Title: Landslide Information Service Based on Composition of Physical and Social Sensors Aibek MusaevSchool of Computer ScienceCollege of ComputingGeorgia Institute of Technologyhttp://www.cc.gatech.edu/~amusaev3/ Date: Wednesday, November 4, 2015Time: 1pm - 3pm EDTLocation: KACB 3100 Committee:Dr. Calton Pu (Advisor), School of Computer Science, Georgia Institute of Technology Dr. Ling Liu, School of Computer Science, Georgia Institute of Technology of Technology Dr. Shamkant B. Navathe, School of Computer Science, Georgia Institute of Technology Dr. Edward R. Omiecinski, School of Computer Science, Georgia Institute of Technology Abstract:Modern world data come from an increasing number of sources, including data from physical sources like satellites and seismic sensors as well as social networks and web logs. The challenge is to analyze and integrate big data from multiple sources instead of studying each data source in isolation. Disaster detection is a representative real problem domain requiring a multi-source integration approach. Detection of natural disasters normally relies on dedicated physical sensors to detect specific events, e.g. using seismometers for early warning of earthquakes. However, there are some events for which there are no physical sensors, such as landslides. An alternative approach explores the big data from social networks, such as Twitter, functioning as social sensors. Nonetheless, despite some initial successes, social sensors have met serious limitations due to noise and lack of geo-tagged data. In this proposal, I will describe LITMUS — a landslide information service that combines data from both physical and social sensors by filtering and then joining the information flow from those sources based on their spatiotemporal features. LITMUS addresses two of the major research challenges in social network based analysis. Specifically, it filters out noise in the presence of irrelevant meanings of search keywords and estimates locations in the absence of geo-tagged data. In my presentation I will focus on the noise filtering problem and propose two classification approaches based on reduced explicit semantic analysis. Finally, I will discuss the extensions to my research, including multi-lingual support and improvements to the noise filtering component.