Detecting Presence of Personal Events in Twitter Streams

Social media has become a prime place where many users announce their personal events, such as getting married, graduating, or having a baby, to name a few. It is common for users to post about such events and receive attention from their friends. Such events are often sought after by social platforms to enrich users timelines, to create life-log videos, to personalize ads, etc. One important step towards accurately identifying an event is learning the signals that indicate the presence of such events. In this paper we generate an event/non-event classification model using a mixture of content and interaction features. We experiment with two categories of interaction features; activity, and attention, and reached a Precision of 56 % and 83 % respectively, demonstrating the higher importance of attention features in personal event detection.