Which is the cited source? A new perspective on article evaluation based on semantic similarity—Citation contribution attribution
提出基于语义相似度的引文贡献归属方法，将引文贡献归因于引用内容的真正来源（被引论文或其参考文献），测试表明该方法优于传统方法，对论文评价和排名有重要影响。
Abstract Citation analysis, an essential method in bibliometrics and scientometrics, has been widely applied to assess scholarly publications. Traditionally, studies attribute citation contributions directly to cited papers, overlooking indirect citations. Citation cascade research improves on this by assuming papers inherit contributions from their citation generations but does not consider the source of the citation content (CC). We argue that citation contributions should be attributed to the source of the CC, which includes not only the cited paper but also its references. This study introduces a semantic similarity‐driven approach (CCA_SS RS ) to allocate citation contributions. CCA_SS RS evaluates semantic similarity between CC in the citing paper (CC FP ) and the reference span in the cited paper (RS FP ), as well as between CC FP and CCs associated with the cited paper's references (CC FP_R ). If similarity between CC FP and CC FP_Ri exceeds or equals that between CC FP and RS FP , the i‐ th reference is credited; otherwise, the cited paper receives full credit. Tested on the CL‐SciSumm 2017 dataset, CCA_SS RS outperformed three established methods in identifying implicit cited sources, enabling references to receive varying contributions based on semantic similarity. This study highlights the significant impact of citation contribution attribution on paper evaluation and ranking.