Directly to content
  1. Publishing |
  2. Search |
  3. Browse |
  4. Recent items rss |
  5. Open Access |
  6. Jur. Issues |
  7. DeutschClear Cookie - decide language by browser settings

Learning Social Links and Communities from Interaction, Topical, and Spatio-Temporal Information


PDF, English
Download (8MB) | Terms of use

Citation of documents: Please do not cite the URL that is displayed in your browser location input, instead use the DOI, URN or the persistent URL below, as we can guarantee their long-time accessibility.


The immense popularity of today's social networks has lead to the availability and accessibility of vast amounts of data created by users on a daily basis. Various types of information can be extracted from such data, for example, interactions among users, topics of user postings, and geographic locations of users. While most of the existing works on social network analysis, in particular those focusing on social links and communities, rely on explicit and static link structures among users, extracting knowledge from exploiting more features embedded in user-generated data is another important direction that only recently has gained more attention. Initial studies employing this approach show good results in terms of a better understanding latent interactions among users.

In the context of this dissertation, multiple features embedded in user-generated data are investigated to develop new models and algorithms for (1) revealing hidden social links between users and (2) extracting and analyzing dynamic feature-based communities in social networks.

We introduce two approaches for extracting and measuring interpretable and meaningful social links between users. One is based on the participation of users in threads of discussions. The other one relies on the social characteristics of users as reflected in their postings. A novel probabilistic model called rLinkTopic is developed to address the problem of extracting a new type of feature-based community called regional LinkTopic: a community of users that are geographically close to each other over time, have common interests indicated by the topical similarity of their postings, and are contextually linked to each other. Based on the rLinkTopic model, a comprehensive framework called ErLinkTopic is developed that allows to extract and capture complex changes in the features describing regional LinkTopic communities, for example, the community membership of users and topics of communities. Our framework provides a novel basis for important studies such as exploring social characteristics of users in geographic regions and predicting the evolution of user communities.

For each approach developed in this dissertation, extensive comparative experiments are conducted using data from real-world social networks to validate the proposed models and algorithms in terms of effectiveness and efficiency. The experimental results are further discussed in detail to show improvements over existing approaches and the applicability and advantages of our models in terms of learning social links and communities from user-generated data.

Item Type: Dissertation
Supervisor: Gertz, Prof. Dr. Michael
Date of thesis defense: 4 August 2014
Date Deposited: 01 Sep 2014 07:36
Date: 2014
Faculties / Institutes: The Faculty of Mathematics and Computer Science > Department of Computer Science
Subjects: 004 Data processing Computer science
Controlled Keywords: regional communities, probabilistic models, social networks
About | FAQ | Contact | Imprint |
OA-LogoDINI certificate 2013Logo der Open-Archives-Initiative