WWW2011-Unified Analysis of Streaming News Amr Ahmed, Qirong Ho, Jacob Eisenstein, Eric Xing, Carnegie Mellon University, Alexander J. Smola, Choon Hui Teo Yahoo! Research
Motivation Clustering News into Stories. – Based on Entity, time, etc.. – Cant identify high-level topics Topic Modeling – LDA, etc – Cant cluster stories well. Propose: Cluster + Topic
Cluster Model: Recurrent Chinese Restaurant Process
Topic Model: LDA
Storyline Model
Inference: Particle Filtering
Sampling Topics
Sampling Stories
Sampling Stories – cont. too expensive: instead, sample s* from with acceptance rate Update particle weight
Experiments Baseline: single-link clustering.
Application: Structured Browsing