Topic Modeling Approach for Rank Aggregation
summary
Claims equivalence between pairwise rank aggregation and topic modeling and uses a Mallows mixture model to generate noisy permutations, proving that for sufficiently large item count q approximate separability appears. Algorithm finds dominant comparison rows as extreme points via random projection extreme-point recovery, giving polynomial sample and runtime guarantees. Implementation on MovieLens matches Bayesian matrix factorization accuracy without MCMC; theory quantifies probability of separability and required sample scaling with users, comparisons per user, and latent rank count.
AI-written summary · about
chapters (8)
- 0:00Introduction to rank models
- 3:30Topic modeling overview
- 6:12Connecting topics and ranking
- 9:23The Mallows model application
- 12:36Algorithmic advantages
- 15:24Concept of separability
- 19:02Theoretical performance results
- 23:34Empirical evaluation and Q&A
related talks
49mTopic Modeling: A Provable Spectral Method
56mTensor Decompositions for Learning Latent Variable Models I
29mThinking of Topic Modeling as Search
13mThinking of Topic Modeling as Search
27mUnveiling Clustering in BERTopic Topic Modeling
45mEvaluating Topic Models
10mThrough the Lens of LLMs: Unveiling Differential Privacy Challenges
42mRobust Mixture Learning when Outliers Overwhelm Small Groups