Lisa Pinero opens Tinder and looks through the profiles on her screen. After swiping through a few, she bores of the task. She closes the app, feeling her time was wasted.
Pinero is not the only Tinder user who has experienced this feeling.
In 2017, Charles Jekel noticed that it takes a lot of time for the average user to review Tinder profiles. Unless someone upgrades to Tinder Gold, he or she may swipe dozens of profiles before finding a match.
Jekel, a UF mechanical engineering doctoral student, decided to build a model to make a Tinder user’s experience faster. He presented his idea to the supervisor of his research group, Raphael T. Haftka. Haftka oversees Jekel’s structural optimization group which focuses on finding material model parameters.
“He really enjoyed the concept of fitting a model to assist my online dating experience,” said Jekel. “He gave me a few points I should consider,” Haftka said.
By the end of that year, Jekel released Tindetheus, a model that detects patterns in a Tinder user’s preferences through facial recognition and can like or dislike profiles for the user. Jekel presented the model to his research group and they received it well.
Jekel reviewed 8,545 profiles — which equates to about 85 hours of swiping — to train his model. He initially aimed to review 10,000 profiles, but a new addition changed his plans: Jekel met his girlfriend.
The couple met on Tinder about three weeks after Jekel began collecting data for his model. Jekel said he had to keep swiping because he was determined to finish his model.
“I told her about the project maybe a month into dating, and she got a big laugh out of it,” Jekel said.
Jekel said that a person interested in using the model could train the program by reviewing 20 profiles. The only requirement is that the person gives the model samples of both likes and dislikes.
Even though Jekel’s model would make browsing Tinder faster, Pinero, a 21-year-old UF biology junior, said she would not want to use a program that does the swiping for her.
“I just feel that a program can learn what I prefer, but ultimately, it can’t determine whether I’m going to like a person enough to establish a connection with them,” Pinero said.
Jekel said the Github page for the model has received around 300 downloads. Some have used it, but most people are just interested in how it works.
“My cousin has used it,” Jekel said. “I think every time he gets on Tinder, he realizes that he doesn’t have time for a girlfriend.”
Jekel said there’s room for improvement in his work, including the prediction accuracy.
He is uninterested in commercial development for the model but can see potential for apps.