Exploring cross-event relations on Twitter datasets via topic recommendation and word embedding

The ability to compute the degree of semantic similarity of real world events represented by social data and tracking the cross-event clues on a huge collection of social messages (i.e., tweets) has proven useful for a wide variety of event-awareness applications. The developed system should be able to overcome the challenge of high redundancy in social corpus (e.g. Twitter messages) and the sparsity inherent in their short texts. In this work, we propose a method to explore implicit relations on Twitter-based detected event datasets using an online event detection and word embedding technique for event analysis. The preliminary empirical result showed that the combined framework in our system is sensible for mining more unknown knowledge about event impacts.