Attribute Extraction by Combing Feature Ranking and Sequence Labeling

Authors:
  • BPBin Peng
    Bin Peng
  • XZXiaoming Zhang
    Xiaoming Zhang
  • YHYueying He
    Yueying He
  • ZLZhoujun Li
    Zhoujun Li

Abstract
Due to the language characteristics, it is a challenge for the knowledge extraction of Chinese text documents. In this paper, an attribute extraction method based on feature ranking and sequence labeling is proposed. Firstly, we obtain the training corpus by annotating Wikipedia texts with the attribute information extracted from the information box of Wikipedia. To improve the quality of training corpus, the trigger keywords are filtered based on the information entropy. The attribute extraction is regarded as a sequence labeling problem, which exploits the multidimensional features such as part of speech and word context. Then, the conditional random field model is trained on the corpus to extract attributes from the unstructured texts. Experiment results show that our method can effectively improve the quality of training corpus using the keyword filtering technique, and hence improve the performance of attribute extraction. Compared with the rule-based attribute extraction methods, our method can be extended to other fields, which has better portability and expansibility.