#Unconfirmed: Classifying rumor stance in crisis-related social media messages

It is well-established that within crisis-related communications, rumors are likely to emerge. False rumors, i.e. misinformation, can be detrimental to crisis communication and response; it is therefore important not only to be able to identify messages that propagate rumors, but also corrections or denials of rumor content. In this work, we explore the task of automatically classifying rumor stances expressed in crisisrelated content posted on social media. Utilizing a dataset of over 4,300 manually coded tweets, we build a supervised machine learning model for this task, achieving an accuracy over 88% across a diverse set of rumors of different types.