Location Oriented Phrase Detection in Microblogs

Saeid Hosseini, Sayan Unankard, Xiaofang Zhou, Shazia Wasim Sadiq

Abstract: As a successful micro-blogging service, Twitter has demonstrated unprecedented popularity and international reach. Location extraction from micro-blogs (tweets) on this domain is an important challenge and can harness noisy but rich contents. Extracting location information can enable a variety of applications such as query-by-location, local advertising, crises awareness and also systems designed to provide information about events, points of interests (POIs) and landmarks. Considering the high throughput rate in Twitter space, we propose an approach to detect location-oriented phrases solely relying on tweet contents. The system finds associated phrases dedicated to each specific scalable geographical area. We have evaluated our approach based on real-world Twitter dataset from Australia. We conducted a comprehensive comparison between strong local terms (uni-word) and phrases (multi-words). Our experiments verify the system’s capabilities using multiple trending baselines and demonstrate that our phrase based approach can better specify locality instead of words.