Relational Bibliographic Information Networks Huan Gui, Yizhou Sun, - - PowerPoint PPT Presentation

Modeling Topic Diffusion in Multi- Relational Bibliographic Information Networks

Huan Gui, Yizhou Sun, Jiawei Han, George Brova UIUC

Multi-relational Information Networks

• In the real word, objects are connected via

different types of relationships, forming multi- relational heterogeneous information networks

• E.g.

– in the bibliographic information network, researchers could be linked together via different types of relationships

• collaboration relationships, citation relationships, sharing

common co-authors, co-attending conferences, etc.

– In the social network case, people are connected

• via friendships, colleague relationships, family relationships,

etc.

Multi-relational Information Networks

Goal of this paper

• They address the problem of modeling

information diffusion in multi-relational information networks

– Propose multi-relational diffusion model

• Propose two models by extending the Linear Threshold

model

– Learn parameters of the diffusion model

• Learning from action log (a sequence of object set

recording when an object is activated)

• Using MLE

Dataset

• They extracted topics from papers’ titles and abstracts:

– 79 topics in DBLP dataset, and 30 topics in APS dataset, – study diffusion of these topics during selected periods when these topics have increasing popularity trends

Distributed Graph Summarization

Graph Summarization

• Give a compressed representation of the graph

Distributed graph processing systems

• Giraph: an open source implementation of

Pregel [8] proposed by Google

– This paper

• Others

– GraphLab: proposed by Carlos Guestrin – Trinity: A Distributed Graph Engine on a Memory Cloud [SIGMOD 2013] by Microsoft Research Asia

• Other distributed system in the database

– Hadoop: Google – Hyracks: by Michael Carey et al (ICDE 2011)

Algorithm

MapReduce Triangle Enumeration With Guarantees

Idea

• Divide graphs into multiple overlap partitions,

and distribute each partition to a mapper

• Based on TTP (Triangle Type Partition)

algorithm [CIKM 2013]

• Using multiple rounds to reduce the memory

cost

Contributions

• They propose Colored Triangle Type Partition

(CTTP), a multi-round MapReduce randomized algorithm for triangle enumeration

– Require rounds in the worst case

• E is the total number of edges
• m denotes the expected memory size of a reducer
• M the total available space.

– use M/E space per mapper, m space per reducer, and M words as total aggregate space

Results

They are the first to get the result for this graph

Component Detection in Directed Networks

Directional community

• They propose a novel concept of communities,

directional community

– nodes play two different roles, source and terminal, in a directed network

Proposed Methods

• They changed Markov Clustering (MCL) and its

variant, R-MCL methods

• Based on a simulation of stochastic flows on the

network

Case Study: Twitter

• Detecting Communities from Twitter Interaction

Network

– a directed edge from a source node to a terminal node is created if any of the following interactions happens

• retweet(forwards) a tweet
• reply to a tweet
• mention someone

Case Study: Twitter

• Source: post some tweets
• Terminal: spread the tweets

This hashtag represents the “No vull pagar” (“I don’t want to pay”) campaign, a protest in Catalonia at early April, 2012