Social media networks had become very popular recently. Statistica, the online statistics portal, estimated that there are 2.22 billion active social network users by the end of the year 2016 [1]. The same source estimates that this number will increase to 2.72 billion active social network user around the globe by the end of year 2019. Publishing personal contents has never been easier with the wide availability of microblogging platforms such as Twitter [2]. This has enabled users to post their opinions swiftly [3]. Recent research shows that Twitter process more than 500 million tweets daily [4]. The number of tweets that has been sent since 2006 when Twitter was founded is more than 300 billion tweets [5]. Data generated by social media users is huge in volume, grows at a very high velocity, varies in its type, and varies in its quality. These characteristics, also called the four V’s, i.e volume, velocity, variety, and veracity, are the main dimensions that characterize big data [6]. The availability of huge datasets representing more than a quarter of the world’s population who are actively interacting creates an opportunity to uncover patterns that could explain a lot of social phenomena [7], [8]. Meanwhile, the availability of these datasets introduces many challenges for researchers who are trying to analyze and process such data [7], [9].

۱٫۱٫ Research Problems

Social media networks are now considered as one of the major news channels. Mainstream media tend to monitor social media networks by looking for breaking news and interesting event. Furthermore, government entities are also relying on social media for the purpose of collecting security related intelligence. On the other hand, social network analysis focuses on individual users and their networks. The problem, i.e. the event detection on social media, attracted researchers attention recently because of the enormous popularity of social media. Existing approaches focus on features that doesn’t reflect the characteristic of the social network. Therefore, it fails to detect events in the context of the social network as a whole, which result in lower accuracy in detecting events. To address the problem, the temporal approach for processing a social network as we can detect an event from multiple temporal images. We define an event as an occurrence that has enough force and momentum that could create an observable change the shape of social network. We can measure such change by comparing the shape of data as time goes by. Therefore, if the shape of data at time t1 is different from the shape of data at time t2 we can conclude that there was a certain event that has an impact on the data and changed its shape. In this study, we show that processing temporal social networks graphs captures the complete complexity of the social network, which results in a higher accuracy of event detection model. We propose a temporal social network graphs event detection framework based on which we propose a novel social network transformation approach that transforms social media streams into temporal images. This allows for building a better event detection predictive model. We validate the proposed approach by performing experiments on streamed social media data collected for the purpose of this research. The ground truth collected data is extracted from mainstream media and labeled the dataset to create training and testing data. We achieve an accuracy rate in detecting events that surpasses existing approaches. We evaluate our proposed approach by using commonly used model evaluation metrics. Accuracy alone could be deceiving especially when data is imbalance. We calculated and compared precision, recall, and F1-score. We also used precision-recall and ROC curves to evaluate the performance of our proposed approach.