"I'm eating a sandwich in Glasgow"

摘要: Social media such as Twitter generate large quantities of data about what a person is thinking and doing in particular location. We leverage this to build models locations improve our understanding user's geographic context. Understanding the context can turn enable variety services that allow us present information, recommend businesses services, place advertisements are relevant at hyper-local level.In paper we create language using coordinates extracted from geotagged data. model varying levels granularity, zip code country level. measure accuracy these by degree which predict location an individual tweet, further with user. find meet performance industry standard tool for predicting both tweet user country, state city levels, far exceed its level, achieving three- ten-fold increase