Research map / Label-efficient and robust learning / Recognition and long-tail
Contrastive Learning for Neural Topic Model
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 contrastive NTM with word-based positive and negative sampling instead of the strongest listed baseline | Neural topic modeling · 20NG, IMDb, Wiki; T=50 and T=200 | NPMI: 0.324 ± 0.006 -> 0.334 ± 0.004 (20NG, T=50); 0.455 ± 0.007 -> 0.478 ± 0.006 (Wiki, T=200) | Helped |
| Remove positive sampling (ablation) | Contrastive NTM · 20NG, IMDb, Wiki; T=50 and T=200 | NPMI: 0.334 ± 0.004 -> 0.320 ± 0.004 (20NG, T=50); 0.497 ± 0.009 -> 0.452 ± 0.012 (Wiki, T=50) | Hurt |
| Remove negative sampling (ablation) | Contrastive NTM · 20NG, IMDb, Wiki; T=50 and T=200 | NPMI: 0.334 ± 0.004 -> 0.331 ± 0.002 (20NG, T=50); 0.497 ± 0.009 -> 0.474 ± 0.010 (Wiki, T=50) | Hurt |
| Use k=15 sampled tokens rather than k=1 or k=5 (ablation) | Contrastive NTM · 20NG, IMDb, Wiki; T=50 and T=200 | NPMI: 0.327 ± 0.006 -> 0.334 ± 0.004 (20NG, k=1 to k=15); 0.465 ± 0.012 -> 0.497 ± 0.009 (Wiki, k=5 to k=15) | Helped |
| Use adaptive triangle scheduling for beta rather than fixed/manual sampling controls | Contrastive NTM · 20NG, IMDb, Wiki; T=50 and T=200 | NPMI | Helped |
| Use word-based sampling rather than 0-sampling or random-document sampling (ablation) | Contrastive NTM · 20NG, IMDb, Wiki; T=50 and T=200 | NPMI: 0.269 ± 0.003 -> 0.334 ± 0.004 (20NG, T=50); 0.448 ± 0.008 -> 0.497 ± 0.009 (Wiki, T=50) | Helped |
| Use word-based sampling rather than topic-based sampling (ablation) | Contrastive NTM · 20NG, IMDb, Wiki; T=50 and T=200 | NPMI: 0.313 ± 0.004 -> 0.334 ± 0.004 (20NG, T=50); 0.467 ± 0.012 -> 0.497 ± 0.009 (Wiki, T=50) | Helped |
| Use tf-idf importance rather than PCA or SVD for sample construction (ablation) | Contrastive NTM · IMDb, 20NG, Wiki; T=50 | NPMI: 0.481 ± 0.005 -> 0.497 ± 0.009 (Wiki, PCA to tf-idf); 0.476 ± 0.014 -> 0.497 ± 0.009 (Wiki, SVD to tf-idf) | Helped |
Closest papers by meaning
- Topic Modeling as Multi-Objective Contrastive Optimization (ICLR 2024 · similarity 0.87)
- Meta-Complementing the Semantics of Short Texts in Neural Topic Models (NeurIPS 2022 · similarity 0.86)
- Representing Mixtures of Word Embeddings with Mixtures of Topic Embeddings (ICLR 2022 · similarity 0.85)
- Neural Dynamic Focused Topic Model (AAAI 2023 · similarity 0.85)
- Neural Topic Model via Optimal Transport (ICLR 2021 · similarity 0.85)
- Document Informed Neural Autoregressive Topic Models with Distributional Prior (AAAI 2019 · similarity 0.85)
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.