Open AccessBiLSTM-CRF Chinese Named Entity Recognition Model with Attention Mechanism
Author(s) -
Zhaolin Wan,
Jie Xie,
Wei Zhang,
Huang Zhao-hua
Publication year - 2019
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1302/3/032056
Subject(s) - computer science , artificial intelligence , conditional random field , word (group theory) , relevance (law) , dependency (uml) , feature (linguistics) , constraint (computer aided design) , named entity recognition , mechanism (biology) , pattern recognition (psychology) , natural language processing , deep learning , feature vector , sequence (biology) , task (project management) , mathematics , engineering , linguistics , philosophy , geometry , systems engineering , epistemology , political science , law , biology , genetics
In order to make up for the weakness of insufficient considering dependency of the input char sequence in the deep learning method of Chinese named entity recognition task, this paper proposes a method, which integrate Bidirectional Long Short-Term Memory (BiLSTM), attention mechanism and add the information of word vector. Firstly, the proposed model obtains the char vector feature extracted from the text corpus, which is then input to the BiLSTM model; Secondly, the attention mechanism is used to calculate the relevance between the current input char and the other input char of the BiLSTM model; Finally, the global feature is obtained according to relevance, concatenating the word vector feature, which is introduced to the Conditional random field(CRF) layer to perform the mutual constraint between tags. Thus, the classified result can be obtained. Based on the corpus of the Chinese Peoples’ Daily Newspaper in 1998, our experiments show that the proposed method can improved the performance and efficiency of named entity recognition, compared to the existing deep-learning method that combines word vector and char vector.