Research on Entity Relation Extraction in Education Field Based on Multi-feature Deep Learning
Mengxue Song, Jingsheng Zhao, Xiang Gao · 2020
Relation extraction is the core task of information extraction and natural language processing tasks. Most of the existing methods stop at extracting the context characteristics of the entity in the sentence, and sometimes ignore the global context characteristics of the entity in the whole corpus. In order to fully consider the local and global information of text, a Chinese entity relation extraction model (Cooc_Att_Bilstm) based on the combination of multi-feature attention Bidirectional Long Short-term Memory (Bilstm) network and entity co-occurrence network is proposed. First, the attention-based Bilstm is utilized to extract sentence-level context features for entity pairs and combine the lexical, syntactic, semantic and positional features. Then, construct an entity co-occurrence network to extract global corpus-level features for each entity. Finally, the combination of sentence-level context features and global corpus-level features is used to extract entity relation extraction. The experimental results show that this method has good performance in the manual labeling data set.