USC researchers help refine eHarmony’s dating survey

eHarmony, a dating website, has recruited two researchers from the USC Information Sciences Institute to shorten the matchmaking process for its users.

Aram Galstyan, a research director for data science and machine learning at the Information Sciences Institute, and Greg Ver Steeg, a research assistant professor, are applying quantum mechanics to the eHarmony algorithm that assess personality traits and psychological qualities.

Consistent with eHarmony’s slogan, “#1 Trusted Dating Site for Like-Minded Singles,” the dating network matches users with compatible individuals through an algorithm. Upon creation of an account, the dating network prompts the users with a series of questions about their personalities and relationship preferences. The information gathered from this diagnostic is then used to build the user’s profile. eHarmony uses the completed profile to find psychologically harmonious matches, which show up in the user’s email the morning after.

eHarmony has decreased their survey from 500 questions to 145 questions, but Galstyan and Ver Steeg hope to get the number even lower in the future.

“The ultimate goal of our research is to develop efficient computational tools extracting useful knowledge from very high-dimensional, complex data,” Galstyan said. “Specifically, we assume that regularities in such data are often described by a small set of hidden factors, so our objective is to infer those factors. When applied to behavioral data such as psychological questionnaires, those factors describe things like personality types.”

According to eHarmony’s “Married Couples by the Numbers” report, 71 percent of female users and 69 percent of male users find their future spouse on eHarmony within a year of creating a profile. Five percent of marriages in the United States started as relationships through eHarmony. The divorce rate for these marriages is 3.86 percent while the national average is about 50 percent, according to the American Psychological Association. This is largely due to the preliminary questionnaire presented to the user. It acts as a curator, finding the right match for each user.

However, at 500 questions long, many users dislike the length of the questionnaire.

“A lot of people give up before they make it through the whole list,” Ver Steeg said. “For eHarmony, it means they lose a lot of customers. The question is how to find the best trade-off. What is the smallest set of questions that will identify a good set of potential partners?”

Cutting certain questions can risk omitting some of the user’s information that is necessary to successfully find a potentially long-lasting partner. This chance proved that there must be a shorter and more direct method to find the right partner while still upholding the comprehensiveness and success of the questionnaire.

Galstyan and Ver Steeg have produced a simplification through a process called Total Correlation Explanation, or CorEx.

“The ‘explanation’ is to find some hidden factor that would explain how all these relationships came about,” Ver Steeg said. “For instance, if you saw that the days your friends carry an umbrella to class are very correlated, you could posit a hidden factor, in this case ‘rain,’ that explains all the correlations you observed. A similar thing happens for questionnaires and surveys. How you respond to the question of how much you enjoy parties is correlated to how well you can strike up conversations. The ‘hidden factor’ that seems to explain your answers to these questions is usually called extroversion.”

A previous version of this article stated that Galstyan and Ver Steeg were responsible for the reduction in eHarmony’s survey questions from 500 to 145. They were not part of this reduction process. The Daily Trojan regrets the error.