Mining the relation and implication of user generated content in social media

Jin, Xin

Permalink

https://hdl.handle.net/2142/34426 Copy

Description

Title

Mining the relation and implication of user generated content in social media

Author(s)

Jin, Xin

Issue Date

2012-09-18T21:16:14Z

Director of Research (if dissertation) or Advisor (if thesis)

Han, Jiawei

Doctoral Committee Chair(s)

Han, Jiawei

Committee Member(s)

• Abdelzaher, Tarek F.
• Zhai, ChengXiang
• Yu, Jie

Department of Study

Computer Science

Discipline

Computer Science

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Date of Ingest

2012-09-18T21:16:14Z

Keyword(s)

• social media
• clustering
• similarity
• ranking
• prediction

Abstract

The phenomenal success of social media sites, such as Facebook, Twitter, LinkedIn, Flickr and YouTube, not only revolutionized the way people communicate and think, but also revolutionized the way how corporations do business. During the current age of social media, web usage can be characterized as the decentralization of online information, which now largely consists of high volume and real-time content generated from the bottom-up, where common users are the contributors and producers of information. The transition from Web 1.0 (represented by static webpages instead of dynamic user-generated content) to Web 2.0 (represented by Social Media which consists of large scale of real-time and dynamic user-generated content), makes Internet information become in larger scale, richer, more interactive and complex. The goal of this thesis is to mine the relation and implication of user generated content in social media. In this thesis, I will present several studies that I have conducted on how to analyze such relation and implication. First, we proposed an approach for similarity computation based on both visual content and link information in social media by a novel way of mutual reinforcement of content similarity and link similarity. Second, we proposed a GAD (General Activity Detection) framework to fully explore the power of activity detection for clustering, which can be used to partition similar content objects into groups. The algorithms (both exact and approximate) developed within this framework can perform fast clustering for large scale content data. Social media content not only relate to each other, but also to outside phenomena and show strong implication with prediction power. In my third work, by aggregating user content information in social media, we developed a unified model to integrate clustering, ranking and regression for the prediction of stock price change.

Graduation Semester

2012-08

Permalink

http://hdl.handle.net/2142/34426

Copyright and License Information

Copyright 2012 Xin Jin

Owning Collections