Research on the Entity Relation Extraction of Field based on Semi-Supervised

Aim at the problem of supervised learning needing much labeled data for training, this paper proposes a new method based on Semi-Supervised learning. Firstly, to construct a classifier of certain accuracy, small-scale training data was used. Secondly, with the self-expanding idea, we applied the method of information entropy to select some new instances of higher credibility from candidate instances, which were to be predicted by the classifier. Finally, with the expansion of training data, training classifier re-iteratively, classification performance tended to be stable iteration termination, which achieved the entity relation extraction of tourism field by semi-supervised learning. The experiments result show that the new classifier which applies information entropy to iteratively expand training data to be trained makes the precision rate increase by 7% and the F-score increase by 15%.