Relevancy assessment of tweets using supervised learning techniques: Mining emergency related tweets for automated relevancy classification

Abstract

Social media provides an abundance of information that can be vital to emergency services. Especially during large-scale emergencies and disasters this amount of information rises even more and emergency services struggle to find relevant information that can support their current operations. The approach described in this paper uses Twitter generated data from an incident in Ludwigshafen, Germany in October 2016 to evaluate machine learning approaches for the relevancy assessment of social media content during emergencies. Not only different classifiers, but also several vectorizers and the use of n-grams are regarded. It is found that machine learning approaches can achieve very good results in the automatic relevancy classification and offer techniques that provide realtime quality assessments to emergency-services.