Facebook description of methodology
This document describes ad measurement, filtration and reporting methodologies for Facebook News Feed and RHS ad impressions.
It addresses how Facebook complies with the IAB Ad Impression Measurement Guidelines, the MRC Minimum Standards and the MRC Invalid Traffic Detection and Filtration Guidelines Addendum (Version 1.0, 15 October 2015). Facebook News Feed and right hand column (RHC) is currently accredited for served display ad impressions (desktop, mobile web and in-app).
Facebook employs "General Invalid Traffic Detection" techniques as defined by the MRC's Invalid Traffic Detection and Filtration Guidelines Addendum.
OVERVIEW
Facebook is an online social networking service headquartered in Menlo Park, California. Our mission is to give people the power to build a community to bring the world closer together. The platform enables users to consume content via the News Feed, chat with friends via Facebook Messenger, create personal (and business) profile pages as well as sharing photos and videos and joining various groups.
Ad placements & ad types
Facebook has three primary ad placements that are under examination for MRC accreditation. Ads are displayed in the News Feed that runs in both desktop and mobile environments (web and app) as well in the right-hand column (RHC) in the desktop environment. Additional information can be found by visiting – https://www.facebook.com/help/407108559393196.
Ads Manager is a self-serve interface for advertisers to create and run ads, target ads, set and configure ad budgets, and for reporting. Refer to the following URL for additional details – https://www.facebook.com/ads/manage/home/
Advertisers can also programmatically access Facebook's advertising platform from their own advertising tools via the Marketing API. This enables developers to use all the same functionality as Facebook tools within a custom solution. Refer to the following URL for additional details – https://developers.facebook.com/docs/marketing-apis.
Ad measurement metrics
Facebook's reporting system counts impressions delivered and measured in News Feed (mobile and desktop) and RHC (right-hand column).
Impressions reported through Facebook's reporting system, Ads Manager, reflect the billable impressions delivered and measured via Facebook's ad delivery system.
To review available ad metrics and learn how users are reaching their business goals, see the Get Started Guide to Reports in Ads Manager in the Facebook Help Centre. Please note that depending on how a customer sets up their ad campaigns, some data might not be collected, or may not be available.
Ad impression measurement
For ad impressions displayed in Facebook News Feed and RHC, across both desktop and mobile platforms (primarily iOS and Android mobile web and mobile applications or native environments), an ad impression is counted the instant any part of the ad appears on the user's screen (i.e. greater than zero pixels and greater than zero seconds).
A JavaScript-based check is performed every 100 milliseconds or a frame rate of 60 frames per second is used to determine whether the ad has satisfied the greater than 0 pixel and more than 0 seconds requirement for measurement. .
- If the check is a SUCCESS, a 1x1 pixel is fired and the impression is counted.
- If the check is a FAIL, the 1x1 pixel is not fired and the impression is not counted.
Note: This additional check step does not exist for certain platforms because Facebook can't detect when stories come on screen so they fire the pixel as soon as they are rendered. Some feature phones do not support the viewport measurement methodology and employ a "rendered" measurement methodology. These platforms and devices only account for a small percentage of the number of ads served on Facebook.
Facebook does allow certain third-party ad tracking tags (i.e. DoubleClick, Atlas etc.) on their ads but Facebook performs all ad serving (via the Facebook ad server). No empty FRAME for dynamic ads serving is involved.
Compound tracking
Compound tracking is not used. Each impression event is individually recorded and measured.
Offline measurement
Offline impressions are stored and re-sent once the user is back online; however, this must occur within six hours of the impression event, otherwise it will be marked as non-billable.
Auto-refresh
Facebook auto-refreshes ad impressions in the RHC placement on desktop browsers after a period of time, except when the individual takes actions that might indicate interest in the ad (e.g. hovering the cursor over the ad). The refresh is triggered when the individual scrolls or resizes the Facebook window, so no refresh occurs when Facebook is loaded on a background tab or the individual is interacting with other windows or applications. Auto-refresh ad impressions make up over half of the total RHC impressions.
Auto-play
The vast majority of Facebook video events are auto-play initiated. Additional information on auto-play functionality can be found by visiting https://www.facebook.com/business/help/846558025399472?locale=en_US
Cache-busting
Cache-busting is a technique to prevent a web browser from re-using an ad it has already shown to someone, and forces the browser to request a new ad each time the page refreshes.
Facebook makes use of the cache-control http header to prevent caching on dynamic pages.
Additionally, the client impression timestamp is included in the request (via the "cts" parameter), which also helps reduce caching in the same manner in which a random number would.
Data logging and processing
Each call to Facebook's http service generates a logging event that is processed through the stream processing infrastructure. Such logging events reach the Facebook reporting service, and the data is aggregated and stored in the Facebook data warehouse. Customers can create and schedule reports based on this aggregated data.
All impressions are logged in the ad logs and counted as "raw" statistics. Facebook processes the raw logs, flags some of the events as non-billable based on a list of rules and outputs the raw data as annotated logs.
Known limitations
Facebook believes the below limitations to be insignificant.
- Caching: Facebook employs appropriate cache-busting techniques; however, defeating all caching is not possible.
- Abandonment: Abandonment is a known measurement limitation; however, due to Facebook's technical measurement methodology of not recording the ad impression until it enters the viewport, the impact is believed to be insignificant.
- Ad blocking tools (desktop only): Facebook has implemented measures that are designed to prevent ads blocked via AdBlock and AdBlock Plus from being measured as billable impressions (but they may still have an insignificant impact on reported metrics as these tools are constantly evolving).
- Image rendering disabled (desktop only): Browsers with images disabled (desktop and mobile web) may result in ad impressions and feed impressions being measured that did not have the opportunity to be seen.
- JavaScript disabled or incapable browsers (desktop only): The ads viewability on Facebook browser-based products rely on JavaScript to be enabled. Internet Explorer versions 7 and previous are also not supported.
- Google's Chrome browser has pre-rendering functionality which may overstate desktop feed impressions and RHC impressions. (Desktop only)
- Facebook's News Feed impression measurement does not consider the placement of other windows that might be open on the user's screen. As a result, if a user were to place another window over an ad (or even move the browser window off screen), Facebook will not be able to detect this. (Desktop only)
- Some feature phones do not support the viewport measurement methodology and employ a "rendered" measurement methodology (as noted above).
- A user's connectivity is not always consistent or persistent, especially in mobile environments. In order to create a seamless user experience, we batch-fetch several posts at a time, ensuring that ad copy appears for all ads, even if images are still loading due to low connectivity. In situations where the ad is partially loaded, ad impression measurement is counted when any part of the ad appears on the user's screen (i.e. greater than zero pixel and greater than zero second), including ad copy.
Partner qualification controls and material business relationships
For the purposes of this document, which speaks to Facebook's procedures as it relates to ads delivered on Facebook.com, Facebook is both the "publisher" and "ad server".
As a result, Facebook does not maintain material business partner relationships with organisations that are part of the transactional chain associated with serving, capturing ad actions or enriching ad impressions or audience measurement.
Additionally, Facebook does not "acquire" or purchase traffic.
Filtration methodology
Facebook employs techniques based on identifiers, activity and patterns based on data in the log files in an attempt to detect and remove invalid activity, including, but not limited to, known and suspected non-human activity and suspected invalid human activity. Browsers that can be identified as bots are excluded from being able to log billable impressions. However, because identification and intent cannot always be known or discerned by the publisher, advertiser or their respective agents, it is unlikely that all invalid activity can be identified and removed from the report results.
Robot instruction files
Facebook uses robots.txt files to instruct robots that they should not follow or index any links within the domain from which the robots.txt file was retrieved. The robot instruction files are present in all tracking web server instances.
Identification of non-human activity
Facebook uses the IAB spiders/robots blacklist to exclude non-human and invalid activities. The IAB lists are reviewed monthly and a Facebook user agent blacklist is updated accordingly. Facebook blocks requests from null (blank) user agents and employs the IAB white-list to filter ad impressions that were not generated from a known browser.
Internal traffic
Facebook filters impressions from Facebook internal locations. Facebook compares an IP address against the entire list of Facebook IP addresses, which is maintained by the Facebook operations teams. The Facebook operations team adds/removes IP ranges where Facebook stores internal traffic generated when it tests its production systems under test campaigns. The test data is stored in a test database and is completely excluded from customer data.
Activity-based filtration
Activity-based filtration is determined through analysis of incoming events in correlation with the identified user source and window of time in which the events occurred. If incoming events from a single user source exceed a defined threshold within a given amount of time, subsequent events will be logged but will be marked as non-billable and will not be reported.
Invalid impressions detection and filtration
Facebook has a number of systems in place to detect and remove General Invalid Traffic, e.g. impressions generated by bots or spiders etc., as well as auto-checking for adware, malware, spam accounts, fake accounts etc.
The filters and thresholds used by Facebook to remove invalid traffic are constantly being monitored and reviewed. The thresholds are updated as needed.
The tools and methods to identify fake and malicious accounts are continuously monitored and evaluated by dedicated teams at Facebook. Formal reviews are performed, in which the accuracy and precision of their methods is evaluated.
The methods used by Facebook are executed both at the impression level and the user level (i.e. a given impression may be marked as non-billable or a given user may be removed from the network if deemed malicious).
Data consistency and accuracy filtration
In addition to non-human and activity-based filtration, Facebook includes various checks to ensure the accuracy and consistency of the data, such as:
- Time check: Impressions that originated more than a certain time frame are excluded from further processing.
- Activity frequency capping: Impression requests that exceed a certain number per hour are marked as non-billable.
The filters described above affect impression counts and will help distinguish between authentic traffic (a person at a computer who is actively engaging with ads) and invalid or non-human traffic, such as bots and spiders.
Data reporting
The Facebook Help Centre has a section on Reporting within Ads Manager, which includes the list of preset reports, available columns for reporting and how to schedule delivery of reports. Please refer to Ads Reporting in the Facebook Help Centre for more information.
Pre-release quality assurance
Facebook has developed tools to run end-to-end data integrity and consistency checks for quality assurance purposes. Furthermore, Facebook has a set of back-office tools that enable the Facebook staff to monitor activity trends. For example, the streaming process is designed to identify and log for analysis anomalies or corruption in processed data. In addition, Facebook runs checks between raw log files and the data warehouses to detect any discrepancies between the data sets.
Data re-issue/ad refund
If Facebook detects or otherwise becomes aware of issues that impact advertiser data or spend, it follows a formal investigation process in order to determine the cause and impact of such issue. Decisions to provide credit or refunds are made on a case-by-case basis and are dependent on the facts and circumstances of the particular issue. In the event that a credit or refund is provided, advertisers are notified through our platform; however, there are circumstances in which the sales team will contact the advertiser directly.
Data retention policies
Facebook's impression data retention policies vary but comply with the minimum standard required by these guidelines. Customers can run and download reports against this data, though they are responsible for storing older reporting data. This policy is subject to change.
* Nguồn: Facebook