2026-10-14T12:46:43Z https://www.scidb.cn/oai

10.57760/sciencedb.j00133.00625 2026-06-29T08:53:23Z

2026-06-29 MGSSR Dataset doi:10.57760/sciencedb.j00133.00625 en The MATRES dataset is constructed based on the TimeBank and AQUAINT corpora, encompassing texts from the news domain, serving as a benchmark for event temporal relation extraction. The TCR dataset is built from news and narrative texts, covering diverse temporal expressions and event relations, focusing on temporal reasoning tasks to evaluate models' ability to comprehend complex temporal relationships. MATRES annotates four temporal relations: BEFORE, AFTER, EQUAL, and VAGUE, comprising 256 English documents with 13,577 event pairs' temporal relations. TCR annotates three temporal relations: BEFORE, AFTER, and EQUAL, including 2,646 event pair temporal relations. First, the optimal model is trained on the MATRES training set, followed by evaluations on the MATRES test set and the TCR dataset. MATRES; TCR; Temporal Relation Extraction liang jin guo PUBLIC https://creativecommons.org/licenses/by/4.0/ dataset Science Data Bank