Research map / Large language models / Retrieval and knowledge
A Deep Generative Framework for Paraphrase Generation
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 |
|---|---|---|---|
| Condition both VAE encoder and decoder on the original sentence and generate from a sampled latent code | VAE-SVG · MSCOCO | BLEU: 37.0 -> 39.2 | Helped |
| Condition both VAE encoder and decoder on the original sentence and generate from a sampled latent code | VAE-SVG · MSCOCO | TER: 51.6 -> 43.6 | Helped |
| Use VAE-SVG-eq with a shared original-sentence encoder instead of VAE-SVG (ablation) | VAE-SVG · MSCOCO | BLEU: 39.2 -> 37.3 | Hurt |
| Use VAE-SVG-eq with a shared original-sentence encoder instead of VAE-SVG (ablation) | VAE-SVG · Quora (150K training, Avg) | BLEU: 25.0 -> 26.2 | Helped |
| Add supervised paraphrase targets and condition generation on the source rather than training an unsupervised VAE (ablation) | VAE-SVG · Quora (150K training, best METEOR) | BLEU: 11.4 -> 30.3 | Helped |
| Add the paraphrase-sentence encoder on the VAE encoder side, compared with VAE-S (ablation) | VAE-SVG · Quora (150K training, best METEOR) | BLEU: 19.7 -> 30.3 | Helped |
| Increase Quora training data from 50K to 150K (ablation) | VAE-SVG · Quora (Avg) | BLEU: 13.8 -> 25.0 | Helped |
| Use beam search with beam size 10 (ablation) | VAE-SVG-eq · Quora (150K training, Avg) | BLEU: 26.2 -> 37.1 | Helped |
| Use beam search with beam size 10 (ablation) | VAE-SVG · MSCOCO (Avg) | BLEU: 39.2 -> 41.3 | Helped |
| Use beam search with beam size 10 (ablation) | VAE-SVG · MSCOCO (Avg) | TER: 43.6 -> 40.8 | Helped |
| Use beam search with beam size 10 (ablation) | VAE-SVG · MSCOCO (Avg) | METEOR: 29.2 -> 30.9 | Helped |
Closest papers by meaning
- Paraphrase Generation with Latent Bag of Words (NeurIPS 2019 · similarity 0.88)
- Unsupervised Deep Keyphrase Generation (AAAI 2022 · similarity 0.88)
- Novelty Controlled Paraphrase Generation with Retrieval Augmented Conditional Prompt Tuning (AAAI 2022 · similarity 0.87)
- Joint Copying and Restricted Generation for Paraphrase (AAAI 2017 · similarity 0.87)
- Dictionary-Guided Editing Networks for Paraphrase Generation (AAAI 2019 · similarity 0.87)
- LAMPAT: Low-Rank Adaption for Multilingual Paraphrasing Using Adversarial Training (AAAI 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.