Research on Document Level Temporal Extraction Method Based on Event Time Argument Extraction
The task of identifying event temporal relationships is a branch of the field of relationship extraction,which has received increasing attention in recent years. At present,research on extracting temporal relationships of chapter level events has not been fully explored in existing neural network methods. Therefore,this paper proposes a method based on the temporal relationship matrix of text events,and uses the time argument theory in linguistics to guide the model extraction effect. By modifying the word embedding in the pre training stage,the tense,aspect,and time adverbs of the sentence are incorporated as additional information into the word embedding expression of event triggered words. At the same time,the model also utilizes event time argument extraction tasks for auxiliary training,thereby constructing text expressions with enhanced temporal features. By incorporating the task of extracting event time arguments into the model training process,the model achieved better performance than the baseline in extracting event temporal relationships.
relationship extractiontemporal relationshipsevent time argumentpre-trainingextraction matrix
NETL
NSTL
万方数据
