Real-Time Local Word Database Construction from Twitter

Recently, geotagged posts to social media such as Twitter have been used to automatically construct a geographical dictionary containing diverse types of local words which indicate specific locations in the real world. The existing methods typically examine the spatial locality of the usage patterns observed in the geotagged posts accumulated for a certain period of time to select the local words; however, how long the geotagged posts need to be accumulated depends on the usage frequency of the word, and additionally, some local words can indicate different locations at different times. Thus, we propose a real-time method for constructing a local word database which consistently keeps the local words and their locations up to date by iteratively adding new local words, removing old temporary local words, and updating the locations indicated by the local words. These functions are realized by adaptively recording/resetting the usage history of each word to properly examine its spatial locality and by assigning the weight for each geotag which is used to represent the locations indicated by the local words according to their temporal variations. The local word database constructed by our proposed method was verified to contain more up-todate local words and locations compared to other types of geographical dictionary constructed by experts, crowdsourcing, and from the geotagged tweets accumulated for a fixed period of time based on the performance evaluations of tweet location estimation as an example of applications utilizing the geographical dictionaries.