Research map / Architectures and efficiency / Attention and transformers
Recurrent Attentional Topic Model
Shuangyin Li, Yu Zhang, Rong Pan, Mingzhi Mao, Yang Yang
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 |
|---|---|---|---|
| add document-level topic distribution G0 (RATM-D vs RATM-N) (ablation) | RATM · Wikipedia T=100 | perplexity: 402.47 -> 392.05 | Improved |
| add document-level topic distribution G0 (RATM-D vs RATM-N) (ablation) | RATM · NYTimes T=50 | perplexity: 576.06 -> 529.08 | Improved |
| RATM-D instead of LDA (add recurrent attention over preceding sentences) | topic model · Wikipedia T=200 | perplexity: 402.98 -> 314.47 | Improved |
| RATM-D instead of LDA | topic model · NYTimes T=100 | perplexity: 768.24 -> 442.65 | Improved |
| RATM-N instead of LDA (ablation) | topic model · Wikipedia T=50 | perplexity: 585.08 -> 553.87 | Improved |
| RATM-D instead of CTM | topic model · Wikipedia T=50 | perplexity: 524.11 -> 532.72 | Hurt |
| increase attention window C from 1 (LDA) to 4 (ablation) | RATM-D · Wikipedia T=100 | perplexity: 424.56 -> 392.05 (C=2 -> C=4) | Improved |
| increase window size C beyond 4 (5, 6) (ablation) | RATM-D · Wikipedia T=100 | perplexity: 392.05 -> 430.15 (C=4 -> C=6) | Hurt |
| increase number of topics from 50 to 200 (ablation) | RATM-D · Wikipedia | perplexity: 532.72 -> 314.47 | Improved |
| RATM-D document embeddings instead of LDA/CTM/RSM/Para2V | SVM classifier · Wikipedia and NYTimes classification | precision | Improved |
Closest papers by meaning
- Knowledge-Aware Bayesian Deep Topic Model (NeurIPS 2022 · similarity 0.88)
- textTOvec: DEEP CONTEXTUALIZED NEURAL AUTOREGRESSIVE TOPIC MODELS OF LANGUAGE WITH DISTRIBUTED COMPOSITIONAL PRIOR (ICLR 2019 · similarity 0.87)
- TopicRNN: A Recurrent Neural Network with Long-Range Semantic Dependency (ICLR 2017 · similarity 0.86)
- Neural Dynamic Focused Topic Model (AAAI 2023 · similarity 0.86)
- A Discrete Variational Recurrent Topic Model without the Reparametrization Trick (NeurIPS 2020 · similarity 0.86)
- FASTopic: Pretrained Transformer is a Fast, Adaptive, Stable, and Transferable Topic Model (NeurIPS 2024 · 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.
