GPU-Accelerated Stream Clustering of Geotagged Text Data for Crisis Management

Social media services are endlessly producing large amounts of streaming data, and one of the most important ways of discovering and analyzing interesting trends in the data is through stream clustering. When clustering streaming data, it is crucial to access incoming data only once, and the clustering model should evolve over time, while not losing important feature statistics of the streaming data. We previously developed a stream clustering system for crisis management that clusters social media data based on their textual content and displays the clusters and their locations on an interactive map. Our system takes advantage of a text stream clustering algorithm called TextClust [6], which uses two-phase clustering, composed of micro-clustering and macro-clustering. The online micro-clustering phase incrementally creates microclusters that represent enough key information about topics occurring in the text stream. The off-line macro-clustering phase clusters micro-clusters for a user-specified time interval and can change macro-clustering algorithms dynamically. However, due to the sheer volume of an infinite data stream, stream clustering techniques on a single processor have inevitably hit their bottleneck. In this paper, we assess the benefits of utilizing the enormous computational power of the Graphics Processing Unit (GPU) to improve the performance of our micro-clustering process. We transformed our previous sequential clustering algorithm into a parallel version that can be executed efficiently on the GPU. Our experiments demonstrated that our GPU-based implementation has an average speedup of 2.9 when clustering multiple temporary micro-clusters.