Document level event argument recognition based on BERT and graph attention network
Event argument recognition is one of the sub-tasks of event extraction.Its purpose is to identify event-related argu-ments and corresponding argument roles in text.The research shows that sentence dependency is helpful to the task of event argu-ment recognition.However,it is easy to introduce irrelevant arguments in the construction of discourse dependency relations to gen-erate noise,which is poorly handled by existing methods.To solve this problem,an article level event argument recognition model based on BERT and graph attention network is proposed.This model solves the noise problem from two perspectives.On the one hand,it constructs more effective discourse dependency syntactic features by obtaining sufficient semantic features of the text.On the other hand,the graph attention network is used to assign different weights to different argument nodes to eliminate invalid argu-ments.The experimental results on RAMS corpus show that the proposed method can effectively solve the noise problem in the context-dependent syntactic relation,and obtain a good result of context-level event argument recognition.
NETL
NSTL
万方数据
