A novel method for geographical social event detection in social media

Popular microblogging service has attracted much attention around the world recently. With tremendous amount of tweets published each day, social event detection is becoming one of the most challenging research topics, especially for geographical social event. This paper proposes a novel geographical social event detection approach by mining geographical temporal pattern and analyzing the content of tweets. For the tweets published by users in the geographical area at each time unit, we first estimate its geographical temporal pattern based on the alternation regularity of tweets. Furthermore, we discovery the unusual geographical area by more frequent alternation of tweet count, and adopt adaptive K-means clustering algorithm for the tweets published in the geographical area. Finally, the geographical social event is detected by the number of the tweets in the cluster. We implement and validate our approach on realistic data collected from real-world social media websites. Experimental results show that our method can detect geographical social event with better performance than traditional methods. In addition, vivid demonstration of geographical social event can be effectively performed by our method.