Research map / Probabilistic and scientific modeling / Variational and generative sampling
CompareLDA: A Topic Model for Document Comparison
Maksim Tkachenko, Hady W. Lauw
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| supervise topics with pairwise comparisons (CompareLDA) instead of pointwise BTL scores (sLDA) | topic model · Wikipedia (AC) | ranking accuracy: 55.0 -> 78.8 | Improved |
| joint training of topics and comparison instead of LDA then BTL-R pipeline (ablation) | CompareLDA · Wikipedia (AC) | ranking accuracy: 75.4 -> 78.8 | Improved |
| joint instead of pipeline (ablation) | CompareLDA · Wikipedia (CC) | ranking accuracy: 74.9 -> 75.6 | No clear effect |
| joint instead of pipeline (ablation) | CompareLDA · Wikipedia (LE) | ranking accuracy: 84.0 -> 85.7 | No clear effect |
| joint instead of pipeline (ablation) | CompareLDA · Product Reviews | ranking accuracy: 76.7 -> 81.4 | Improved |
| joint instead of pipeline (ablation) | CompareLDA · Movie Reviews | ranking accuracy: 63.8 -> 68.8 | Improved |
| add comparison supervision to LDA (ablation) | CompareLDA · Wikipedia (AC) | held-out log-likelihood: -10.084 -> -10.084 | No clear effect |
| add comparison supervision to LDA (ablation) | CompareLDA · Product Reviews | held-out log-likelihood: -10.389 -> -10.391 | No clear effect |
| pointwise supervision from ranked list (sLDA*) instead of BTL-converted comparisons (ablation) | sLDA · Wikipedia (AC) | ranking accuracy: 55.0 -> 55.6 | No clear effect |
| pointwise supervision from ranked list (sLDA*) instead of BTL-converted comparisons (ablation) | sLDA · Wikipedia (LE) | ranking accuracy: 52.2 -> 50.8 | No clear effect |
| BTL conversion instead of ranked-list supervision (ablation) | sLDA · Wikipedia (AC) | held-out log-likelihood: -12.367 -> -11.132 | Improved |
| BTL conversion instead of ranked-list supervision (ablation) | sLDA · Wikipedia (CC) | held-out log-likelihood: -12.368 -> -12.370 | No clear effect |
Closest papers by meaning
- A Statistical Approach for Optimal Topic Model Identification (JMLR 2022 · similarity 0.85)
- Learning Supervised Topic Models for Classification and Regression from Crowds (TPAMI 2017 · similarity 0.85)
- Online Bayesian Moment Matching for Topic Modeling with Unknown Number of Topics (NeurIPS 2016 · similarity 0.84)
- A Discrete Variational Recurrent Topic Model without the Reparametrization Trick (NeurIPS 2020 · similarity 0.84)
- Algorithms for Generalized Topic Modeling (AAAI 2018 · similarity 0.84)
- Autoencoding Variational Inference For Topic Models (ICLR 2017 · 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.
