Fishery standard named entity recognition with integrated attention mechanism and BiLSTM+CRF
CHENG Ming1,2,3, YU Hong1,2,3*, FENG Yanhong1,3, REN Yuan1,2,3, FU Bo1,2,3, LIU Jusheng1,2,3, YANG He1,2,3
1.College of Information Engineering, Dalian Ocean University, Dalian; 2.Key Laboratory of Environment Controlled Aquaculture, Ministry of Education, Dalian; 3.Key Laboratory of Marine Information Technology of Liaoning Province, Dalian
Abstract: A named entity recognition model of BiLSTM+CRF(BiLSTM+Attention+CRF)is proposed based on E-BIO annotation and attention mechanism to deal with the problems of context sensitivity and semantic dilution of long sequences in fishery standard texts.The E-BIO tagging method can effectively learn the context structure features by introducing the structured information in the fishery standard text, and the attention mechanism can output the changing semantic vector to effectively solve the problem of long sequence semantic dilution.A comparative experiment was carried out on the corpus annotated with E-BIO method in order to verify the effectiveness of the proposed method, with accuracy of more than 90% for BiLSTM+Attention+CRF in the recognition of different types of fishery standard named entities, and the recall rate of over 85%.The findings indicate that the proposed method can effectively utilize the features of context structure, avoiding the problem of semantic dilution, and that has better recognition performance for fishery standard named entity recognition.
2020, Vol. 35 
