Loading presentation...

Present Remotely

Send the link below via email or IM

Copy

Present to your audience

Start remote presentation

  • Invited audience members will follow you as you navigate and present
  • People invited to a presentation do not need a Prezi account
  • This link expires 10 minutes after you close the presentation
  • A maximum of 30 users can follow your presentation
  • Learn more about this feature in our knowledge base article

Do you really want to delete this prezi?

Neither you, nor the coeditors you shared it with will be able to recover it again.

DeleteCancel

Make your likes visible on Facebook?

Connect your Facebook account to Prezi and let your likes appear on your timeline.
You can change this under Settings & Account at any time.

No, thanks

Predicting Trending Messages and Diffusion Participants

No description
by

on 6 August 2014

Comments (0)

Please log in to add your comment.

Report abuse

Transcript of Predicting Trending Messages and Diffusion Participants

BIAN JINGWEN
FEB 26 2014

Predicting Trending Messages and Diffusion Participants
in Microblogging Network

Microblog Streaming
In this work, we presented a novel approach for predicting trending microblogs and the subsequent diffusion action in microblogging network.
Conclusion
Evaluation
Overview of the whole framework
Dataset
Questions?
Social network
Users' microblogging history
New microblog streaming
User Interest Profile Learning
Diffusion-targeted Influence Learning
Microblog Diffusion Modeling
Diffusion Prediction
Microblog Data
Texts
Images
External Knowledge
Multi-task transfer learning
Generally, a user will show different level of interest and possess different level of expertise for various interest categories.
Definition
:
Interest Profile.
A microblog could belong to various interest categories. The interest profile of a user represents the user's interest distribution over all interest categories.
Microblog data
Related interest category
Microblog classification problem.
Challenges
It is difficult to collect labeled training microblog data.
Microblog contains multimedia data.
Import external knowledge:
well-edited articles from portal websites.

Handle different data distribution:
Domain adaption.
Handle multimedia data:
Multi-Task Learning.
Two desirable properties for domain adaption.
Well-categorized: labels are available.
Contents cover nearly all aspects.
Rich multimedia information available.
Objective 1
: Distribution Matching.
Objective 2
: Locality Preserving.
Objective 3
: Multi-task learning.
Overall optimization problem:
Three types of influences
Interest-oriented Influence
Social-oriented Influence
Epidemic-oriented Influence
Interest-related feature function:
Social-related feature function:
Epidemic-related feature function:
Diffusion-targeted Influence Model
A factor graph model to distinguish and quantify the degree of the three types of influences.
With the learned influences between users, the next target is to utilize these influences for microblog diffusion analysis and prediction.
Configuration maximization to learn weighting matrix for all influences.
What we have: diffusion action history (who retweets what from whom.)
+ various influence between users.
One Problem to Consider
Negative sample selection.
User u did not retweet the microblog m posted by user v.
Active: u tweeted something
Only consider this time period.
Given a new coming microblog, predict whether it will become trending in the near future. Besides, also predict which of the users will participate in the diffusion process.
Use the
Independent Cascade
model.
Initial set of active nodes.
Each node is given a single chance to activate its inactivate neighbors.
If it fails, no further attempts to activate other nodes.
Predict whether a further diffusion action will be performed.
A social network containing 16.2 million users from Tencent Weibo.
Distribution of repost number for all microblogs in our dataset.
Average number of post per day: 58 million
Average number of post per day: 117 million
Cui, Peng, et al. "Cascading outbreak prediction in networks: a data-driven approach." Proceedings of the 19th ACM SIGKDD international conference on Knowledge discovery and data mining. ACM, 2013.
Trending Microblog Prediction
Trending:
Large number of reposts.
Why
do people perform repost action?
What is our
target
?
Predict whether a microblog will become trending in the near future.
Target
In addition, predict which users may participate in the diffusion process.
Prevent the widely spread of malicious content.
Increase user's reputation by participating early in the diffusion process.
Choose better advertisement strategy.
Microblog recommendation to accelerate information propagation.
Why Using Microblog?
Social media:
manage interpersonal relationships with friends, update daily activities to keep socially active.
News media:
official accounts and some ordinary users keep publishing latest news about all aspects of the society.
Interest discovery tool
: the users explore and disseminate the content conforming to their personal interests.
Factors That Contribute to a Diffusion Action.
The content of the microblog is in accordance with the user's interest.
The microblog is posted by the user's close friend.
The content of the microblog is epidemic (e.g., breaking news).
Influence from interest
Influence from friend
Influence from microblog
External Knowledge with 0.65 million articles from Sina.com, covering 20 categories.
Predict Trending Microblogs
Predict Diffusion Participants
Component Contribution Analysis
Three types of influence are defined, namely, interest-oriented influence, social-oriented influence, and epidemic-oriented influence, which jointly affect the user diffusion action.
A diffusion-targeted influence model was proposed for quantifying different types of influences.
The labeling task is tedious and expensive.
The contents of microblogs are highly dynamic.
Cross-media classification problem.
Objective likelihood function:
Full transcript