On information filtering in social sensing

For decades, from the invention of Sensor Networks, people envisioned a global sensing platform with millions of sensors deployed globally. The platform has finally become real recently with the advent of multiple online social network services where humans act as sensors and the social networks act as sensor networks, a practice named Social Sensing. Social sensing was born with the advances of high-level semantics sensing (since humans are the “sensors” with texts or photos as the sensing data) and (almost) zero-cost real-time data infrastructure, which makes this new sensing paradigm very promising in multiple real-world applications including disaster response and global event discovery. However, its global scale results in a massive amount of data generated and collected in applications that far exceeds normal people’s cognitive capability of information consumption, thus we desire a system that can filter the massive sensing data and delivers only information and intelligence to the users with a human-consumable amount. In this thesis, I focus on designing an information filtering system for social sensing; specifically, I focus on three levels of information filtering. In the first level, we focus on untruthful information removal, also known as fact-finding, where the challenge lies in the unknown reliability of each individual social sensor (i.e. human) a prior. In the second level, we focus on event-level information summary, also known as event detection, where the challenge lies in de-multiplexing different event instances and fusing social events detected in multiple social networks that previous approaches do not perform well. In the third level, we focus on information-maximizing data delivery to social sensing users, especially on redundancy removal by diversifying the information feed, where the challenge lies in algorithm design that not only works well empirically but also has a theoretical performance guarantee. We address the above challenges by algorithm design and system implementation and real-world data evaluations verify the efficiency of our proposed solutions.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2017

Author dc:creator

• Wang, Shiguang

Contributors dc:contributor

• Abdelzaher, Tarek F.
• Caccamo, Marco
• Han, Jiawei
• Kaplan, Lance M.

• Social sensing
• Information filtering

Statement dc:rights

• Copyright 2017 Shiguang Wang

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/97351

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/97351