‍

Introducing Anonymised Topics

In January 2022, Google announced Topics (and its API) as the Privacy Sandbox's primary solution for interest-based targeting within digital advertising. Today, we announce the release of Anonymised Topics - and we believe there are striking differences between them.

‍

‍

Google Topics Limitations

Whilst Google Topics brings understandable privacy, the capabilities for targeting - when compared with convention - are many and obvious:

‍

Website focus: The classifier will map topics to hostnames only, not full URLs, leading to over-generalisation. For instance, browsing the sports section of a news website might assign a "news" topic due to the overall domain. 

‍

Topic latency: The topics API will assign a web user 5 topics a week, but each user may have a different week start. Up to 3 topics will be passed to ad tech providers per call based on the last 3 weeks' activity, which can present a considerable delay between user action and user targeting.

‍

Taxonomy limitation: At present, only 350 categories are included, and although this may increase the concept will be limited by privacy considerations; as too many categories will lead to cross-site identification.

‍

Limited context: Topics lack deeper context about a user's intent. Someone browsing a recipe website might be interested in cooking or just researching for a school project - Google Topics would not be able to differentiate.

‍

Even with the positive privacy outcomes, these broad categories combined with the k-anonymity techniques do not offer advertisers the targeting capabilities they are accustomed to. In addition, there’s concern over the exclusive custodianship of the classifiers.

‍

‍

Advantages of Anonymised Topics

None of those limitations apply to Anonymised Topics. Our classifier works across all URLs, with an application layer on top of our existing NLP machine-learning models, that computes the statistical relevance of entities and keywords with one another, to suggest clusters or groups which are contextually and semantically related.

‍

Anonymised Topics connect the most appropriate keywords and entities in a semantically comprehensible group, maximising clarity, coherence, audience fit and reach with no restriction to the number of topics that can be assigned to a targeting cohort, and with over 16000 topics, at the time of writing‍ [the classifier adds more Topics as our publisher portfolio increases], there is no latency or taxonomy limitations to limit the capability.

‍

Anonymised Topics are born from the realisation that the contextual gap between IAB interest categories and NLP-extracted keywords and entities can be far too broad. When you process millions of pages, the list of keywords and entities grows exponentially, with tens of thousands of entries each delivering a limited audience size.

‍

As an example, typing "Samsung" for mobile phones, may result in a long list of keywords such as individual products or other unrelated concepts such as smartwatches and chip manufacturing. Sifting through these variations to select those that are both campaigns appropriate and deliver audience scale is both cumbersome and time-consuming.

‍

So Anonymsied Topics replaces this monotonous exercise with machine learning to calculate semantical coherence and understanding deep contextual relevance across thousands of entities and keywords to return results as clusters with contextual labels assigned. However, unlike interest categories that are broad and fixed, Anonymised Topics offer both flexibility as well as granularity.

‍

For example, as a marketer of mobile devices you may wish to target consumers with a high interest in the brand name “Samsung”, the results returned will be clearly labelled as follows:

‍

By choosing just the Android Smartphones topic classified as Technology & Computing you can curate a scaled-up device tech audience whilst confidently excluding those interested in stocks, shares, trading and legal issues.

‍

‍

When to use Anonymised Topics

Advertisers should use Anonymised Topics to build scalable bespoke audiences that combine the size of IAB interest categories with the granularity of keywords. Here are some of the examples:

• You wish to curate a highly-performing audience using specific interests with a high degree of granularity e.g. brand or product names.
• You wish to scale; an interest-based audience quickly without having to research and type individual keywords or entities that contribute to the interest.

• You receive a media brief and wish to accelerate the manual process of interpreting the audience criteria.
• You wish to utilise an organic Topic taxonomy which is not bound by any standardisation
• You wish to build a high-performing audience demonstrating a strong intent signal towards a brand or product type.

‍

How to use Anonymised Topics

1. Create a new Anonymised targeting segment
2. Select Topics under Interests
3. Type a search term and press enter
4. Choose the topics with the most appropriate classification, you may select more than one
5. Estimate the size of the audience
6. Click save if you wish to keep the segment, or activate to initiate the segment

‍

For example:

‍

Download the Anonymised Topics product sheet to learn more about this feature.