research
My research interests include problems relating to dynamic multi-relational social network analysis – in particular, community dynamics, social information summarization and representation. My current research, community discovery in dynamic, rich media social networks, focuses on extracting human communities that collaborate around certain topics or shared media artifacts. I have applied community analysis to summarization of time-evolving, social media stream that enables exploring and searching shared media and tracking community interests.
I have also conducted work on spam blog ("splog") detection based on link connectivity and temporal properties of blogs. There I combined traditional content based features with temporal and link signatures with excellent results.
Prior to my Ph.D. study, I was interested in the area of computer graphics, with a particular concentration on non-photorealistic rendering. In my master thesis, I proposed a rendering framework that renders three-dimensional models in a synthetic Chinese painting style.
See also: released code and data.
community discovery in dynamic, rich media social networks
My work studies several aspects of community analysis in dynamic, rich media social networks: (1) mutual awareness, (2) transitive awareness, (3) community evolution, and (4) multi-relational community analysis.
 |
multi-relational community analysisHow can we extract communities from data of interactions with rich contexts and track the dynamics of membership and topics of interests within communities? We propose MetaFac, a graph-based multi-tensor factorization framework for analyzing the dynamics of heterogeneous social networks.
see also:
|
|
 |
community evolutionHow can we extract sustained communities in dynamic networks and analyze the evolution of these communicates? We introduce FacetNet, a probabilistic generative model to analyze communities and their evolutions in a unified process.
see also:
|
|
 |
transitive awareness and community dynamicsWe capture the amount of mutual awareness expanding on the entire network using a random walk based distance measure and propose an efficient iterative mutual awareness expansion algorithm for community extraction based on this distance measure.
see also:
|
|
 |
mutual awareness - from actions to relationHow do we discover communities from online social actions? We propose mutual awareness as fundamental property of a community, which is computationally defined by contextual use of links in blog media. The effectiveness of mutual awareness is verified on two different blog datasets.
see also:
|
|
summarization of time-evolving, social media stream
 |
temporal patterns in shared media streamsHow can we characterize time-evolving patterns in community-shared media? We propose a joint matrix factorization framework that incorporates image content features and contextual information to discover distinct temporal patterns of group photo streams.
see also:
|
|
 |
summarization of media shared activitiesHow do we summarize community generated media content that allows exploring shared media and tracking community interests? Our methods include a novel social media summarization framework that leverages a syntactic structure of the social activity in a multi-graph mining algorithm, which extracts community activities along multiple important facets (who, what, when, etc.)
see also:
|
|
 |
community trajectoryHow can we reflect on our everyday collaboration? Understanding activities of one’s peers become particularly difficult when people collaborate across time, locations or even disciplines. We propose to support collaboration by extracting and representing collaborative patterns, thus improve awareness.
see also:
|
|
 |
blog antennaHow can we capture the content dynamics in a blog? We propose a framework to analyze and summarize the temporal dynamics within personal blogs, based on self-similarity matrix factorization of blog content. Summaries based on large real-world blog datasets reveals interesting temporal characteristics for four blog types – personal blog, cooperative blog, power blog and spam blogs.
see also:
|
|
archived projects
 |
adversarial information retrieval - splog detectionI have also conducted work on spam blog (splog) detection based on link structure and temporal properties of blogs. Our methods combine traditional content based features with temporal and link signatures with excellent results.
see also:
|
|
 |
CG & animation - non-photorealistic renderingPrior to my Ph.D. study, I was interested in the area of computer graphics, with a particular concentration on non-photorealistic and artistic rendering. In my master thesis, I proposed a rendering framework that renders three-dimensional models in a synthetic Chinese painting style.
see also:
|
|