Researchers Develop a Matchmaking Program

Online dating

Results obtained by researchers mining data from a major online dating service in China may help people find the perfect match. 

02/09/2016
By Edwin L. Aguirre

Finding love over the Internet is a thriving industry here in the U.S., according to computer science Assoc. Prof. Benyuan Liu. Online dating services, such as Match.com, eHarmony.com and Tinder, have exploded in popularity over the years, generating more than $2 billion in revenues in 2014 alone. According to a recent survey, 40 million single people (out of 54 million) in the country have signed up with various online dating sites, and around 20 percent of currently committed romantic relationships began online.

Liu, along with UMass Lowell computer science Ph.D. graduate Peng Xia and Assoc. Prof. Cindy Chen and Northeastern University Asst. Prof. Yizhou Sun, recently used data from Baihe.com, one of the leading dating websites in China with 60 million registered users, to develop a recommendation system that would help those seeking romance find a good match. Here are Liu’s comments about their findings:

Q. How do men and women differ in their online dating behavior? 

A. When it comes to looking for potential dates, men in particular tend to be focused on their own interests and are oblivious of their “attractiveness” toward their possible dates, while women are more conscious of their own attractiveness. Attractiveness in this context refers not only to the person’s physical characteristics, such as appearance, gender, age, height and weight, but also to his or her personal profile such as geographic location, occupation, income, education, religion, marital status, number of children, home and car ownership, hobbies, interests, ambitions, smoking and drinking habits and so forth. In China, people also pay attention to their respective astrological animal signs to check for compatibility. 

Q. When writing the algorithm for online dating, what data is the most useful in predicting a good match?  

A. It’s much more effective to make recommendations based on a user’s past activities, that is, who they have contacted and replied to, than on their profile information. These activities allow us to learn a lot about a user's taste and attractiveness. This also enables us to identify other users with similar taste and attractiveness so we can make recommendations based on the behavior of these “similar” users. Our algorithm takes into account the mutual interests and attractiveness in both directions.

Q. Which is more successful — online dating or meeting people the traditional way, through friends?

A. This is hard for us to gauge because of lack of data. The problem is that once people meet their partners and start a relationship, they go offline. Or, if they are not successful in finding a date, they get frustrated and they also go offline. It’s hard to keep track of people’s lives as they go through marriage and family.

However, a research paper published in the Proceedings of the National Academy of Sciences in 2013 states that online dating is, on average, “associated with slightly higher marital satisfaction and lower rates of marital break-up than meeting a spouse through traditional (offline) venues.”

Q. What’s the future of online dating?

A. Because of people’s busy lifestyles nowadays, many of them don’t have time to join the dating game the traditional way. Unlike social media such as Facebook, wherein connecting with friends’ friends can lead to serendipitous meeting/dating, online dating is a specialized, targeted and customized service. I believe this trend will continue and the market will keep growing.

Q. What prompted you to start on this online-dating research?

A. I always like to have real data for my research. It so happened that one of the founders of Baihe.com is a friend of mine in grad school, and he offered to give me access to their company’s dataset. We analyzed the profiles and behavior of 200,000 individuals who use the site so we can help the company design a better recommendation system. So far, my colleagues and I have published four papers using the dataset. 

Q. How can people access your program? Is it an app they can download online?

A. Unfortunately, the algorithm we developed will not be available for download. Our plan is to work with a major online dating service so the company can incorporate our program to improve the company’s recommendation system. This is a big project involving a lot of time and resources, but we’re hopeful this would happen in the near future.