GARM Brand Suitability: Developing a Collaborative Interpretation of the Standard

In recent years, brands are becoming increasingly interested in understanding the types of content their ads run in front of. This is especially true today, when many brands and agencies view aligning with suitable content as not just a ‘nice to have’, but also as part of their corporate responsibility. Until 2019, there was no standard way of defining this — leading to companies defining their own strategies and building their own opaque custom taxonomies. 

IAB Tech Lab’s Content Taxonomy has existed for a long time and provides the necessary basis to label content; however a standard way to define the level of risk a brand is going to take, or has already taken when advertising in a specific environment did not exist until 2019.

In 2019, the Global Alliance for Responsible Media, GARM, released the brand safety and suitability standard. This standard, spreads out 11 categories of content against 4 risk levels (low, medium, high, and floor) and was created to allow brands to have a clear and concise vocabulary to define their particular brand suitability needs. Tech Lab thus used this framework to evolve their content taxonomy to accommodate the GARM standards to help align the industry on the solution to this challenge, and we at Zefr applaud the Tech Lab for doing so. 

In July 2021, Tech Lab announced a new initiative: the Open Source Repository for the industry. Zefr has contributed to this initiative by providing rich data-sets based human reviews of video content against the GARM standards, to in turn help both the buy-side and the sell-side have a common source when “brand suitability” is being referred to in video environments. After extensive testing, we know that this data set is an effective tool when used as training data to build strong and robust machine learning systems to evaluate data signals for video content. We have embarked on this project to make this open-source data, once reviewed, accessible for the industry to build on top of, in an effort to improve the quality and consistency of the GARM framework applications for activation and verification.

While the GARM standard is leaps and bounds better than what was available previously (i.e. nothing), no one framework is comprehensive enough to be able to apply to all use cases and channels; and we found this to be especially true when dealing with user generated content (UGC). For that reason, this project is highly focused on this critical arena. UGC is highly challenging due to the nuance of the videos, and the scale at which it is consumed and uploaded. Simply relying on keyword analysis and Natural Language Processing (NLP) without human reviews leaves room for oversights like false positives and low precision. We acknowledge that this is wholly different from the open web, in which text is bountiful and descriptive from some of the greatest publishers in the world, and thus requires different toolsets and data.

When addressing complex challenges of building towards the GARM standards, it is important that all parts of the industry know the role they play in the solution. In this case, the existing GARM standard takes an initial top down approach by working with the buy-side to create brand suitability definitions; and while this is an important first step it is equally important to support the sell-side by providing content examples that match these definitions. This is a role that Zefr has played for the industry, by labeling a large volume of content, based on manual reviews, according to the GARM standards, and upon doing so we are contributing our work to the IAB Tech Lab Brand Suitability Testing Benchmark open source repository for the industry to use and benefit from.

Doing so enables two functions:

  1. It allows the maintainers of the GARM standards & taxonomies to refine their verbiage and categories based on reviews of actual content, making the framework tangible for all parties.
  2. It allows practitioners to have a golden labeled set publicly available to judge their own content understanding algorithms by. This means working towards steady progress in improving precision and recall of brand suitability technology, with a common data set serving as ground truth. This in turn helps to ensure that the adoption and roll out of the standards are progressing with the speed required for our industry, and prevents us from losing momentum by getting side-tracked by isolated brand suitability incidents.

The work that was done by the GARM committee is a major first step towards unifying the process by which brands communicate their suitability preferences. While the content itself will always evolve and grow based on current events, this project will allow for the beginning of a data set that helps inform technology solutions against a common policy and reviewed content.  

It’s time for the next step, which is to build on the framework and ensure that publishers, platforms, brands, agencies and tech providers work towards constant improvements based on complicated mediums. We, at Zefr, hope that this work we have contributed to the open source repository helps to move this conversation forward and can create innovations in the brand suitability space — tied to the GARM standards.

Author:

Jon Morra, Zefr

Jon Morra, PHD
Chief Data Scientist, Zefr