页面图文模型与元素特征归纳

Picture-text webpage model and page element feature induction

针对以图文内容为核心的页面信息抽取,以形式化的方式提出了对页面进行元素分析的理论模型。通过定义基础元素集与变换规则,页面图文模型简化了页面DOM树结构,并展现出页面内元素的图文特征。在此基础上,通过定义元素分类相似度,从页面图文模型的元素特征中进行优选,归纳最佳分类特征,提出并实现了获取最佳分类特征集与识别阈值的算法。实验结果表明,页面图文模型简化了页面元素的规模,特征集归纳算法能够在较小的学习成本下获得理想的分类精度。

According to the graphic-text content as the core of the page information extraction, this paper in a formal way forward on the page for elemental analysis of theoretical model. Through the definition of basic elements and rules of transformation, graphic-text page model with tree structure to show the page elements within the text and graphic features. The graphic-text page model elements in many features, by defining the elements classification of similarity, is proposed in this paper to obtain the best classification feature set and the recognition threshold method and gives the algorithm implementation. The experimental results show that, the graphic-text page model simplifies the page element size, feature set in smaller learning costs induction can achieve ideal classification accuracy.