Research map / Image generation / Autoregressive and latent generation
Fast and Constrained Absent Keyphrase Generation by Prompt-Based Learning
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 |
|---|---|---|---|
| Use prompt-based constrained mask-predict generation instead of unconstrained autoregressive generation | Absent keyphrase generation · KP20K test set | time per document (ms): 915.79 -> 105.75; 8.67× | Helped |
| Add constrained generative absent keyphrases rather than unconstrained ones to the retrieval index | BM25 and BM25+RM3 retrieval · ACM-CR | Recall@10: 36.17 -> 36.59; 33.97 -> 36.02 | Helped |
| Use ground-truth keywords instead of automatically extracted keywords (ablation) | Our approach · Inspec | F1@M (absent keyphrases): 0.022 -> 0.176 | Helped |
| Use ground-truth keywords instead of automatically extracted keywords (ablation) | Our approach · NUS | F1@M (absent keyphrases): 0.042 -> 0.086 | Helped |
| Use ground-truth keywords instead of automatically extracted keywords (ablation) | Our approach · SemEval | F1@M (absent keyphrases): 0.032 -> 0.062 | Helped |
| Use ground-truth keywords instead of automatically extracted keywords (ablation) | Our approach · KP20K | F1@M (absent keyphrases): 0.042 -> 0.156 | Helped |
| Add task-specific keyword prompts rather than using vanilla mask-predict without prompts (ablation) | Absent keyphrase generation · KP20K, Inspec, NUS, and SemEval | F1@5 | Helped |
| Increase the number of top keywords used to construct K-Prompt (ablation) | Our approach · Validation data | F1@5 (absent keyphrases) | Helped |
| Use more than four keywords when constructing K-Prompt (ablation) | Our approach · Validation data | F1@M (absent keyphrases) | Hurt |
Closest papers by meaning
- Reflection-Window Decoding: Text Generation with Selective Refinement (ICML 2025 · similarity 0.83)
- Learning to Rank in Generative Retrieval (AAAI 2024 · similarity 0.82)
- The Hyperfitting Phenomenon: Sharpening and Stabilizing LLMs for Open-Ended Text Generation (ICLR 2025 · similarity 0.82)
- Correlation and Navigation in the Vocabulary Key Representation Space of Language Models (ICLR 2025 · similarity 0.82)
- Prompt Sketching for Large Language Models (ICML 2024 · similarity 0.81)
- Superposition Prompting: Improving and Accelerating Retrieval-Augmented Generation (ICML 2024 · similarity 0.81)
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.