基于多学习策略的网页信息抽取方法

A METHOD FOR WEB INFORMATION EXTRACTION BASED ON MULTI-LEARNING STRATEGIES

由于网页信息具有异构和动态的特点,致使现有的大多数网页信息抽取方法都存在适用性差的问题。为此,将传统的文本分类器和隐式马尔可夫学习策略结合起来,提出了一种基于多学习策略的网页信息抽取方法。该方法在获得网页文本记录的局部最优分类抽取结果基础上,还利用了整个网页文本结构信息对抽取结果进行进一步优化。实验结果表明,该方法不需要对新的站点进行学习,就能获得较高的信息召回率和抽取精度,具有较强的适用性。

The current information extraction methods exist in the problem of poor applicability, since the content on the internet are heterogeneous and dynamic. A method based on multi-learning strategies was proposed for Web information extraction （IE） by combining two types of algorithms based on conventional text classifier and Hidden Markov Models （HMM）. The method can refine the IE result by using the relevant structural information present in the document, based on locally optimal classification of each fragment. Experiment result show that MLS method achieves higher accuracy and recall rate of IE without learning new Websites, and has strong applicability.