The recommendations of the CNIL will come into force on 1 April 2021, putting new barriers to the collection of personal data from Internet users. In order to offer them relevant advertising messages, contextual targeting is an innovative solution. In particular, because it uses "deep learning" tools that enable it to constantly optimize its performance.
The deadline is getting closer: as of April 1st, the famous "by continuing to browse this site, I accept the use of my data" will no longer appear. The user's consent will have to be collected separately, and will be uncorrelated to his browsing on a site. Moreover, he or she may withdraw his or her consent at any time.
This new CNIL regulation is part of a long-term, global and probably irremediable logic: the personal data of Internet users will become increasingly protected, and sending them targeted advertising messages will prove to be increasingly difficult. To counter this evolution, the web giants are each developing their own proprietary - and competing - solutions to continue, in spite of everything, to offer advertisers alternative targeting solutions. But for media sites, the equation looks more complicated: they cannot live without advertising, and all do not have sufficient means (nor do they have enough data "on their own") to develop their own advertising solutions.
In this context, there is growing interest in contextual targeting, i.e. targeting that is linked to the content consulted by the user, and not to his or her personal data. At first glance, the principle may seem relatively basic: displaying, for example, a campaign for a mascara on a site dedicated to make-up seems quite intuitive, and this is the principle that traditional media have been implementing since the 19th century.
However, when transposed to the digital world and its technological possibilities, contextual targeting proves to be much more sophisticated than a page of advertising in a magazine. Not least because digital contextual targeting uses tools built on artificial intelligence. This is what allows it to significantly refine its approach, and the very notion of context.
The most recent "deep learning" tools allow, based on the analysis of key words present on a web page, to understand in which context these words are used, what links them together, and finally what overall message is delivered by the page: a positive or negative feeling, a more or less thorough analysis of a subject... by linking these keywords to other data (is it a frequently requested keyword, what images illustrate it, who is the author of the text...) we can build not only a very precise vision of the context, but also a system of vectors: the latter, which resembles a neural network, enables the database of analysed keywords to be constantly enriched.
Finally, the tool will understand, beyond words, which concepts, which notions, are present in the content. This will enable it to deduce two important elements: firstly, to define to which advertising categories (family, health, entertainment...) the content can be linked. Secondly, the tool will be able to determine to which audience segment the individual who consults this content belongs: without using personal data or cookies, it will be able to determine the user's age group, gender and interests. This will avoid, for example, offering an advertisement for an industrial pizza to a fan of organic food.
Refining the analysis in this way makes it possible to offer each user advertising content that is more relevant, and therefore better accepted: artificial intelligence chooses an advertising message that usefully complements the content consulted, without redundancy or intrusion, and without "following" the user in his future browsing. It focuses on the moment when the message will be useful to him. As a result, the campaigns proposed in this framework achieve performances that are around 30% higher than traditional digital advertising. In the future, these semantic analysis tools will be able to "understand", in addition to words, the concepts developed in video or audio format, and to offer an ever finer understanding of the meaning of each content.
This article was first published on CB News by Jean-Philippe Caste. Find the original article here.