Joint Extraction of Triple Knowledge Based on Relation Priority

摘要

Triple knowledge extraction from texts is an important type of information extraction and plays a crucial role in such domains as knowledge base construction. To do so, most of existing methods usually first extract entities or entity pairs, which may lead to the redundant combination problem of entities and cannot tackle with the overlapping problem. To solve these problems, this paper proposes a new model, called RFTE, based on relation priority. The model first classifies texts according to the classification of relations to be extracted, then combines the predicted relations with texts to perform entity recognition based on the sequence labelling technique, and finally combines the extracted head and tail entities and the corresponding relations to obtain triple knowledge. Our model decouples entity recognition under different types of relations and significantly reduces the impact of the large search space. In addition, this paper also employs data augmentation and rules to further promote the performance of the model. Experiments on three benchmark datasets show that the proposed model successfully overcomes the overlapping problem and significantly outperforms the traditional methods.