Template detection and content extraction are two of the main areas of information retrieval applied to the Web. They perform different analyses over the structure and content of webpages to extract some part of the document. However, their objective is different. While template detection identifies the template of a webpage (usually comparing with other webpages of the same website), content extraction identifies the main content of the webpage discarding the other part. Therefore, they are somehow complementary, because the main content is not part of the template. It has been measured that templates represent between 40% and 50% of data on the Web. Therefore, identifying templates is essential for indexing tasks because templates usually contain irrelevant information such as advertisements, menus and banners. Processing and storing this information is likely to lead to a waste of resources (storage space, bandwidth, etc.). Similarly, identifying the main content is essential for many information retrieval tasks. In this paper, we present a benchmark suite to test different approaches for template detection and content extraction. The suite is public, and it contains real heterogeneous webpages that have been labelled so that different techniques can be suitable (and automatically) compared.
DEGAP: Dual Event-Guided Adaptive Prefixes for Templated-Based Event Argument Extraction with Slot Querying
School of Information Management, Jiangxi University of Finance and Economics(江西财经大学信息管理学院);University of Montreal (UdeM), Department of Computer Science and Operations Research (DIRO)(蒙特利尔大学(UdeM)计算机科学与运筹学系);School of Computer Science and Engineering, University of Electronic Science and Technology of China(电子科技大学计算机科学与工程学院);Department of Statistics, Division of the Physical Sciences, the University of Chicago(芝加哥大学统计学系)
Joint Bootstrapping Machines for High Confidence Relation Extraction
Siemens AG(西门子股份公司);University of Munich (LMU)(慕尼黑大学(LMU))
用于从文本中提取关系的半监督自举技术迭代地扩展一组初始种子实例。由于缺乏标记数据,自举中的一个关键挑战是语义漂移:如果在迭代期间添加了假阳性实例,则所有后续迭代都将受到污染。我们介绍BREX,一个新的自举方法,防止这种污染的高度有效的信心评估。这是通过联合使用实体和模板种子(而不是在以前的工作中只有一个),通过并行扩展实体和模板,并在每次迭代中相互约束的方式,并通过引入更高质量的模板相似性措施。实验结果表明,BREX在四个关系中实现了比现有技术好0.13(0.87 vs. 0.74)的F1。
Comparison of Feature Learning Methods for Metadata Extraction from PDF Scholarly Documents
Fraunhofer Institute for Applied Information Technology FIT(弗劳恩霍夫应用信息技术研究所FIT);University Hospital of Cologne(科隆大学医院);University of Koblenz(科布伦茨大学);Soochow University(苏州大学)