Identifying burstiness of local topic using location-based burst detection with a classifier technique

In recent years, a huge number of researchers have been paying attention for developing new data mining techniques to detect topics and events in posted data on social media sites. In this paper, we focus on geotagged tweets posted on the Twitter site, which are referred to as georeferenced documents including not only short text message but also posted time and location. Burstiness is one of the simplest and most effective criteria for identifying hot topics and events. Kleinberg's burst detection algorithm is the simplest but most remarkable algorithm for detecting bursts; however, it does not consider the localities of bursts in topics and events. In this paper, we propose a new method for identifying the burstiness of a local topic using the location-based burst detection algorithm with a classifier technique. The aim of this study is to develop a new location-based burst detection technique that can identify the local burstinesses of topics such as, natural disasters, weather, and diseases. To evaluate the proposed method, an actual sequence of batched georeferenced documents composed of crawling geotagged tweets was used. The proposed method could identify the local burstiness of a local topic related to weather in Japan.