The “Ehime Method”: Using Big Data to Support Matchmaking

14,000 Users Worth of Data

The Ehime Marriage Support Center is near the JR station serving Matsuyama, the prefectural capital of Ehime on Shikoku, Japan’s fourth largest island. Many men and women visit it every day to sit in the private booths at the back of the center’s office and use its computerized system to seek potential mates.

After entering search criteria like place of residence, age, height, and other desired qualities, a man peruses the list of matches that appear on a tablet screen. He clicks another button at the right of the screen to search for partners suggested by the system, which uses “big data” analysis to provide its recommendations.

The next search brings up a list of several women who hadn’t appeared with his original search criteria. Pressing the “details” button, the man can now see photos of the women on the list; he selects one candidate he would like to meet.

(From top) The Ehime Marriage Support Center aims to help users along the road to matrimony; individual booths allow visitors to perform their searches in privacy; the tablet interface that visitors use to narrow their searches. (© Kōda Hideyuki)

The center, commissioned by Ehime Prefecture, began marriage support operations in 2008. In 2011, it began offering support for face-to-face arranged meetings. In March 2015, the center began analyzing the information it had amassed on a cumulative 14,000 registered users—a total of more than 1.5 million data points, including the qualities that users were looking for in a mate, the sorts of people users had asked to meet, and the number of times meetings had not led to a relationship.

Hitting a Dead End Is the Start

So far, 435 couples have married after meeting through the center. There were just 177 marriages in the first four years of the center’s operation; during the three years after rolling out big data analysis, though, the center added 258 more couples to its total. With this kind of success, the “Ehime method” began attracting attention nationwide.

Center director Iwamaru Hirotake was the one who suggested using big data. Fourteen other prefectures, including Ibaraki and Tokushima, now use the system. In 2016, groups from 28 municipalities and local assemblies around Japan toured the center, which has also been featured on news programs in China, Vietnam, and other countries.

At private-sector matchmaking agencies or online matchmaking sites, users set their criteria and look for a person who will suit their preferences. They will then ask for a matchmaking meeting or participate in match-up parties. Alternatively, “supporters” who act as intermediaries will suggest matches based on searchers’ criteria or interests shared with prospective partners.

Like for-profit enterprises, the center holds match-up parties and allows users to search its database for a partner. One thing that differentiates it, though, is the use of big data to support match-making. As Iwamaru explains, “Usually, when someone asking for a meeting with a prospective partner is turned down, the process ends there. But the center’s system takes that as a cue to start the process anew. The higher the number of meeting requests that don’t pan out, the longer the list of possible matches the system recommends.”

Finding People You Like Who Like People Like You

How does it work? Take Mr. A, for example, who expresses a liking for Ms. B, who came up as a match in his search. He asks for a meeting but is turned down. This is usually where things end, and the man might have a few drinks to heal his wounded pride.

But it is at this point that big data comes into its own. While accumulating data on the type of woman that Mr. A likes, the system is also analyzing other people’s behavior.

First, the system will search for other men who asked for a meeting with Ms. B, grouping together users with preferences similar to Mr. A’s. It then extracts data on which women the men in that group had asked to meet in the past and uses them to draw up a priority “recommend” list of women who might be Mr. A’s type.

Simultaneously, the system produces another group of women with similar preferences in men to those in the first group. From among this second group, a subset that have expressed preferences for men similar to Mr. A are picked up and recommended to Mr. A as potential matches for his own preferences.

Big Data as a Matchmaker

Amazon and other online retailers have an “other customers who bought this item also viewed . . .” function. The center’s big data works in a similar way, but when it comes to human beings, both sides need to be interested in dating the other person. This means that the center’s system needs to do more than just offer follow-up recommendations that a user might like, even though they aren’t good matches for the first set of search terms; its strength lies in its ability to find someone in the candidate pool who might like the searcher in return.  The more meetings that are arranged but don’t work out, the more data on the traits those people like are accumulated, so the bigger the pool of prospective candidates grows.

Iwamaru continues: “Private-sector matchmakers believe that their main mission is to find the best partner possible for their clients. This builds their reputation and helps their bottom line. But we use big data to find ways to bring people together whose search parameters don’t necessarily match. That’s the biggest difference between our service and those companies. I think of our service as a matchmaking friend, who might suggest ‘How about that person? She may not be exactly who you’re looking for, but you might find something about her to like.’”

Why does the service go out of its way to match people with someone who isn’t what they’re looking for? The answer to that came from facts uncovered by analyzing big data.

Before the introduction of the Ehime method, the center’s marriage support service had seen some success. But there was criticism about using public money for matchmaking, so the center needed to improve its record. In fact, many users were on the verge of dropping out of the service. For example, one man who had spent three years hoping for an arranged meeting and had never had one acceptance was ready to throw in the towel.

Weaknesses of Search Parameters Exposed

Before big data, the chances of having an arranged meeting were 6% for men and 13% for women. However, among pairs that actually did meet, about half went on to develop a dating relationship. If an arranged meeting did take place, capable volunteer supporters were on hand to further things along.

“In other words,” says Iwamaru, “the key was getting to the arranged meeting stage. We felt that if the data gave us answers or hints as to why some users never got to that stage, we could boost our success rate. We also thought the data could encourage those who were burned out from trying to find a mate to give it another try.”

Experienced volunteers are on hand to support a couple following an arranged meeting. (© Ehime Marriage Support Center)

After concealing personal data, the center asked algorithm researcher Uno Takeaki of the National Institute of Informatics to analyze the behavior patterns of center users. He discovered that always using the same search criteria was the fatal misstep of users who were repeatedly unsuccessful in moving to the arranged meeting stage. Everyone, naturally, has personal preferences in the qualities they seek in a mate. That is certainly not a bad thing, but this search behavior highlights a weakness of computerized searches.

Big Data Helps Change Searchers’ Perspective

When the age criterion chosen is “twenties,” the system does not take the possible option of “someone who’s in his thirties but looks younger” into account. Likewise, it is either “yes,” “no” or “not a consideration” where smoking or drinking alcohol are concerned. Again, that does not consider the possibility that a person would accept a prospective partner’s smoking as long as it did not take place in the home. In fact, if married couples searched according to their preferred criteria, some things they do not like about their current partner would surely turn up. In this way, strictly defined computerized searches appear to throw up barriers to finding a partner to one’s overall liking.

“Tweaking the search criteria could substantially improve the odds, but somehow searchers can’t bring themselves to do that,” notes Iwamaru. “And for those past the age of thirty-five, the pool of prospective mates dwindles, yet the requirements get even more onerous. For example, a woman may be looking for a partner who owns a home or may concerned about the possibility of having to live with aging in-laws. These additional requirements can really diminish the chances of finding someone suitable.

“When we were testing the system, we asked a forty-year-old man who was looking for a woman in her twenties, but had never been successful in initiating a meeting, to use the system. The system suggested a woman slightly older than him, but the man responded positively and said right away that he’d like to meet her. As more similar successful examples followed, we gained confidence that the system could be useful. Big data helps people who are set on certain criteria realize that alternatives do exist.”

The success rate for arranged meetings rose to an average of 29% for both men and women—and, in the case of requests to meet someone recommended by big data, reached nearly 40%.

The system had another effect. Under the previous approach, there were around five requests by men for arranged meetings to every request made by a woman, but more women began asking to set up meetings after big data was adopted. Many women noted that they felt at ease requesting a meeting, seemingly feeling that since the big data system had made the choice of prospective partner for them, a possible rejection would not be so painful.

Opportunities on the Rise

With these types of user reactions in mind, this year the center added a bookmark feature to its tablets to allow users to keep track of “favorites.” Even though meetings with a view to dating might not materialize, the system keeps track of data on partner preferences and recommends prospects matching those criteria to the users. The system acquires about 500,000 new data points every year and its evolution is driven by this new information.

Will this system prove useful for reversing Japan’s singles rate and flagging birthrate? When a user requests a meeting, the other person only receives a notice that a request has been made; that individual has no way of knowing whether he or she came up in search results according to preselected criteria or was suggested by big data. Some people might be shocked to learn that it was the computerized system, not the prospective partner, that selected them.

In this day and age, when people are free to find a partner on their own, some people are concerned that the system may be promoting old-fashioned values that view marriage as a desirable objective. Even so, Iwamaru firmly believes in the center’s mission.

“It’s not everyone who has opportunities to meet members of the opposite sex and develop relationships. Ehime Prefecture, for example, has many sparsely populated islands. Some young men on farms there are surrounded by old people and say they haven’t had a chance to chat with a young woman in years. And in the industrialized eastern part of the prefecture, workplaces are male-heavy and workers have no time off on weekends, so some men don’t have a chance to meet anyone or go on a date.

“To people who wish to marry, it’s rather heartless to say ‘you’re on your own’ in the face of such social impediments. If citizens are in situations like that, I believe it’s up to the public authorities to do something to help. Big data is one alternative. I hope that people who want to marry will find a good partner and experience the joys of being in a supportive relationship.”