中国科学院自动化研究所机构知识库

With the development of information technology and the Internet, especially the proliferation of mobile Internet, social media is a new tool and platform that allows people to create and share information. Its rapid development has attracted attentions from hundreds of millions worldwide users. Meanwhile, along with the advancement of geographic positioning technology, especially the popularity of smart mobile phones, location-based services have become mainstream applications. In this context, the hybrid of geo-location and social media forms georeferenced social media, which includes various types of social media services featured with geographical locations. Georeferenced social media that enables users to access and share information “anytime, anywhere“, has generated a huge amount of geo-tagged social media data, which is heterogeneous, multimodal, and spatio-temporal. There is a “knowledge gap” between the massive georeferenced social media data and user information needs. Therefore, how to effectively and efficiently conduct data mining to harvest knowledge for end-user services, becomes the key problem to the development of modern Internet.

Georeferenced social media contains geo-locations, users, and data, which interrelate with each other. On the one hand, users have generated massive geographical media data, which can be aggregated to mine knowledge for understanding geographical locations. On the other hand, users have contributed rich online social behavior data, which can be explored to understand users. In this thesis, we investigate the research on georeferenced social media mining and application, which aims to explore the user-sensed georeferenced social media data to harvest knowledge for understanding geo-locations and users, thus providing valuable applications and services. In particular, we have combined semantic understanding, knowledge mining and application services in a principled and unified framework to conduct the research work concerning the following three issues: • Geo-location computing. User-generated geographical data contains geograph
ic related knowledge and information of our world and human society. Geolocation computing aims to exploit the geo-tagged media content for modeling geo-locations and discovering geographic knowledge by combining geographical information and multimedia semantics in a unified way. The problems of geo-location computing include geographical location estimation, location theme discovery, multimodal aspect-opinion mining, and sentiment analysis. The harvested geographic knowledge can enable potential applications such as location recognition, visualization, and exploration. • User understanding. To truly bridge the gap between data and user information requirements, besides multimedia semantic analysis, another key issue is user interest and intent understanding. This research line attempts to explore rich user-contributed multimedia content and social behaviors to model users for harvesting user profiles and behavior patterns. The detailed research problems include user attribute inference and influence modeling. • Jointly modeling users and geo-locations. The knowledge discovered from georeferenced social media data should be integrated with appropriate application mechanisms to satisfy users’ needs. This research direction combines users, geo-locations, and data in a unified framework to mine knowledge of users and geo-locations, which is further incorporated into an established recommendation system to support location exploration.