Advertise-meant For You

Inspiration

In present day, social media platforms (Instagram, Google+, Facebook, etc.) create virtual identities for each user. They use these virtual identities to cater specific content towards users based on their interests. This model works better when the disparity between the virtual identity and the individual's actual identity is minimised. Our team noticed an untapped resource that could further decrease this margin: use of uploaded user images to better understand the person behind the screen.

What it does

Application is a proof of concept that demonstrates how uploaded images (ex: Instagram posts) can be used to tailor advertisements to specific users. It is not meant to be a full scale application; it allows account creation/login, viewing a feed and uploading images from your gallery. The feed (which for our purpose only consists of ads) becomes more specific to user interests as more images are uploaded.

How we built it

For our use case, we decided to focus on building an Android application. We used Firebase authentication and Firebase realtime database to manage users and their information. We used the Google Cloud Vision API to process and glean information from user uploaded images. For the purpose of this hack, we limited ourselves to focus on 8 categories (running, swimming, sports, vacation, music, camping, dance, cars). Advertisements are chosen from a small set of advertisements for each of the 8 categories (10 per category). Advertisement selection is purely based on the information gained from the uploaded images.

Challenges we ran into

One of our largest limiting factors were the significantly fewer labels available using the on-device version of the Google Cloud Vision API which restricted the granularity of the categories we focused on.

Accomplishments that we're proud of

We are pleased with the results and the fact that we have a running, functioning application that meets our initial goals. In its current state, our application can support multiple users and uniquely tailor advertisements to each of them.

What we learned

More than anything, we learnt the breadth of tools available to do interesting things with artificial intelligence and machine learning. We enjoyed the experience of playing with the plethora of services Google Cloud Platform offers. Furthermore, we learnt the importance of setting achievable goals as this was a very short timeframe, which required prioritisation to achieve a single goal.

What's next for Advertise-meant For You

Next steps would include scaling the number of categories by moving from the on device API to the cloud API. In addition, we would want to change the fixed small ad database that we have to a wider and more comprehensive database.This solution can be easily implemented into a new application or integrated into an existing platform.

Built With

• android
• android-studio
• computer-vision
• firebase
• gcp
• google-cloud
• google-cloud-vision
• java
• machine-learning