Multi-party Dialogue Entity Relation Extraction Incorporating Speaker Position Heterogeneous Graph
Dialogue Relation Extraction aims to predict the relationship between entity pairs in dialogue texts.In multi-party dialogue scenarios,traditional methods of entity relation extraction face accuracy challenges due to the uncertainty of role information.To address the issue of unclear role information,the model constructs a heterogeneous graph of dialogue texts to capture the complex structure of multi-party dialogue.Moreover,in order to handle the speaker information effectively in different rounds of dialogue,the Long Short-Term Memory network is introduced for feature processing and these features are embedded into the corresponding sentence nodes.Experiments on the DialogRE dataset demonstrate the outstanding performance of model,specifically reflected in achieving F1 and F1c scores of 66.8%and 62.2%respectively,thus confirming its superiority.
国家科技期刊平台
NSTL
万方数据
