Closed thekingofkings closed 8 years ago
The link-prediction problem for social netowrks
Measure of influence is not important, but the ranking is. E.g. Pagerank.
You are who you know: inferring user profiles in online social networks Use community detection technique, profile categorical attributes.
We know where you live: privacy characterization of foursquare behavior Use visited venue to predict home city. Majority vote.
Towards social user profiling: unified and discriminative influence model for inferring home locations Use both network and user-centric data (tweets) to predict home location. Influence model.
Estimating age privacy leakage in online social networks Estimate age in Facebook, iterative process.
Composite Social Network for Predicting Mobile Apps Installation
Predict apps installation with multiple networks. Subnetworks are treated uniformly.
ComSoc: Adaptive Transfer of User Behaviors over Composite Social Network
Combine multiple large network to predict user behavior (user - item interactions). Method: transfer learning + hierarchical bayesian network. Different users will be influenced by their neighbors in each subnetwork with personalized weights.
Erheng Zhong composite social network
Pervasive Sensing to Model Political Opinions in Face-to-Face Networks
Use multiple network connections (with observations: count) to predict political opinion. multiple flow, weighted graph
Relationship strength measure (link-prediction) with similarity of nodal features, plus different interaction strength observations.
I model the interaction as a latent variable. The observed flows on the link is always biased sample.
Explain why with the PGT paper.
What kinds of models are there?
Drawbacks of those models in my problem?
How are they different from mine?