KoTSQA-v.2.0: Korean Time-Sensitive Question Answering Dataset

Korean Time-Sensitive Q&A Dataset (KoTSQA) is a split version for machine learning, containing 6,750 test examples and 750 train examples intended for reinforcement learning. The dataset was created by ETRI LIRS and last updated on 2026-05-29.

Use Cases

• Fine-tuning language models for Korean question answering based on time-sensitive queries.
• Evaluating model performance on time-sensitive fact retrieval in Korean.
• Training reinforcement learning agents for question answering using the provided train split.
• Analyzing model robustness on questions where answers change over time.

Strengths

• Contains 6,750 test examples, providing a substantial evaluation set.
• Includes a dedicated train split of 750 examples specifically noted for reinforcement learning use.
• Provides detailed statistics on answer categories (unchanged, changed, new) and false-premise questions.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the full dataset is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

ETRI LIRS

Freshness

Last updated 2026-05-29 00:52:02; freshness should be verified.

Geography

South Korea (implied by Korean language focus)