SciEntsBank¶
Product or Programme · All products and programmes
Evidence · no claims cited
Description¶
SciEntsBank is a benchmark dataset of human-annotated student science short answers and semantic-correctness labels used to develop and evaluate automated short-answer assessment systems.
Components¶
- SciEntsBank Test_UD evaluation split for confidence estimation in ASAG: SciEntsBank is described as "a widely used benchmark dataset for ASAG containing nearly 11,000 student responses to 197 science questions across 15 domains". The study uses the Test_UD split with 4,562 responses (1,917 correct, 2,645 incorrect), treating correct responses as positive and grouping all remaining labels as incorrect. The split was further divided into a 10% calibration subset and a 90% held-out test subset via stratified sampling. Its diverse error types make it suitable for studying confidence estimation. (Longwei Cong et al. (2026))
Claims¶
Related Products and Programmes¶
- Rubric-aligned concept-annotated grading benchmarks (REC-CBM)
- SelfCode2.0 dataset of student and expert line-by-line code explanations
- NSMQ Riddles
- Gold and ultra-gold benchmark subsets
Key Sources¶
- Longwei Cong, Sonja Hahn, Sebastian Gombert, Leon Camus, Hendrik Drachsler, and Ulf Kroehne. (2026). Confidence Estimation in Automatic Short Answer Grading with LLMs. https://arxiv.org/abs/2605.00200
Page metadata
| Field | Value |
|---|---|
type |
product |
id |
scientsbank |
title |
SciEntsBank |
description |
SciEntsBank is a benchmark dataset of human-annotated student science short answers and semantic-correctness labels used to develop and evaluate automated short-answer assessment systems. |
status |
review |
product_kind |
dataset |
generated |
by process:settle-candidates on 2026-10-10 |
trust |
unverified — no one has confirmed this page's content |
Frontmatter is the page's machine-readable record — see the Schema & Guide.