关键词: 增强约束条件随机场, 条件随机场, 属性标注, Web对象, 信息抽取

Abstract: Liner conditional random fields model is difficult to handle the relationship between Web data objects and characteristic of labeling attributes. To solve this problem, an improved sequence labeling model named Boosted Constrained Conditional Random Fields （BCCRFs） is proposed. Confidence constraint is introduced to the reasoning procedure by improved Viterbi procedures in the liner conditional random fields model. And the theory of large margin is applied to the random conditional fields model to improve the labeling accuracy. The proposed model is compared with the conditional random fields and hierarchical conditional random fields. The experimental results show that the BCCRFs model is effective on Web object information extraction and improves labeling accuracy.

Key words: boosted constrained conditional random fields, conditional random fields, attribute label, Web object, information extraction