FAST COMPANY - June 25 - A new deep learning algorithm called CoupleNet analyzes the tweets of two users and estimates how compatible the duo would be. To train the algorithm, the team behind CoupleNet created a dataset of ~2M tweets pulled from ~4,600 likely couples. The "couples" were found by identifying public accounts who sent messages like "good night baby love you :-*" to each other. After training, CoupleNet could identify real couples in the testing data better than other recommendation tools that were given the same data. The algorithm was able to spot what mutual interests between users it thought suggested a good match, like tweets about K-pop bands. But it's more than just word matching. CoupleNet could find pairs correctly even when there are no obvious matching signals.