摘要
基于模板的上下位关系抽取是上下位关系自动获取的重要方法。目前学术界通常采用两类模板,但目前中文上下位关系抽取研究尚缺乏对这两类模板的客观比较。针对这一问题,提出基于大规模语料的匹配抽取和高质量人工标注,以实现对两类模板的优劣进行客观对比和评价,为上下位关系自动获取提供坚实基础。对于四种主流的上下位相关句型,在3800万句中文语料上进行匹配抽取,并人工标注2800个随机抽取的上下位候选词对。实验结果表明,依存句法模板的准确度比词法模板高出9.07百分点,而召回度则略低了0.32百分点。
Pattern-based hypernym-hyponym relation extraction is an important method to automatically obtain hypernym-hyponym relation.Nowadays,there are two kinds of patterns,but there is still a lack of performance comparison between the two patterns.To solve this problem,this paper proposes the matching extraction and high-quality manual annotation of large-scale corpus to objectively compare and evaluate the two patterns,which provides a solid foundation for automatic acquisition of hypernym-hyponym relation.Aimed at four mainstream sentence structures that imply hypernym-hyponym relation,matching extraction was carried out on 38 million Chinese sentences,and 2800 randomly selected hypernym-hyponym candidates were manually labeled.The experimental results show that the accuracy of the dependency pattern is 9.07 percentage points higher than that of the lexical pattern,while the recall is slightly lower by 0.32 percentage points.
作者
章岳
黄赛豪
陆凯华
李正华
Zhang Yue;Huang Saihao;Lu Kaihua;Li Zhenghua(School of Computer Science and Technology,Soochow University,Suzhou 215006,Jiangsu,China)
出处
《计算机应用与软件》
北大核心
2022年第1期126-134,共9页
Computer Applications and Software
基金
国家自然科学基金项目(61876116)
江苏省大学生创新训练计划项目(201910285142H)。
关键词
上下位关系
词法模板
依存句法模板
Hypernym-hyponym relation
Lexical pattern
Dependency pattern
