Research map / Probabilistic and scientific modeling / Variational and generative sampling
Conic Scan-and-Cover algorithms for nonparametric topic modeling
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use CoSAC to estimate topics without being given the topic count | LDA topic modeling · Simulated corpora, K=15 | minimum matching Euclidean distance | Mixed |
| Use CoSAC with short documents | LDA topic modeling · Simulated corpora, N_m=25 | estimated topic count: underestimating on average by 4 topics | Hurt |
| Increase document length to at least 75 | CoSAC · Simulated corpora, M=30000 | minimum matching Euclidean distance | Helped |
| Use CoSAC rather than Bayes factor model selection | Topic-count estimation · Simulations, K in [5,50] | topic-count recovery | Helped |
| Increase the angular threshold from 0.6 to 0.75 (ablation) | Document CoSAC · Simulations, vocabulary size V above 10000 | topic-count recovery: 0.6 -> 0.75 | Helped |
| Use angular threshold 0.75 (ablation) | Document CoSAC · Simulations, varying vocabulary size | minimum matching distance | Helped |
| Increase the Dirichlet parameter alpha (ablation) | CoSAC · Simulations, alpha in [0.01,1.5] | topic-estimation quality | Hurt |
| Add weighted spherical k-means post-processing to CoSAC with mean-shifting (ablation) | Document CoSAC · Simulated corpora with short documents | minimum matching Euclidean distance | Helped |
| Use weighted spherical k-means by itself instead of CoSAC initialization (ablation) | Topic direction estimation · Simulated corpora | minimum matching Euclidean distance | Hurt |
| Estimate topics using cscRecoverKL rather than RecoverKL | Anchor-word topic modeling · Simulated experiments | minimum matching Euclidean distance | Helped |
| Estimate the topic count using cscRecoverKL, which counts anchor words | Topic-count estimation · Simulations at higher K | topic-count recovery | Hurt |
| Use CoSAC rather than cscRecoverKL | Topic modeling · NYTimes | Perplexity: 2603 -> 1568 | Helped |
Closest papers by meaning
- Scalable inference of topic evolution via models for latent geometric structures (NeurIPS 2019 · similarity 0.89)
- Geometric Dirichlet Means Algorithm for topic inference (NeurIPS 2016 · similarity 0.87)
- Provable Algorithms for Inference in Topic Models (ICML 2016 · similarity 0.85)
- Algorithms for Generalized Topic Modeling (AAAI 2018 · similarity 0.85)
- Discovering Discrete Latent Topics with Neural Variational Inference (ICML 2017 · similarity 0.84)
- A Reduction for Efficient LDA Topic Reconstruction (NeurIPS 2018 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.