Precautions for Data Processing - Dating App

Data Processing Processes and Risks

Byoungjoon Min
Precautions for Data Processing - Dating App

Finding a user's true needs in a sea of data is at the heart of data processing. In recent years, businesses have begun to emphasize the importance of this capability, and much of marketing is now using data processing to create services that appeal to users. Examples include curated shopping or movie recommendations, and another classic example is dating apps.

While the stigma of marriage is intensifying by the day, the willingness to date seems to be timeless. Spring is coming, which means peak season for dating apps as well as matchmakers. AI is struggling to keep up, too, as it claims to be the Cupid of many. But while that girl and guy on your app might be the good guy or gal thatAI matched you with based on big data, they might not even exist.



Crunching data to find your ideal match


Anyone who's been on a blind date can relate to the difficulties of arranging one. Even those who say they don't have a particular ideal date are often picky eaters when it comes down to it. "Do you want jajangmyeon or kimchi stew for lunch today?" "How about kimchi stew?" "No, actually, I wanted jajangmyeon." For people who couldn't or wouldn't reveal what their true ideal was, dating apps were likely to fail, but AI was able to uncover users' hidden tastes through the data processing process.


Launched in 2019, is a platform that develops and provides AI-based matching services.Based on technologies such as AI, algorithms, and big data analysis, it analyzes individual preferences and tastes. In addition, matching advisors and psychological counselors help people find a partner to further increase consumer confidence and matching rates. We operate a marriage information matching method using a Bi-LSTM-based preference prediction model, and have also completed a patent application for a service that finds a match through conversation keywords obtained through text analysis of voice and messages.

Glam's"Recommendation of the Day" feature also recommends matches based on machine learning. It measures and quantifies the attractiveness of users' profiles and provides personalized recommendations that take into account general preferences by gender, age, country, and culture, as well as the likelihood of a real connection. Glam's operator, Cupid, has also shown its willingness to share its know-how as much as possible with other companies in the social dating industry who are struggling with the technical aspects of building customer trust.


소개팅앱에서 찾은 데이터가공 과정과 리스크_01


Data processing methods


In the early days of dating apps, users would register a profile and use filters to screen and match their preferences. Nowadays, machine learning is used to analyze objective profile criteria such as occupation and age, as well as activity data on social media accounts, matching history, and the likability of conversations with the opposite sex. The goal is to increase the success rate by matching users with people who have similar subconscious tendencies, and even analyze DNA samples to match users with the most chemically compatible matches.

It's not just about meeting good people, it's also about coaching you to look good. While some of us spend weeks trying to pick the perfect photo for a dating app, few of us put much thought into how we write our introductions. In fact, Tinder noted that people who made two spelling mistakes in their About Me section received 14%fewer responses. That's a pretty big percentage of what makes someone like you, which is why they use AI-powered tools to correct typos and grammatical errors in profiles and messages.



The shadow of the data processing process


Is she 'real'?


Nowadays, it's possible to mass-produce images of people for use as models in online advertisements. Customers of Argentinian design firm Icons8 can enter their gender, age, ethnicity, eye or hair color, hairstyle, and facial expression to receive a photo of a person who doesn't actually exist. Based on the photos of70 real people, the company is able to generate 1 million virtual portraits a day using AI, from which the 100,000 highest quality photos are selected and distributed.


Generating virtual portraits is a recent trend among AI researchers. But what if these fake photos are being exploited by dating apps? If you're looking at a photo of someone you've been matched with and having a conversation with, but they don't exist in the real world, it can feel like an emotional betrayal and can be used against you in many ways.


This is why the social dating industry is working hard to catch malicious and fake users. Glam applies machine learning-based data processing technology to analyze information such as profile photos, occupation, country, and OS registered by users to detect the truthfulness of the information. The company has introduced a feature that identifies and blocks malicious users by determining whether photos are stolen, duplicate accounts, or impure activities in real time.Previously, administrators had to manually inspect, which was time-consuming and inaccurate. However, after introducing AI, the company said it was able to inspect user profiles for dating fraud, such as theft, spam, and money, with an accuracy rate of 99.5% within 5 seconds.


Meanwhile, 72% of users of social dating apps say they've been blocked by someone's misbehavior, offensive content, or profile, so many companies are looking to use AI to prevent user frustration. Tinder has a reporting system that asks users how they feel about inappropriate language and sends data when it detects it, and algorithms that can spot bot accounts with 99% accuracy based on IP addresses, messages, and stolen images are also on the rise.


소개팅앱에서 찾은 데이터가공 과정과 리스크
Serve my heart on valentine’s day


Dangerous organizers


AI solves problems by learning rules based on training data and then inferring similarities between the data it is fed. However, if there is a bias in the accumulated data, the AI performing the processing will inevitably have a confirmation bias. For example, if the accumulated data shows a higher preference for white people than black people, black people will be relegated to the back of the queue.While it's safe to say that AI matching services are the need of the hour in terms of meeting people who match your preferences, experts have pointed out that they can reinforce bias.

More recently, they've also been collecting data on your social media activity to improve your matching satisfaction. The idea is to find matches based on the likes you've given, photos you've posted on your SNS, places you've visited, and conversations you've had. While this is positive as an attempt to improve your chances of success, it also puts you at risk of "identity theft," a slang term for the indiscriminate and widespread collection of personal data to improve your odds. Deepfake technology, which utilizes "synthetic data," the latest big data trend, can mimic a person's face, body, and even voice and accent.


These are just a few of the dangers of AI-powered matching systems. However, depending on the application, studies have shown that AI performs best when it is biased in a certain direction rather than a neutral model. As a result, it is the quality of the data rather than the bias that is important for matching systems usingAI. If the quality of the data trained is such that it can meaningfully draw associations between the required data without including too much personal information, you'll get much more accurate results. This requires a high level of analytical power over the data you have.

Companies have no choice but to utilize AI that can provide high user satisfaction, even if it involves the risk of personal information leakage. In the process of conductingAI business, it will be important to have a high level of data analysis and collaboration to create quality data.


Talk to Expert