基于篇章结构和引用动机的引文推荐研究

Research on Citation Recommendation Based on Text Structure and Citation Motivation

[目的/意义]在考虑科研人员引用动机差异性的基础上,结合深度学习中的层次注意力网络模型,提出一个基于文本篇章结构和引用动机的引文推荐模型。[方法/过程]通过构建科研人员的引用动机模型,将其划分为科学性引用动机和战略性引用动机。首先,将论文的篇章结构与科学性引用动机进行映射,采用层次注意力网络模型对论文章节内容进行科学性动机分类;再者,分别通过对章节内容和施引文献篇关摘(篇名、关键词和摘要)与参考文献篇关摘的相似度进行加权计算,得到基于科学性引用动机的推荐结果;在此基础上,引入战略性引用动机对推荐结果综合排序,得到最终的推荐结果。最后,以情报学领域的1443篇文献为例来对本文所提出的推荐方法进行验证。[结果/结论]实验结果证明该推荐方法能较好地实现基于科研人员不同引用动机的引文推荐,具有一定的可行性和准确性,并为后续相关研究提供借鉴思路。

[Purpose/Significance]Considering the differences of scholars'citation motivations,a citation recommendation model based on text structure and citation motivation is proposed based on the hierarchical attention network model in deep learning.[Method/Process]By constructing the citation motivation model of researchers,it can be divided into scientific citation motivation and strategic citation motivation.Firstly,the text structure of the paper was mapped to the scientific citation motivation,and the hierarchical attention network model was used to classify chapter contents according to the scientific citation motivation.In addition,the similarity weighted calculation was carried out by the similarity of the paragraphs and the titles,keywords and abstracts of the citing literatures and references,and the recommendation results based on scientific citation motivation were obtained.On this basis,strategic citation motivation was introduced to comprehensively rank the recommendation results to get the final recommendation results.Finally,1443 references in the field of information science were taken as examples to verify the proposed method.[Result/Conclusion]Experimental results show that this recommendation method achieves citation recommendation based on different motivation of researchers,has certain feasibility and accuracy,and provides reference ideas for subsequent related research.