Every year, companies spend billions of dollars on market research and product evaluation. However, it is difficult for testers to provide an objective rating of how much they like or dislike a product, and extremely easy to instead provide biased answers. To solve this problem, we decided to use an EEG to objectively rate a users reaction to a given stimulus (video advertisement, movie, using a product). The EEG allows for an objective numerical analysis of a tester's thoughts, not influenced by variables like order of stimulus or other irrational cognitive biases.
What it does
Our project uses raw data from a NeuroSky-based EEG toy, MindFlex, to measure a user's reaction to a given piece of media over time. The intention of this is to quantify peoples' reactions to commercials, movies, ads, logos, and other forms of media which invokes emotions difficult to put into words. The system helps remove individuals'
How we built it
An Arduino processes the serial output of the NeuroSky chip and return an attentiveness scale, meditation scale, and the magnitudes of 8 different brain frequencies (Delta through High-Gamma). We then turn this data into Matlab matrices that we analyze and compare through wave comparison and magnitude similarity. The result is a zero to one value estimating how similar two reactions are.
Challenges we ran into
One of the largest challenges we ran into was in streaming usable data from a closed source headset. After hours of attempts, we abandoned the Bluetooth module route due to baud rate incompatibilities. Eventually, we settled on a wired headset and got it effectively communicating with a Processing sketch.
Accomplishments that we're proud of
We are especially proud of the hardware involved in our project, as we essentially modified a children's toy to provide raw data by looking at it's serial output data stream. This innovative usage of the MindFlex headset is the backbone of our solution.
What we learned
Our team improved our knowledge of serial communications and signal processing greatly over the course of the competition. Serial communications form the backbone of the hardware component, and getting the serial working taught us about how the underlying protocol worked. We tried many different signal processing metrics once our data was created, each test granting us insight into the field of signal processing.
What's next for DeterMind
With so limited time, we weren't able to implement neural networks and deep learning like we had hoped. However, we have the outline of a supervised neural network and given more time and data we hope this will improve the effectiveness of our data clustering in the future.