Research map / Vision-language and multimodal / Visual question answering
Semi-Distantly Supervised Neural Model for Generating Compact Answers to Open-Domain Why Questions
Ryo Ishida, Kentaro Torisawa, Jong-Hoon Oh, Ryu Iida, Canasai Kruengkrai, Julien Kloetzer
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 question word marking to pointer-generator network (ablation) | PG · Why-QA compact answers test set | ROUGE-1: 36.5 -> 38.2 | Improved |
| add query attention to PG (ablation) | PG · Why-QA compact answers test set | ROUGE-1: 36.5 -> 36.1 | Hurt |
| add 250K auto-generated training data to manual data (ablation) | Proposed · Why-QA compact answers test set | ROUGE-1: 38.2 -> 44.9 | Improved |
| add 250K auto-generated training data to manual data (ablation) | Proposed · Why-QA compact answers test set | ROUGE-2: 21.6 -> 31.0 | Improved |
| increase auto-generated data from 250K to 500K (ablation) | Proposed+N · Why-QA compact answers test set | ROUGE-1: 44.9 -> 44.1 | No clear effect |
| increase auto-generated data from 250K to 1M (ablation) | Proposed+N · Why-QA compact answers test set | ROUGE-1: 44.9 -> 44.4 | No clear effect |
| train on 250K auto-generated data only, without manual data (ablation) | Proposed+250K · Why-QA compact answers test set | ROUGE-1: 44.9 -> 28.6 | Hurt |
| compare PG against DrQA span extraction (ablation) | PG · Why-QA compact answers test set | ROUGE-1: 36.5 -> 24.7 | Hurt |
| pointer-generator vs CRF causal-relation extraction (ablation) | Proposed · Why-QA compact answers test set | ROUGE-1: 32.7 -> 38.2 | Improved |
| use Proposed+750K instead of Proposed (ablation) | Proposed · Fluency (human) | Good label ratio (%): 61.2% -> 65.7% | Improved |
| add question word marking (ablation) | PG · Fluency (human) | Good label ratio (%): 52.7% -> 61.2% | Improved |
| use Proposed+750K instead of Proposed (ablation) | Proposed · Informativeness (human, question only) | Good label ratio (%): 17.1% -> 24.1% | Improved |
Closest papers by meaning
- Exploiting Background Knowledge in Compact Answer Generation for Why-Questions (AAAI 2019 · similarity 0.90)
- A Semi-Supervised Learning Approach to Why-Question Answering (AAAI 2016 · similarity 0.89)
- Generative Question Answering: Learning to Answer the Whole Question (ICLR 2019 · similarity 0.85)
- Conclusion-Supplement Answer Generation for Non-Factoid Questions (AAAI 2020 · similarity 0.85)
- Improving Question Generation with Sentence-Level Semantic Matching and Answer Position Inferring (AAAI 2020 · similarity 0.85)
- SuRe: Summarizing Retrievals using Answer Candidates for Open-domain QA of LLMs (ICLR 2024 · similarity 0.83)
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.
