文法推断与HMM相结合的信息提取

Combining Grammatical Inference with HMM for Information Extraction

本文提出了一种结合文法推断和HMM进行信息提取的方法。首先将待提取的原始文本转换为相应有意义的一个小的抽象符号集合,然后通过使用文法推断(GI)获取一个合适的HMM拓扑结构,最后利用所得的HMM拓扑结构,使用经典的Viterbi算法提取出用户感兴趣的信息。实验结果表明,针对半结构化文档,该方法在某些领域能够有效地提高提取的精确度。

This paper describes a method of information extraction which combines grammatical inference with HMM.Firstly, the raw text is translated into a small set of abstract symbols, and then by using grammatical inference, an optimal topology of HMM is obtained. Now we can extract the interesting information to users by using the classic Viterbi algorithm throughout the obtained topology of HMM. Results show that this method can effectively improve the precision of information extraction in some fields for semi-structured documents.