Challenges we want to participate
- Open Web Technology
While no one enjoys ads, it's hard to disagree on their effectiveness. Just take a look at the business models of Google and Facebook and you'll be convinced.
But what if these ads could actually be used in a more beneficial way for you?
That is where our team comes in.
What it does
Ad-blockers remove ads altogether. We decided to keep the virtual billboards and replace the content by truly personalized messages. These could be reminders, calendar events, motivational quotes or even nice images. The real magic part is that we are able to guess what content shall be displayed in any specific context using zero-shot learning. In order to avoid stressing the user, we used our teams modest background in cognitive psychology to design our billboards in a minimal and non-invasive way.
How it works
The frontend is a chrome extension that is built on top of an adblocker. It grabs relevant information from the current page, lists all the available advertisement spots, and asks the background service to do the alchemy.
The background service was built using Python and provides an API over HTTPS. In its core it uses a zero-shot learning model that is capable of classifying text into a topic, which in our case means that it understands what the user is watching based on the content of the page so that it can recommend a personalized message.
Data we use
There are three main data sources that we use: the content of the page a user visits, his todo list and his calendar.
As described above, the page content is used to personalize the message we show but we also wanted to provide the user with information that is useful to her. Hence, we also fetch data from the user's Todoist and Google Calendar accounts and use it to customize our messages even more!
For example, if a user is visiting a sports website and has a "do laundry" entry in his todo list with 10PM due date, we would show him a "Sports are cool but don't forget to do laundry tonight!".
Even though our application only uses Todoist and Google Calendar at the moment, our modular architecture enables us to easily extended it to further sources.
Challenges we ran into
For this hack, we relied on different data sources and services: the tricky part was to make them all work nicely together.
Developing a Chrome Extension was easy, but the round-trip times to test changes were comparably slow. In addition, ads are not easy to work with.
Accomplishments that we're proud of
Remote collaboration did not stop us from producing crazy ideas. In fact, we wanted to prove otherwise.
What we learned
What's next for AdMotiv
We only tackled a fragment of the possibilities that are offered by machine learning to create personalized content. Also, the more apps the user connects the better the content we can provide. Moving forward maintaining the privacy of our users will be in the center of our product.