Classification of URLs Citing Research Artifacts in Scholarly Documents based on Distributed Representations
Classification of URLs Citing Research Artifacts in Scholarly Documents based on Distributed Representations
复制标题
DOI：
--
复制
发表时间：
2021
期刊：
影响因子：
--
通讯作者：
Masaya Tsunokake;S. Matsubara
中科院分区：
文献类型：
--
作者：
Masaya Tsunokake;S. Matsubara
本文描述了对学术论文中涉及研究文献的URL进行分类的方法，并检验了它们的分类性能。这些方法区分URL是否涉及研究成果，并将识别出的URL分类为“工具”或“数据”。该方法使用从URL的引用上下文获得的分布式表示。URL的每个组成部分可以被视为一个词，并且整个URL的含义可以通过使用组合函数合成每个组成部分的分布式表示来生成。本文从分类性能的角度对几种类型的合成函数进行了评价。在国际会议论文中使用URL的实验表明了我们所提出的组合函数的有效性。
This paper describes methods for classifying URLs referring to research artifacts in scholarly papers, and examines their classification performance. The methods discriminate whether a URL refers to a research artifact or not and classify the identified URL into “tool” or “data.” The methods use distributed representations obtained from citation contexts of the URL. Each component of a URL can be regarded as a word, and the meaning of the entire URL can be generated by synthesizing the distributed representation of each component using compositional functions. This paper evaluates several types of compositional functions from the view-point of classification performance. Experiments with using URLs in international conference papers showed the effectiveness of our proposed compositional functions.