科学文献中参考文献影响力评估方法研究

Reference Impact Evaluation of Scientific Literatures

从科技文献的若干参考文献中寻找到一篇对其最具影响力的参考文献对研究人员获取科技文献的研究目的和研究思路是十分有益的.基于文献引用关系分析科学文献与其参考文献之间的相似性和统计特征,对参考文献进行评估;使用具有人工标记训练集的分类器,通过考虑文献的相关特征从参考文献中区分出非常重要的参考文献;使用ArnetMiner和Libra下载的会议文献和期刊文献及作者信息进行了测试实验,测试结果表明,本文对参考文献影响力的评估方法效果良好.

Generally, there are many references in scientific literatures. Researchers may desire to find the research motivation and relevant contents of the literature so as to better understand it. It will be helpful if we can figure out the most important references that mainly affect the purposes and ideas of the literature. In order to find out the most influential references of a scientific literature, we analyze the similarity and statistics features between the citing literature and its references based on the citation behavior,which can be used to help evaluating the references. Then, we use a classifier with training set labeled manually to differentiate the very-important references from important ones by considering the relevant features of the literature. We test our approach on academic collections containing literatures, authors and conferences downloaded from ArnetMiner and Libra separately. The experimental result shows that this method can get a better result.