Event Detection in Social Media - Detecting News Events from the Twitter Stream in Real-Time

Abstract

The proliferation of social media and user-generated content in the Web has opened new opportunities for detecting and disseminating information quickly. The Twitter stream is one large source of information, but the magnitude of tweets posted and the noisy nature of its content makes the harvesting of knowledge from Twitter very hard.

Aiming at overcoming some of the challenges and extract some of the hidden information, this thesis proposes a system for real-time detection of news events from the Twitter stream. The first step of our approach is to let a classifier, based on an Artificial Neural Network and deep learning, detect news relevant tweets from the stream. Next, a novel streaming data clustering algorithm is applied to the detected news tweets to form news events. Finally, the events of highest interest is retrieved based on events' sizes and rapid growth in tweet frequencies, before the news events are presented and visualized in a web user interface.

We evaluate the proposed system on a large, publicly available corpus of annotated news events from Twitter. As part of the evaluation, we compare our approach with a related state-of-the-art solution. Overall, our experiments and user-based evaluation show that our approach on detecting current (real) news events delivers state-of-the-art performance.