Learning Topic Map from Large Scale Social Media Data

Abstract

Geo-tagged social media data are hugely generated every day. Those content provide rich sources to explore keywords and topics in any region of the world thanks to the widely popularized mobile internet services. The association between geographic regions and their keywords/topics in social media has raised a lot of research attentions. Such association provides important information to event prediction, source detection, and news propagation in various applications. For instance, the disaster control and management, and evaluation of sales marketing campaign are just two of the many examples. The association between regions and keywords/topics are analogous to that between geographic coordinates and spatial features, such as streets intersections, building compounds and so on in a conventional street map, and we propose a new model, called “topic map” to mimic such an analogy and to encode and represent text features with geographic regions that reside in Geo-tagged social media data. Applications based on topic map will be explored during my research, and extensions to temporal data will be investigated further.