基于证据理论的论文元数据抽取算法研究

针对PDF文件的特点，应用pdfbox开源库对PDF文件进行解析，去除PDF文件的文件头、交叉引用表以及文件尾等额外的文档描述信息得到目标信息。在研究不确定性理论的基础上，确定初始证据各种特征的可信度计算方法，通过推理网络及证据理论的推理算法，得到各个证据的可信度，最后比较各个证据可信度，对论文元数据进行抽取。实验表明，各类元数据的查全率都在87％以上，查准率都在92％以上，与常用的正则表达式方法相比准确率提高了10％以上，大幅提高了工作效率。

Research of paper metadata extraction algorithm based on theory of evidence

Aiming at the characteristics of PDF files, PDF files were parsed by the open source library of pdfbox, the body of the PDF files were get by removing the additional information which describes document,such as the header,the cross-reference table and the trailer.Extracted metadata using the algorithm of reasoning with uncertainty,which based on theory of evidence by definiting the calculation of the initial evidence and got the credibility of evidence.Test result shows that the recall reaches 87% and the precision reaches 92% in paperopen.Aceuracy increasing of more than 10% compared with the general method which uses the regular expressions, greatly enhances the efficiency.