A Novel Text Classification Method for Emergency Event Detection on Social Media

• Published in: 2018 IEEE SmartWorld, Ubiquitous Intelligence & Computing, Advanced & Trusted Computing, Scalable Computing & Communications, Cloud & Big Data Computing, Internet of People and Smart City Innovation (SmartWorld/SCALCOM/UIC/ATC/CBDCom/IOP/SCI) pp. 1106 - 1111
• Main Authors: Liu, Yanfang, Niu, Jianwei, Zhao, Qingjuan, Lv, Jianghua, Ma, Shilong
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.10.2018
• Subjects: event detection; one class SVM; social media; text classification; variable n gram
• DOI: 10.1109/SmartWorld.2018.00192

Currently event detection based on social media becomes a more and more active research field. Text classification is a challenging task in event detection on social media because a large number of irrelevant texts are provided at the same time of collecting relevant texts. Most of existing text classification methods are based on the supervised machine learning which requires labeled and balanced training dataset. However, it is very difficult or even impossible for emergency event detection to collect balanced and labeled training data. In this paper, we propose a novel text classification method based on one-class Support Vector Machine (SVM) for emergency event detection, which takes only irrelevant texts as training data to detect the relevant texts to the emergency event. More importantly, we propose to extract variable length n-gram features for representing texts that takes the sequential relations and semantic relations among words into account. Finally we conduct two set of experiments to evaluate our method and experimental results show our method can perform text classification for detecting emergency event on social media with a comparable accuracy.