.png){kind=icon}
.png){kind=icon}
Return
In-context curriculum learning for enhanced semantic understanding in scientific text mining
.png){kind=icon}
PRE
AI
Y
J
X
Y
W
Q
DOI：10.1007/s11192-026-05767-y.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Large language models (LLMs) are increasingly employed in scientometric studies for scientific text analysis and knowledge discovery. A key factor influencing their in-context learning (ICL) performance is the organization of demonstrations, which guides how LLMs interpret complex scientific texts. However, existing organization strategies often involve high computational costs and lack interpretability, limiting their effectiveness and transferability across models. We propose In-Context Curriculum Learning (ICCL), a lightweight and explainable framework that organizes demonstrations in ascending order of complexity to enhance LLMs’ deep semantic understanding of scientific text. ICCL comprises three stages: demonstration retrieval, label recall, and curriculum-guided ordering based on a novel Instruction Alignment Score (IAS) to quantify contextual difficulty. We evaluate ICCL on three representative scientific text mining benchmarks, SciCite, SciNLI, and SciERC, covering citation intent classification, scientific language inference, and entity extraction tasks. Relative to few-shot prompting, ICCL yields consistent average F1-score gains of 4.65% on SciCite, 3.29% on SciNLI, and 7.29% on SciERC, while demonstrating strong overall performance against the other baseline methods. We further explore the mechanism behind ICCL, showing that LLMs benefit when lower-difficulty demonstrations are positioned closer to the instruction, while more complex ones are placed nearer to the test input. These findings support ICCL as a practical and generalizable method to improve LLMs’ deep semantic understanding of scientific texts. The code is publicly available at https://github.com/61peng/sci-iccl .
Keywords:
In-Context Learning
Curriculum Learning
Large Language Models
Scientific Text Mining
Demonstration Organization
Journal
IF:
3.5
Papers:
8.0K
Citations:
2.2W
Organization
Cited Papers