关键词: 约束条件随机场, 增强因子, 属性标注, Web对象, 最大间隔

Abstract: Conditional random fields model is one of the best statistical models of attribute labeling for Web objects. To overcome the problem that conditional random fields model does not take full advantage of the relationship between Web objects and attribute labels, this paper proposes a boosted constrained conditional random fields model. Motivated by the maximum margin criterion, the proposed model introduces constraints and boosting factor into the conditional random fields model to improve the accuracy of attribute labeling. Maximum likelihood estimation is used for the weights of characteristic functions and the Viterbi algorithm is applied for labeling. To get the best boosting factor, the concept of validation set is introduced in the dataset. The experimental results show that the labeling accuracy is improved effectively.

Key words: constrained conditional random fields, boosting factor, attribute labeling, Web object, maximum margin