HUKB at ChEMU 2022 Task 1: Expression-Level Information Extraction
HUKB at ChEMU 2022 Task 1: Expression-Level Information Extraction
复制标题
DOI：
--
复制
发表时间：
2022
期刊：
影响因子：
--
通讯作者：
Kojiro Machi;Masaharu Yoshioka
中科院分区：
文献类型：
--
作者：
Kojiro Machi;Masaharu Yoshioka
本文描述了我们在ChEMU 2022上的三个任务的结果：任务1a（命名实体识别），任务1b（事件提取）和任务1c（回指解析）。我们采用了一种混合的方法，使用深度学习模型和一套小的后处理规则来完成这些任务。对于任务1b和1c，我们采用管道方法进行关系提取，将提及检测与关系分类相结合。此外，我们提出了Task 1c的后处理方法，该方法考虑了Task 1a的结果。我们的系统得到Task 1a的精确匹配f分数为0.9412，松弛匹配f分数为0.9572，Task 1b的精确匹配f分数为0.8865，松弛匹配f分数为0.9027，Task 1c的精确匹配f分数为0.7232，松弛匹配f分数为0.8053。尽管我们的方法试图考虑文档级上下文和任务之间的关系，但局限性仍然存在。
This paper describes our results for the three tasks at ChEMU 2022: Task 1a (named entity recognition), Task 1b (event extraction), and Task 1c (anaphora resolution). We adopted a hybrid approach using deep learning models and a small set of post-processing rules for these tasks. For Tasks 1b and 1c, we adopted a pipeline approach for relation extraction, which combined mention detection with relation classification. In addition, we proposed post-processing methods for Task 1c that considered the results of Task 1a. Our system obtained an exact match F-score of 0.9412 and a relaxed match F-score of 0.9572 for Task 1a, an exact match F-score of 0.8865 and a relaxed match F-score of 0.9027 for Task 1b, and an exact match F-score of 0.7232 and an F-score of 0.8053 for Task 1c for each test set (private score). Although our approaches tried to consider the document-level context and relationships between the tasks, limitations remained.