Event Detection in Social Media Within the Arabic Language

Abstract

The rise of social media platforms makes it a valuable information source of recent events and users’ perspective towards them. Social media platforms have been recently exploited as a valuable source of information for event detection. Event detection, one of the information extraction aspects, involves identifying specified types of events in the text. The recent increase of real-world events number that is disseminated over Twitter, which is one of the most important communication platforms in recent years; has attracted researchers to utilize tweets for the event detection system. In this research, we introduce FloDusTA, Flood, Dust Storm, Traffic Accident Saudi Event which is a dataset of tweets that we have built for the purpose of developing an event detection system. The dataset contains tweets written in both Modern Standard Arabic and Saudi dialect. We focus on the flood, dust storm, and traffic accident events according to their significant influence on human life and economy in Saudi Arabia. FloDusTA, are built based on three main steps, data collection, data cleaning and filtering, and data labeling process. The tweets are labeled with four labels: flood, dust storm, traffic accident, and non-event. The necessity to detect flood, dust storm and traffic accident events effectively and extract events from highly noise tweet content is paramount importance. For such events, it is crucial to obtain good result for the task of detecting events tweets from non-related events tweets. To this aim, we investigate the effectiveness of dividing the problem of event detection into two classification steps. This study explores a two-step approach of performing event and non-event classification on FloDusTA and then classifying the resulted events into specific types of events. Two-step approach compares with one-step approach of doing one multiclass classification for detecting flood, dust storm, and traffic accident event. The experimental evaluation shows that the two-step event detection approach is promising.