Research map / Vision-language and multimodal / Human and embodied modeling
Document Summarization with VHTM: Variational Hierarchical Topic-Aware Mechanism
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 |
|---|---|---|---|
| replace pg + cov with VHTM | pointer-generator + coverage · CNN/Daily Mail | ROUGE-1: 39.53 -> 40.57 | Helped |
| replace Google word2vec with BERT embedding (ablation) | Base seq2seq · CNN/Daily Mail ablation | ROUGE-1: 24.14 -> 26.76 | Helped |
| add topic embedding (ablation) | Base+BERT · CNN/Daily Mail ablation | ROUGE-1: 26.76 -> 26.88 | Helped |
| add topic embedding (ablation) | Base+BERT · CNN/Daily Mail ablation | ROUGE-2: 8.24 -> 9.40 | Helped |
| add paragraph-level topic attention (ablation) | Base+BERT · CNN/Daily Mail ablation | ROUGE-1: 26.76 -> 28.42 | Helped |
| combine topic embedding and topic attention (ablation) | Base+BERT · CNN/Daily Mail ablation | ROUGE-1: 26.76 -> 29.13 | Helped |
| train topic inference jointly with summarization (joint training loss) vs independent topic inference (ablation) | VHTM · CNN/Daily Mail ablation | ROUGE-1: 27.53 -> 28.42 | Helped |
| reuse topic probability between topic embedding and attention (joint topic prob) (ablation) | VHTM · CNN/Daily Mail ablation | ROUGE-1: 28.42 -> 27.16 | Hurt |
| increase topic number K from 50 to 150 (ablation) | VHTM · CNN/Daily Mail | ROUGE-1: 38.69 -> 37.25 | Hurt |
| increase paragraph number from 3 to 5 (ablation) | VHTM · CNN/Daily Mail | ROUGE-1: 38.01 -> 37.19 | Hurt |
Closest papers by meaning
- A Discrete Variational Recurrent Topic Model without the Reparametrization Trick (NeurIPS 2020 · similarity 0.87)
- Topic-VQ-VAE: Leveraging Latent Codebooks for Flexible Topic-Guided Document Generation (AAAI 2024 · similarity 0.86)
- Autoencoding Variational Inference For Topic Models (ICLR 2017 · similarity 0.86)
- On the Affinity, Rationality, and Diversity of Hierarchical Topic Modeling (AAAI 2024 · similarity 0.85)
- Distilling Structured Rationale from Large Language Models to Small Language Models for Abstractive Summarization (AAAI 2025 · similarity 0.84)
- Video Summarization with Large Language Models (CVPR 2025 · 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.