Implicit Chinese Complex Sentence Relation Recognition Based On Deep Learning

The recognition of the relationship in complex sentences is an essential fundamental research task in natural language processing(NLP)to distinguish between different semantic relationships.Recognizing the semantic relationship in complex sentences can provide a basis for research in machine translation,text classification,automatic summarization,and other fields,improving their performance.Based on the presence or absence of explicit conjunctions,Chinese complex sentences can be divided into marked and unmarked ones.The research on the classification of complex sentence relationships is based on the interaction of words,semantic relationships,and semantic role structure information.Linguistic rules or statistical methods are used to study the relationship types of complex sentences.The former is based on the rules extracted by linguists from a large number of complex sentence corpora to establish corresponding rule libraries for relationship classification,while the latter is based on the features of words extracted from a large-scale corpus,utilizing feature analysis to construct methods.These two methods have poor adaptability to the complex structure of complex sentences,low recognition accuracy,and require a lot of manual work and time.Currently,the introduction of deep learning methods into natural language processing is showing great potential.Specifically,in the field of Chinese complex sentences application,the use of word embedding models to model the words in complex sentences is employed to explore the characteristics and semantic information of complex sentences in-depth.This paper builds on this research method and uses the latest deep learning models to construct models for complex sentence relationship recognition to improve recognition accuracy.The work in this paper includes selecting complex sentences from materials such as Changjiang Daily,People’s Daily,internet news,etc.to build a corpus,which includes both marked and unmarked complex sentences.Additionally,the paper proposes two models: the Att-SGCN model that combines attention mechanisms and graph convolutional networks(GCN)to capture semantic information and the SRLSGCN model that combines semantic role structure and graph convolutional networks to focus on important semantic roles.The experimental results show that these two models are more effective and have better performance than rule-based and statisticalbased methods,and they are also more extensible.Compared with the current mainstream deep learning models,their performance is also improved,indicating the effectiveness of this method.