The Development of a Search Engine for Academic Papers in Web

网上的开放获取科学论文可以通过几个搜索引擎进行搜索。例如，谷歌学者对文献的覆盖率更高，尽管它不一定保证免费获取全文。我们开发并评估了学术论文全文搜索引擎“Aletheia”。该系统获取广泛主题的PDF文件，并使用基于文本和结构特征的分类器自动检测学术论文。我们已经建立了包含300万个日语PDF文件的PDF数据库集合，五种类型的Weka分类器(AdaBoost，Decision Tree(C4.5)、朴素贝叶斯、随机森林和支持向量机)分别针对20,000个测试集进行训练，使用10次交叉验证来自动检测学术论文。这些特征是使用手动构建的规则生成的，由三种类型的特征组成：结构、URL和内容。

Open access scientific papers available on the Web could be searched through several search engines. For example, Google scholar has higher coverage of literature, although it does not necessarily guarantee free access to full text. We have developed and evaluated the "Aletheia" search engine for full text academic papers. The system obtains PDF files on a broad range of topics and automatically detects academic papers using classifiers based on text and structure features. We have built PDF database collection containing 3 million Japanese PDF files, five types of Weka classifiers(AdaBoost, Decision Tree(C4. 5), Naive Bayes, Random Forest, and Support Vector Machine) were separately trained for 20, 000 test collection using 10-fold cross-validation to automatically detect academic papers. The features were generated using hand-built rules and consisted by the three types of features : structure, URL, and content.