Research map / Vision-language and multimodal / Visual question answering
Dynamic Neuro-Symbolic Knowledge Graph Construction for Zero-shot Commonsense Question Answering
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 |
|---|---|---|---|
| Add dynamic commonsense graph construction and multi-hop reasoning instead of scoring answers directly (ablation) | COMeT on SocialIQa · SocialIQa | accuracy: 48.7 -> 50.1; 49.0 -> 52.6 | Helped |
| Add dynamic commonsense graph construction and reasoning instead of direct answer scoring (ablation) | COMeT on StoryCS · StoryCS | F1: 37.2 -> 39.1 | Helped |
| Use COMeT-DynaGen rather than the strongest listed zero-shot language-model baseline | SocialIQa · SocialIQa test | accuracy: 42.4 -> 52.6 | Helped |
| Use COMeT-DynaGen rather than the strongest listed zero-shot language-model baseline | StoryCS · StoryCS | F1: 36.4 -> 39.1 | Helped |
| Replace argmax decoding with beam search (beam size 5) (ablation) | COMeT-DynaGen · SocialIQa dev | accuracy: 50.1 -> 49.5 | Hurt |
| Replace argmax decoding with beam search (beam size 10) (ablation) | COMeT-DynaGen · SocialIQa dev | accuracy: 50.1 -> 50.0 | No clear effect |
| Replace argmax decoding with top-5 sampling (ablation) | COMeT-DynaGen · SocialIQa dev | accuracy: 50.1 -> 49.0 | Hurt |
| Replace argmax decoding with top-10 sampling (ablation) | COMeT-DynaGen · SocialIQa dev | accuracy: 50.1 -> 49.3 | Hurt |
| Use larger or more varied generated candidate sets instead of argmax decoding (ablation) | COMeT-DynaGen · SocialIQa dev | accuracy: 50.1; 49.5; 50.0; 49.0; 49.3 | No clear effect |
| Set thresholds at the 50th score percentile rather than using CDF-label thresholds (ablation) | COMeT-DynaGen on StoryCS · StoryCS dev | F1: 39.5 -> 38.5 | Hurt |
| Tune prediction thresholds on 4 examples rather than use CDF-label thresholds (ablation) | COMeT-DynaGen on StoryCS · StoryCS dev | F1: 39.5 -> 39.4 | No clear effect |
| Tune prediction thresholds on 10 examples rather than use CDF-label thresholds (ablation) | COMeT-DynaGen on StoryCS · StoryCS dev | F1: 39.5 -> 41.0 | Helped |
Closest papers by meaning
- Knowledge-driven Data Construction for Zero-shot Evaluation in Commonsense Question Answering (AAAI 2021 · similarity 0.94)
- (Comet-) Atomic 2020: On Symbolic and Neural Commonsense Knowledge Graphs (AAAI 2021 · similarity 0.90)
- Amplifying Diversity and Quality in Commonsense Knowledge Graph Completion (Student Abstract) (AAAI 2024 · similarity 0.87)
- Debate on Graph: A Flexible and Reliable Reasoning Framework for Large Language Models (AAAI 2025 · similarity 0.87)
- Harnessing Large Language Models for Knowledge Graph Question Answering via Adaptive Multi-Aspect Retrieval-Augmentation (AAAI 2025 · similarity 0.86)
- Neuro-Symbolic Integration for Reasoning and Learning on Knowledge Graphs (AAAI 2024 · similarity 0.86)
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.