An Improved Method of Joint Extraction

Named entity recognition (NER) and relation extraction (RE) are the basic tasks of Natural Language Processing (NLP). However, previous works always treat them as two separated subtasks, a novel improved method of joint extraction was present in this paper to solve the problem of internal relationship and error propagation in traditional pipeline model, Named entity recognition is regarded as a sequential annotation problem. In relation extraction task, the relationship between two entities and their relationship types are predicted at the same time, the possible multiple pairs of relationships in sentences are identified. Finally, the work innovatively use sequential patterns to correct the results. The experiment on two authoritative datasets verifies the advancement in our model.