Bumble: Is Machine Learning the Future of Online Matchmaking?
Bumble: can online-dating apps use machine learning to substantially increase its ability to accurately matchmake and create values for its users?
Online dating overview (and Bumble)
As access to the Internet and mobile devices became increasingly prevalent across the globe in the last 20 years, online dating has become widely popular, socially accepted, and even essential for many urban professionals. Bumble, one of the new comers in the industry, operates similarly to Tinder where users will indicate their preferences for other users’ profile by swiping either to the left or to the right. The difference is that only female members can initiate conversations after matching, leading the “feminist movement” in the dating apps scene. 
The online dating industry amounts to 2.9 billion USD last year, and it is estimated that the current players only capture as little as 10% of singles worldwide, which I believe serve as a strong indicator of its potential growth.  As many have experiences, while online dating opened up the pool of candidates for chatting and dating, it has also created a platform for many disappointing experiences- both when the app is not correctly understanding your preference and sending you the matches you would liked, or when other members on the app are not acting respectfully, which causes users to drop out and become disillusioned with the idea of the online dating. This is where Machine Learning comes to play.
Machines make the best matchmakers
In the short term, in order to grow and retain users, the competitive landscape of the online dating industry is posing two important questions to Bumble. The first is to to make better matches and recommendations. Secondly, Bumble needs to better protect its community values on the platform by weeding out users who are disrespectful of others.
Some dating apps have already used big data to help users dynamically display their profile photo based on the number of “right swipes” to help maximize their chance of getting matches.  In my opinion, these improvements are tactical and short term focused and only scratches the surface of what Machine Learning can achieve. With Machine Learning technology, Bumble is able to significantly better understand your dating preference, not only through the profiles everyone create and the “interests” that you indicate, but also by digging out the implications and insights through a plethora of members’ mobile “fingerprints” by reading your swipe pattern, initiation rates of certain conversation, response time to messages. Because of the amount data that Bumble obtains, as well as the increasing processing speed of machine, Bumble has the potential of understanding your human heart and emotions even more than you do yourself, hence more efficiently serving the purpose of finding you the ”one.“
However, the ability for Bumble to capitalize on Machine Learning to improve its matching algorithm is much contingent on the size of the network and the amount of interactive data it obtains. Therefore, Bumble needs to better address issues with its customer experiences so that they can continuously grow its user base. Many users dropped out of Bumble after experiencing verbal abuse from other members. By design, because Bumble only allows female users to initiate conversations, the app is already filtering out many unwelcome messages that jeopardizes users experiences and causes user churn. However, the problem is not eradicated. Bumble can leverage Machine Learning capability to better understand the behavioral patterns from users. By understanding and verifying good behaviors, solely based on user’s interactive data on the platform, such as whether someone swipes judiciously or responds to messages appropriately, the system can more effectively predict and reward those that would help maintain the trustworthiness of the platform, hence building a virtuous cycle for scaling its network. 
In the long term, when Machine Learning technology is being developed, Bumble would need to focus even more on user’s privacy protection. Research has shown that users of online dating apps are generally more concerned about institutional privacy safety (social media companies selling personal data to third parties) than social privacy (others users see your information).  When machines can understand more about users preferences and the complexities of individual users’ sexuality expressions, companies need to do more about disclosing the privacy information to users and actively enforcing on strict procedural and technical methods to prevent these hyper sensitive information from being unlawfully extracted and revealed.
- What is the maximize capability for machines to capture the complexity of human sexual and emotional attraction? Research has indicated that machines, even after fully trained with some data, are not very good at predicting human attraction in experimental settings .
- As social media giant Facebook is also getting in the online dating real, how can Bumble and alikes fend off the competition where its competitor has 185 million daily active users in US and Canada alone.  Is Facebook’s entry an immediate threat to Bumble? Or is Facebook’s entry more of a industry wide validation?
- Charlotte Alter, “Whitney Wolfe Wants To Beat Tinder At Its Own Game,” Time, Http://Time.Com/3851583/Bumble-whitney-wolfe/, May 2015
- Leigh Gallagher, “Match Is The Sweetheart Of Online Dating—but Can It Fend Off Facebook And Bumble?”, Fortune Http://Fortune.Com/2018/06/27/Match-dating-tinder-facebook-bumble/, June 2018
- Ananya Bhattacharya, “Tinder Is A/B Testing Your Face”, Quartz, Https://Qz.Com/809681/Tinders-machine-learning-algorithms-can-now-serve-your-most-appealing-photos-to-potential-dates/,October 2016
- Christoph Lutz And Giulia Ranzini, “Where Dating Meets Data: Investigating Social And Institutional Privacy Concerns On Tinder”, Social Media + Society, January 2017
- Samantha Joe, Paul W. Eastwick, And Eli J. Finkel, L, “Is Romantic Desire Predictable? Machine Learning Applied To Initial Romantic Attraction”, Psychological Science, July 2017
- Kurt Wagner And Rani Molla, Facebook’s User Growth Has Hit A Wall, Recode, Https://Www.Recode.Net/2018/7/25/17614426/Facebook-fb-earnings-q2-2018-user-growth-troubles, July 2018
Word Count: 800
Student comments on Bumble: Is Machine Learning the Future of Online Matchmaking?
Thank you for shedding light on the role that ML plays in online dating. Your argument that Bumble has the potential to leverage its large amounts of data in order to truly understand an individual’s dating preferences and desires was very compelling. However, as you noted in your article, this argument hinges on Bumble’s ability to continue to collect that data and grow its user base. In addition, it is based on the underlying assumption that the insights from “your swipe pattern, initiation rates of certain conversations, and response time to messages” are correlated and linked to an individual’s romantic feelings for someone else.
That said, in response to your first question, I think that it is a stretch to argue that machines are able to capture the complexity of human sexual and emotional attraction. Even if Bumble was able to analyze your swipe pattern, I am not sure that it would be able to better predict who “the one” is for you. People can be incredibly indecisive and inconsistent in their decision making, particularly when it comes to romance. As a result, I would argue that ML can only take a company like Bumble so far.
The biggest concern I have for Bumble when it comes to using Machine Learning is the amount of data per user they can actually record. In order for Machine Learning to accurately predict who a user might want to connect with they need to provide the user more ways to interact with the platform. What is bumble currently optimizing for when the data only includes swipes and matches? Thus, Bumble may be forced to heavily alter their model or maybe offer a premium option which requires more information from users. This would allow more serious users to self-select into a new platform but also may dilute the product. Anecdotally, having seen some of my friends use bumble in the past it is apparent that some people “swipe right” on every person in order to save time and create more matches. This ultimately only provides useless data and perhaps demonstrates what many people are really seeking from their online dating experience, a good time.
AI vs. the Human Heart – I agree with John Smith’s point here about machines being able to rightly capture the complexity of human emotions. Along with security concerns that you rightly pointed out I believe that Artificial Intelligence is not neutral. The machine will feed off of the biases of its users and generalise recommendations based on that. The predictability of the matchmaking options also reduces the concept of ‘chance’ which is probably the more exciting part of dating. I wonder if these apps also have processes where it suggests options not so aligned with past interactions/interests data.
Regarding your second question, if I were Bumble (or any of the other apps that are focused exclusively on dating), I would be very concerned about Facebook as a competitive threat. Though I am not familiar with the details of Facebook’s dating app’s set-up, it strikes me that Facebook has some significant, immediate advantages:
1) Customer acquisition, which is critical to amass data to feed the machine learning algorithms, is not a concern for Facebook, as 185M already are active daily users and many more have a Facebook account. Basic information from a person’s Facebook account can be ported over to create the dating account.
2) When a successful match is made and a person decides to leave a dating app platform, that may be a long-term or permanent departure on Bumble. In contrast, Facebook dating app users will continue to use Facebook, and should those users decide to start using a dating app again, will be much more likely to use the Facebook dating app, as this brand/product will still be integrated into their lives.
These dynamics lead to Facebook having a) more data per user to leverage in the matchmaking process and b) more users / more active time per user on the platform, which are critical advantages in the machine learning / matching space.
This piece raises the really interesting question of whether machines will be able to become better than humans at something that is intensely personal. This reminds me of an episode of Black Mirror from the most recent season. If you’re unaware, Black Mirror is an eye-opening, thought-provoking series currently produced on Netflix which examines modern society and in particular, unintended consequences and impacts of new technologies. In the episode “Hang the DJ”, it follows two people in a dating simulation. A virtual simulation runs a 1,000 times to see if two people are compatible, which then shows up on their dating app in the real world. In the show, it follows two people who in the simulation match 998 times, providing their real world selves a probability of 99.8% that they are meant to be together. While an extreme extension of this, it raises the question of how machine learning and computer simulations can impact this area. What works well about the simulation model is it overcomes the challenge of determining what variables to analyze. By effectively being able to analyze “everything” they free themselves of the challenge of only analyzing swipes and likes.
To your second question, I agree that there are limits to the ability of machine learning to predict romantic relationships. In addition to those you mentioned, two reasons stand out to me.
First, I think that getting beyond swiping patterns into meet-up patterns and whether a relationship successfully formed would required Bumble to collect additional data from the user (e.g. pop-ups such as “Did you and X meet up? How did it go?”). In this case, Bumble becomes less a platform for connecting people and more a “relationship manager” which would likely feel intrusive to many, per the privacy concerns raised above. With privacy concerns, you will likely have fewer users, thus reducing the amount of data you have available. This tension between ‘scale’ of data versus ‘depth’ of data and ability to monetize that data is quite interesting.
Second, I think that because the human experience is always evolving, the proper “match” for someone at one point in their lives might not be the proper match at another point in time. I also think experiences with one person can change what you’re looking for in the next person in unpredictable ways.
Prachi also brings up a great point about “chance” – I hadn’t thought about that!