부산대학교 도서관

On online dating websites, a match is accomplished only when two users have mutual interests. Directed contact is a sign of users' interest and our research intends to provide recommendations by predicting whether two users will have reciprocal contact. We learn matching patterns by supervised machine learning and evaluate our framework on a large dataset extracted from a popular dating website. We propose a novel computational pipeline to extract features from both users' profiles and dating social networks to know users' basic information and their preferences. Our results show that both profile-based features and graph-based features can be used to effectively predict whether two users are likely to make reciprocal contact. Best results are achieved when using all of the features as the input to a Support Vector Machine (SVM) and its Matthews correlation coefficient is 0.372. We also prove that users' facial features provide effective information about them and can be used to predict whether two users match each other.