The hard part about every recommendation system is getting good recommendations right from the start. Getting data to tackle this cold start problem is an incredibly hard problem that usually involves tradeoffs regarding privacy and/ or usability. Hence, we strive to use a new source of data with a high level of expressivity while reducing the user burden to a minimum.
What it does
HubsiClips gets the best movie and series recommendation for you. A neural network tracks your facial expression, while watching a variety of pre selected video clips. HubsiClips tracks your emotions, personal characteristics and user engagement. The aggregated data is matched with Burda's databank and a recommendation engine suggest you the best fitting match of movies and streaming services.
How we built it
Blood, sweat and tears. And lots of caffeine. The integrated camera tracks the user's facial expressions and sends the picture data to the backend for low latency processing. A state of the art machine learning model evaluates it and creates a personalized user profile. A matching engine then can merge the user profile and Burda media's streaming data set to find the best customized recommendation. The frontend receives backend's output and visualizes it for the user in a gamified and intuative way.
Challenges we ran into
The teammembers developed their workpackages on different laptop models. When merging the front and backend, we experienced issues with installing the other's libararies. This was lead back to different processor architectures.
Accomplishments that we're proud of
Amazing team collobaration with laser focus on the team's succes. The workpackage distribution in the beginning of the project was effortless and tailored to everyones skillset perfectly.
What we learned
Merge the backend and the frontend earlier to recognize problems earlier. Make sure to get every free merch there is, especially cool thsirts and hoodies ! Team work makes the dreamwork! And its a lot of fun.
What's next for HubsiClips
As a next step HubsiClips will take the iPhones facial 3D scanner into account. This will increase the input data's quality and will refine the facial interpretation greatly.