Coreference Resolution Research Based On Chinese Frame Semantics

As a kind of frame semantic resource,CFN is widely used in natural language processing research fields such as reading comprehension and question answering system.Constructing the connection between frame elements in Chinese text sentences is an effective means of semantic understanding,but the mention co-reference inside the frame elements hinders the establishment of this connection.This article takes the mentions in the frame elements as the research object and uses the semantic information of the frame to study the co-reference relationship between the mentions.This paper is based on the social science fund project "Research on the Expansion of the Semantic Knowledge Base of Chinese Framework for Machine Reading",aiming at the problem of co-reference between the mentions among the different frame elements in the constructed Chinese frame semantic text corpus.Research on the co-reference resolution method based on frame semantics,and apply the co-reference resolution method to the annotation and testing of Chinese frame common-reference corpus.The paper,through the fusion of the semantic information of the Chinese frame,using traditional machine learning and the current popular deep learning and other different methods,experiments are carried out on the sentence frame corpus of Chinese frame text marked with common reference information.(1)Traditional machine learning co-reference resolution method based on frame semantics.This method analyzes the frame element labeling information and the relationship of frames,and obtains the frame feature set as the frame semantic information,combined with the co-reference basic feature set,training the co-reference classification model.(2)Deep-learning co-reference resolution method based on framesemantics.Learned the expressed contextual text information,and then use the frame information marked by the Chinese frame to extract the frame information triplet and use it as external information to integrate it into the LSTM deep learning classification model.Finally,based on the above two methods,a co-reference management system based on the Frame Net knowledge base is designed and implemented.The co-reference management system applies the method to the annotation process of the corpus,and provides tools for the management and annotation of the co-reference digestion corpus in the Chinese framework.