Location-Based Temporal Burst Detection Using Outlier Factors in Geo-Tagged Tweets

Burst detection is one of the most popular techniques for extracting remarkable keywords in online social documents posted through social media. With the growing interest in geosocial media these days, many researchers are focusing on extracting geolocal keywords related to local topics and events from such social documents because of the increasing number of geo-annotated documents (e.g., geo-tagged tweets on Twitter). In our previous work, we proposed a method for identifying local temporal burstiness to detect local hot keywords considering a user's location. Our method is based on moving-average convergence/divergence(MACD)-histogram-based temporal burst detection, and the experiments indicated that our method can sensitively identify the local temporal burstinesses of keywords on the basis of public awareness. However, daily fluctuations in the occurrence rates of keywords have an effect on the qualities of location-based burst detection. To tackle this issue, we propose a new method for identifying local temporal burstiness using quartile-based outlier factors in this paper. Utilization of the quartile-based outlier factors allows for location-based burst detection to eliminate daily fluctuation phenomena. To evaluate the proposed method, we conducted experiments using actual geo-tagged tweets posted on Twitter. The experiments revealed that the proposed method can identify local temporal burstiness more sensitively compared with our previous method.