语义引文的实践进展研究——以Semantic Scholar和Scite为例

Research on the Practical Progress of Semantic Citation:Taking Semantic Scholar and Scite as Examples

随着计算机技术的进步,语义引文逐渐从依赖手工分析的理论走向自动化处理的实践,并显现出改变学术发现与评价方式的可能。Semantic Scholar和Scite两个平台在生产环境中建立了相关实践,并应用于学术发现增强、评价指标拓展等场景。它们在语义引文建立过程中呈现出通用的技术步骤,但是在核心环节分别采用了监督分类方法、神经多任务分类框架、深度学习方法等不同路线。推动语义引文的发展和应用,需要关注这些实践和相关研究进展在语义引文方法的选择、类别数量的控制、交叉研究的加强、规模化数据样本集的建设、学术开放水平的提升、现阶段语义引文的应用推广等方面的启示。

With the advancement of computer technology,semantic citations have gradually shifted from the theory relying on manual analysis to the practice of automated processing,which shows the possibility of changing the way of academic discovery and evaluation.Two platforms,Semantic Scholar and Scite,have established relevant practices in production environments and have been applied to scenarios such as enhancement of academic discovery and expansion of evaluation indicators.They present common technical steps in the process of establishing semantic citations,but adopt different routes such as supervised classification method,neural multi-task classification framework and deep learning method in the core step.In order to promote the development and application of semantic citations,it is necessary to pay attention to the inspirations obtained from these practices and related research progress at the present stage in peacts of the selection of semantic citation method,control of category quantity,strengthening of cross-over research,construction of large-scale sample datasets,improvement of academic openness and application,and promotion of semantic citation,etc.