Pinpoint Geolocation using Multi-Source Multi-Dimensional Big Data from Social Media
Abstract
Automatically discovering geographic locations at longitude-and-latitude level (pinpoint) from multi-source and multi-dimensional social media with limited geographic information is at the core of the multi-INT intelligence community. Early successes for geolocation have focused on the non-visual data, such as text, audio, or hyperlinks. Big heterogeneous metadata from current social space have greatly expanded the volume of information. However, there exists no comprehensive toolset that enables the intelligence analyst to quickly exploit and gain geographic information in the context of the massive heterogeneous collections, where timeliness is essential. Social media data differ from conventional laboratory data due to the fundamental challenges of distribution uncertainty, sample uncertainty, and label uncertainty. This research is to address these fundamental research challenges; systematically and rigorously formulate pinpoint geolocation problems under the real-world conditions when these uncertainties appear; leveraging the design of future social media computational systems by novel machine learning methodologies that can process large-scale, multi-source, multi-dimensional and multilabel data, allowing for applications to operate at their optimal or near-optimal performance levels. The success of this research at Northeastern University will lead to new approaches and publications, which will transform the field of multi-INT intelligence. Our method will actively query the user when it needs input in order to simultaneously achieve the learning goal and minimize the amount of user effort.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Aug 08, 2016
- Source ID
- N002441510041
Entities
People
- Yun Fu
Organizations
- Northeastern University
- United States Air Force