Photofeeler algorithms and AI

Photofeeler uses sophisticated algorithms, artificial intelligence, and battle-tested engineering that accounts for every contingency in order to boost the reliability of every human opinion we gather.

Here are some of the ways we use these proprietary technologies on Photofeeler.


Vote quality

Real-time Voter Fraud Detection

Photofeeler detects all manners of voter fraud in real time so that poor-quality votes are thrown out before they reach the photo owner.

This includes everything from lazy, thoughtless voters to truly malicious bad actors, and it is done using a combination of advanced score distribution analysis and machine learning techniques invented by the Photofeeler team.

System Precautions

Photofeeler safeguards against repeat votes. That is, even if a test is paused and restarted again a year later, users who voted on it previously cannot vote on it again. The system also inhibits, limits, and shut downs multiple accounts by the same person.

Everywhere possible, we ward off situations that would decrease reliability. For example, if someone is running multiple tests at the same time, Photofeeler spaces out those photos in the voting queue. We even have checks and systems for handling false demographic information.

Accuracy-boosting AI

VoterStyles

The VoterStyles algorithm, invented by the Photofeeler team, accounts for the way that people vote differently from one another.

People vary in how they assess what looks smart, trustworthy, fun, etc. Beyond that, some people need a photo to be the worst they've ever seen to give a 0, and others hand out 0s liberally.

By taking information about how voters vote and factoring it into our scores in real time, we're able to dramatically boost the statistical value of every single opinion.

Increasingly, neural networks are being incorporated into this process in order to suss out the finest possible differences among voters.

Photofeeler-D3

Photofeeler-D3 is a neural network that gets its name for representing the three Dating-category traits on Photofeeler.

Photofeeler-D3 was trained on over 100 million Dating-category opinions on Photofeeler in order to predict perceived smarts, trustworthiness, and attractiveness from real-world photos or videos as well as 15 human opinions averaged together.

While still brand new, this accuracy-boosting AI is being rolled out into the platform to increase the statistical significance of human opinions even further, with gains most noticeable for tests with very few votes.


Need more info?

For more information about Photofeeler test results, we recommend reading:

Isn't the voting system easily gamed (making my results worthless)?
Is Photofeeler accurate?
What your results mean

To understand why results vary for different photos of the same person, check out:

Do You Look Different in Pictures Than in Real Life? Yes, and Here’s How
Are My Bad Selfies What I Actually Look Like? How Do People See Me?


Want information about Photofeeler as a whole, not just the technology? Try our About page.