New system improves extraction of breeding information from grain studies
Retrieval-Augmented Multi-Prompt Ensemble for Minor-Grain Breeding Information Extraction
Computation and Language
Summary
Extracting detailed information from scientific papers about grain breeding is challenging because it involves many different kinds of data and relationships. The authors developed a system called RAME that uses multiple tactics to get better information by combining results from several carefully designed prompts and verifying them against each other. This approach does not require additional training and improves accuracy by fetching the most relevant examples and filtering guesses through majority voting. Their system ranked first in a competition, outperforming a strong baseline powered by GPT-5.5.
What this means in practice
- •For agricultural data teams: Automatically extract detailed breeding data from minor-grain research papers to build structured knowledge bases for crop improvement.
- •For natural language processing engineers: Create high-accuracy information extraction pipelines without additional model training by combining retrieval and multi-prompt ensembling techniques.
Authors
Hang Zhao, Jiahao Wang
Abstract
This paper presents our system for CCL2026-Eval Task 5: Minor-Grain Breeding Information Extraction (MGBIE), which jointly extracts 12 entity types and 6 relation types from minor-grain breeding literature. We propose RAME (Retrieval-Augmented Multi-Prompt Ensemble), a training-free framework that elicits multiple LLM outputs under controlled diversity and aggregates them by majority voting to obtain high-confidence predictions. RAME combines (i) retrieval-augmented few-shot selection via a hybrid BM25-embedding retriever, (ii) a three-prompt ensemble (Strict, Relaxed, Balanced) spanning the precision to recall spectrum, and (iii) large-scale repeated sampling with majority voting to filter noisy predictions. Built on DeepSeek-V4-Flash, RAME achieves a Total Score of 0.499 (NER 0.730, RE 0.346) on the leaderboard, ranking 1st and surpassing the official Track-A baseline powered by GPT-5.5 (0.448), representing an 11.4% relative improvement. Code is available at https://github.com/king-wang123/CCL26-RAME.