be construed to refer exclusively to hardware capable of executing software, and can implicitly include, without limitation, digital signal processor hardware, ROM, RAM, and non-volatile storage. Other hardware, conventional and/or custom, can also be included. Similarly, any switches shown in the figures are conceptual only. Their function can be carried out through the operation of program logic, through dedicated logic, through the interaction of program control and dedicated logic, or even manually, the particular technique being selectable by the implementer as more specifically understood from the context.
All examples and conditional language recited herein are intended for pedagogical purposes to aid the reader in understanding the principles of the invention and the concepts contributed by the inventor to furthering the art, and are to be construed as being without limitation to such specifically recited examples and conditions. Moreover, all statements herein reciting principles, aspects, and embodiments of the invention, as well as specific examples thereof, are intended to encompass both structural and functional equivalents thereof. It is intended that such equivalents include both currently known equivalents as well as equivalents developed in the future, i.e., any elements developed that perform the same function, regardless of structure.
The present invention is described in the foregoing examples, which are set forth to aid in the understanding of the invention, and should not be construed to limit in any way the scope of the invention as defined in the claims which follow hereafter. While the foregoing has been described in some detail for purposes of clarity and understanding, it will be appreciated by one skilled in the art, from a reading of the disclosure that various changes in form and detail can be made without departing from the true scope of the invention.
| Number | Date | Country | |
|---|---|---|---|
| 62014344 | Jun 2014 | US |
