Training NQG models on multiple datasets with different answer types improved performance over single-dataset training¶
Claim · All claims
Evidence · 1 study · reviewr2·q2
Currency · newest study 2024 · not yet checked against its sources · not reviewed by a person
Subclaims¶
Reading the evidence codes
q is the design tier, i the impact, n the sample, and the kind with its r
the rigour judged within that kind: what each code means.
This claim: no standardized effect is recorded in observations/ for it yet, so no benchmark reading is possible.
q2 i? Murakhovs'ka et al. (2021) showed models trained on diverse datasets containing different answer types performed better than models trained on a single dataset. → Said Al Faraby 2024
Evidence¶
Said Al Faraby 2024¶
Said Al Faraby, Adiwijaya Adiwijaya, Ade Romadhony. (2024). Review on Neural Question Generation for Education Purposes. Int J Artif Intell Educ 34:1008–1045. https://doi.org/10.1007/s40593-023-00374-x
q2 · i? · review · r2
The review attributes this finding to Murakhovs'ka et al. (2021): training on various datasets "each containing different types of answers" improved performance versus single-dataset training. Related work by Yuan et al. (2022) adapted models to different answer types via lifelong learning.
"Murakhovs’ka et al. (2021) demonstrated that mod- els trained on various datasets, each containing different types of answers, yielded improved performance compared to models trained on a single dataset."
Discussion¶
Related Claims¶
- Multi-paragraph educational contexts require key content identification for useful NQG — related
- A model trained only on the 300-conversation BEA dataset plateaus at 74.84% average macro F1 due to limited data size and diversity — related
- Multi-task learning on a large multi-topic dataset outperforms separate topic-specific essay scoring models — a narrower finding that bears on this claim
Page metadata
| Field | Value |
|---|---|
type |
claim |
id |
multi-dataset-training-improves-nqg-answer-types |
title |
Training NQG models on multiple datasets with different answer types improved performance over single-dataset training |
description |
Training NQG models on multiple datasets with different answer types improved performance over single-dataset training |
status |
review |
evidence_strength |
weak |
newest_evidence |
2024 |
generated |
by process:wiki-ingest on 2026-10-27 |
trust |
unverified — no one has confirmed this page's content |
sources |
1 — said-al-faraby-2024 |
Frontmatter is the page's machine-readable record — see the Schema & Guide.