Activity Prediction: A Twitter-based Exploration

Social media platforms allow users to share their messages with everyone else. In microblogs, e.g., Twitter, people mostly report on what they did, they talk about current activities, and mention things they plan to do in the near future. In this paper, we propose the task of activity prediction, that is, trying to establish a set of activities that are likely to become popular at a later time. We perform a small-scale initial experiment, in which we try to predict popular activities for the coming evening using Dutch Twitter data. Our experiment shows the feasibility and challenges of the task, with a simple method resulting in human-readable activities. This exploration also identifies several issues (e.g., temporal phrases and activity classification) that need to be addressed in future work.