How Facebook can trump Google in advertising
A major percent of internet advertising revenues are shared by Facebook and Google. Facebook makes it advertising money using re-targeting, and latent targeting on user profiles. Google uses an auction model built on search queries to fill its coffers. Search query is one of the top most actionable intents from the user on internet. Google has built an amazing business around it.
Facebook has got more or less monopoly around social communication. Social tweets and messages don’t have a merchandising intent, similar to search queries. Facebook has trickily used the re-targeting mechanism (Disclaimer: I patented the idea, before anyone has implemented it) to make the ads more actionable.
We will eventually be at a stage, where Facebook and Google will be fighting for the same ad dollars. Who has got a strategic advantage to win the war? In my view it will be Facebook.
Elaborating more, Facebook controls the user’s interests and influences across its social properties. Facebook can use the personal data from the users to predict search queries and information that the user will “Google” in the near future and make it as part of the user stream. You might ask, is it possible?
How can Facebook predict search queries before they happen? Facebook has got a search engine on its page and has got a partnership with Microsoft’s Bing. They have access to both what the user is doing at any given point of time, what their influences are and what their search queries will be through its popular properties and partners. They also have significant information on a user through their re-targeting program, about their activities outside of the social walls.
Using above data, one can use variations of Sequence to Sequence algorithms to give search queries. The input sequence can be the aggregated behavior. We can use social profile embeddings, image embeddings in the social stream, previous search queries, the location information as inputs for the Sequence to Sequence algorithm. We can us a Variational encoder for representing the input data. The output sequence can be a list of search queries that the user will type on Google. One can also pose the query prediction as a recommendation problem. We can train a wide and deep neural net on the user’s data and search queries to predict search queries. We can also borrow techniques from Zero Query search engine techniques to do the predictions and generate information in Social Streams, so that users don’t have to go to Google to get information.
It would be a great win for users and Facebook, if they can stop the interruption on social browsing by 50%. Facebook can make money by asking the advertisers to bid on predicted search queries. It might be an easy sell, to the advertisers, with their relationships and engagement numbers.
If I were Google, I would be really scared of this possibility and eventuality (Most probably in the next two years). I would break the Facebook’s monopoly on communication as early as possible.
Disclaimer: My friends at Facebook and other social networks, if you decide to implement this idea, I would appreciate, if you can pay me royalty for the patent I filed with title “Advanced techniques to improve content presentation experiences for Businesses and Users”. Please don’t ignore legal notices from a poor innovator :).
Visit us at www.voicy.ai for more information.