Classification of Incident-related Tweets: Exploiting Word and Sentence Embeddings

Abstract

In this paper, we present our five approaches submitted to the Text REtrieval Conference (TREC) Incident Streams (IS) 2019B edition. The goal is to classify crisis-related tweets into a variable set of information classes and to provide an importance score. This multi-class, multi-label and multi-task problem turns out to be even more challenging because of extremely unbalanced training data available. We use recently proposed, publicy available word and sentence embeddings and deep neural network models for this task.