When a friend online is more than a friend in life: Intimate relationship prediction in microblogs

Microblogging services such as Twitter and Sina Weibo have been an important, if not indespensible, platform for people around the world to connect to one another. The rich content and user interactions on these platforms reveal insightful information about each user that are valuable for various real-life applications. In particular, user offline relationships, especially those intimate ones such as family members and couples, offer distinctive value for many business and social settings. In this study, we focus on using Sina Weibo to discover intimate offline relationships among users. The problem is uniquely interesting and challenging due to the difficulty in mining such sensitive and implicit knowledge across the online-offline boundary. We introduce deep learning approaches to this relationship identity problem and adopt an integrated model to capture features from both user profile and mention message. Our experiments on real data demonstrate the effectiveness of our approach. In addition, we present interesting findings from behavior between intimate users in terms of user features and interaction patterns.