Embedding Time Differences in Context-sensitive Neural Networks for Learning Time to Event

We propose an effective context-sensitive neural model for the task of time to event (TTE) prediction, which aims to predict the amount of time to/from the occurrence of given events in streaming content. We investigate this problem in the context of a multi-task learning framework, which we enrich with time difference embeddings. To conduct this research, we develop a multi-genre dataset of English events about soccer competitions and academy awards ceremonies, as well as their relevant tweets obtained from Twitter. Our model is 1.4 and 3.3 hours more accurate than the current state-of-the-art model in estimating TTE on English and Dutch tweets respectively. We examine different aspects of our model to illustrate its source of improvement. 1