传统PDF论文抽取方法主要是单独基于规则的方法或单独基于机器学习的方法,其中基于规则的抽取方法在处理格式固定的数据方面具有明显的优势,通过制定简单的抽取规则即可准确定位并抽取数据;而在处理格式灵活的数据时,则需要制定相当复杂的规则,且不具备对论文格式的适应性,因而明显缺乏机器学习抽取方法的灵活性和准确性.为此,提出了一种基于规则与SVM相结合的PDF论文抽取方法.该方法充分利用规则方法与机器学习在信息抽取时的优点,在用简单的规则抽取格式固定的信息的基础上,选取样本特征构建训练集,并选择最优的核函数生成SVM模型,从而完成基于SVM方法的信息抽取.以SVM的抽取结果为主体,通过合理利用基于规则抽取的结果并制定适当的规则的方式对该方法进行验证.实验结果表明,该方法在论文元数据和章节标题等信息抽取方面具有较好的效果.%Traditional extraction methods for PDF format papers are mainly based on either rules or machine learning. The extraction method based on rules has obvious advantages in processing fixed format data,which can accurately locate and extract data by making some simple rules of extraction. However it needs fairly complex rules to deal with flexible data and is lack of the adaptability of paper format,which cannot do better than the extraction method of machine learning in terms of flexibility and accuracy. For this,an extraction method for PDF papers via integration of rules with SVM is proposed which makes full use of the advantages of rules and machine learn-ing when extracting information. On the basis of extracting fixed format information via simple rules,the sample characteristics is chosen to build the training set and the optimal kernel function is selected to generate the SVM model for implementation of information extrac-tion based on SVM. By taken extraction results of the SVM as the main body,the verification experiments is conducted based on rules ra-tionally and some appropriate rules made. The experiment results show that it can achieve better results for extracting metadata and chapter headings of PDF papers.
展开▼
