To 100 (Non-existed) Boys I've Dated Before
A Tinder Clone tells my frustrating dating experiences with 100 non-existed men whose dating profiles and stories are machine learning generated online.
What is this project?
Men all have a well-framed dating app bio. However, we all had a weird and frustrating dating story behind it. No matter what happened on our date, there's only one ending - we never talked again.
The project explores using existing datasets on the Internet and brings awareness of dating data privacy to the audience.
How to play with the project?
Swipe! Use it as a regular Tinder. If you swipe right, you will get my dating story with the man you matched. The chat stores all the men you connect with and the dating story behind them.
Tech Stack?
React, RunwayML, Tracey
Where do we get the data?
- Profile photos are from StyleGan2 for FFHQ dataset, one of NVIDIA Research Projects.
- Names are aggregated from the social security administration.
- Dating Bio is trained from Ok-Cupid profiles datasets from Kaggle, where contains around 60k users' dating profile information. From the open-ended questions, we picked the three most related ones, bio, fun fact, and "what are you looking for" Then, we trained the models with GPT2 using Runway ML.
- Dating stories are trained from the 'what are you looking for' and fun fact categories of the Ok-Cupid profiles dataset. We put the training result into Tracey, a context-free grammar, and generated the story with particular grammar.
