TREC 06:

1
Modeling Influence Opinions and Structure in
Social Media
Akshay Java
Advisor Tim Finin
Modeling Influence
Thesis Statement
An accurate model of influence on the Blogosphere
must analyze and combine many contributing
factors, including topic, social structure,
opinions, biases and time. We will develop,
implement and experimentally evaluate such a
model to demonstrate its improved accuracy over
models based on any of these factors.

• Epidemic Based Influence Models
• Linear Threshold Model
• S bwv ?v
• w is the active neighbor of v,
• ?v intrinsic threshold for a node
• Greedy Heuristic
• Assign random ?v
• Compute approx influenced set
• At each step, add the node that increases the
  marginal gain in the size of the influenced set
• Limitations
• Selected nodes may belong to different topics
• Social structure not considered
• Static View of the network
• Extended Model
• Finds influential nodes for a topic
• Models opinions, bias and trust

Influence is Topical
Opinions and Bias Influence Readers
TREC 06 Finding opinionated posts, either
positive or negative, about a query 2006 TREC
Blog corpus 80K blogs, 300K post 50 test
queries BlogVox opinion extraction
system Document and sentence level
scorers Combined scores using an SVM
meta-learner Data cleaning splogs and post
identification
Popular Topics in Feeds That Matter
Tech
Slashdot Gizmodo Wired
Bias towards MSM sources
Politics
Dems
Reps
Topic Ontology derived from 83K user feed
subscriptions consisting of 500K feeds. Provides
a readership-based metric for influence.
Dailykos Talkingpoints
Michellemalkin RightwingNews
Ongoing Research
Temporal Trends Indicate Influence
A generalized framework for influence in Social
Media Predictive Models for topical trends and
influence Link Polarity and Trust Improved
sentiment analysis Generative Models for the
Blogosphere
Who started talking about the topic first? Who
were the early adopters? Who were the
influencers? Who was the source of the
information? What are the future trends to watch
out for?