#1 - Raw
#2 - Filtered
#3 - PSD
By using the Muse headband, we are able to automate the autoswipe feature for Tinder solely based on brain wave data.
The Muse headband takes in brain EEG data which we can analyze further. The headband has four nodes located near the left ear, left forehead, right forehead, and right ear. These nodes collected raw EEG data every second. The raw data can be seen in image #1. Wearing the Muse headband, the user's EEG is being tracked, however in order for it to be useful to us, it must be filtered (see image #2). The algorithm runs the raw data through a FFT (Fast Fourier Transformation) to eliminate volatile data.
From there, the data is broken down into the different wavelengths - delta, theta, alpha, beta, and gamma, respectively lengthwise. This can be seen in image #3. By consulting with a neurotech engineer at midnight, we've determined that "attraction" at the most primitive level can be measured by extracting the alpha value from the PSD (Power spectral density). The next step would be to test this on a subject (a friend).
Next we started to explore relationships between these brainwave values and the subject's preferences. To do so, the subject viewed samples that they deemed attractive for 90 seconds. Their brain EEG data was recorded. Next, the subject looked at samples that they deemed unattractive for 90 seconds.With this experiment we were able to gather valuable data. We compared the ratios between the types of brainwaves for each node, the averages between nodes, and even the variances in our data. Later we instantiated a calibration portion to determine a baseline to compare our brainwaves to. After some fiddling with the numbers, alpha brainwaves that were below this calibration average are understood as the subject is attracted to a sample and alpha brainwave readings above the average is understood as the subject does not deem the sample attractive.
Later we decided to disregard the readings for all brain waves except for alpha waves because they seemed to be the most prominent indicator of attraction.
Problems we had
- Navigating an undocumented Tinder API
- Retrieving the Tinder Auth API by reading the auth_token logs from our sessions
- Creating an algorithm to determine how attracted you are to the person
things we've found
- When we're more calm and focused, the results are fairly accurate
Fun moment's we've had
- Validating our "swipe right" feature by matching with each other
- Receiving matches from confused men
- "Swiping left" on people we find attractive
- Finding out each other's tastes