基于LBSN用户生成短文本的细粒度位置推测技术 (Fine-grained Geolocalisation of User Generated Short Text Based on LBSN)

Recently, the fine-grained geolocalisation of User Generated Short Text (UGST) has been attracting much attention from academia. Most of existing methods rarely introduce the semantic information about a location in UGST, and do not prioritize the entities according to their importance. These reduce the performance of existing approaches. To tackle these problems, we propose a Fine-grained Geolocalisation of user-generated Short Text based on LBSN (FGSTL), which consists of three key components: 1) Using Foursquare as a source to build the tight coupling between entity and location, which can address the location-annotated sparseness problem. 2) Filtering out UGST if it does not contain any location-specific entities, which allows us to eliminate the interference of noisy UGSTs at the early stage. 3) Ranking the candidate locations for each remaining UGST based only on its textual data, and selecting the top-ranked location ( or top n locations ) for UGST. The experimental results show the effectiveness of FGST-L.