Chinese Frame Semantic Role Labeling Based On Associated Memory

Semantic analysis is an important research content in the field of artificial intelligence,and it is also one of the core tasks of natural language processing technology.Semantic role labeling is an important research foundation of semantic analysis,and it has important significance in the fields of reading comprehension,automatic question answering,abstract generation,and information extraction.Based on the semantic knowledge base of Chinese Frame Net,this paper studies the task of frame semantic role labeling.The main work of this paper is as follows:(1)Aiming at the long-distance dependence and semantic importance in the frame semantic role labeling task,this paper proposes a Chinese frame semantic role labeling method based on the Self-Attention mechanism.The existing Chinese frame semantic role labeling model lacks consideration of long-distance dependence.This method uses the Bi-LSTM encoder to encode the context of each word in the sentence,and at the same time introduces the Self-Attention mechanism to model the semantic importance of each word in the sentence.Experimental results show that the model in this paper can effectively improve the performance of Chinese frame semantic role labeling.(2)Aiming at the problem of data sparseness in the Chinese frame semantic knowledge base,this paper uses the frame relations expand the experimental data of related frames.The experimental results show that using the labeled data of the related frames to expand the data of the target frame can improve the labeling performance of the target frame,especially to help the target frame withou labeled data in the sentence database to initially label the Chinese frame semantic roles.(3)Aiming at the unbalanced distribution of frame semantic role labels in the data set,this paper proposes a frame semantic role labeling method based on associated memory.This method uses associated memory to select sentences and related information similar to the target sentence from the labeled data set without introducing external resources to help the target sentence complete the frame semantic role labeling.Among them,because the Chinese frame semantic role labeling task has a strong dependence on syntax,and the existing frame semantic role labeling models are mostly based on sequential methods.When encoding,the information transmission of these models is linear,and they can not make full use of the structural information in syntax.Therefore,this paper uses the tree structure long and short-term memory network(TreeLSTM)to encode sentences according to the structure of the dependency syntax tree.Experimental results show that this method can improve the performance of semantic role annotation in Chinese frame to a certain extent.This paper studies the task of semantic role labeling in Chinese frames.The main contributions are:(1)Introducing a self-attention mechanism to model the semantic importance of each word in a sentence;(2)Without introducing external resources,use associated memory to deal with the problem of unbalanced distribution of frame semantic role tags;(3)Use Tree-LSTM encoder to encode sentences structurally according to dependency parsing tree.