Automatic Recognition Of Causal Complex Sentences Based On DPCNN Model And Fusion Of Sentence Features

Causal complex sentence is the most widely used and frequently used complex sentence in Chinese,which is an indispensable part in the study of Chinese complex sentence.The first step in the study of complex sentences is to understand the semantics of complex sentences,which requires the recognition of the relationship between complex sentences.For marked complex sentences,we can recognize the relationship of complex sentences according to the relative words,but the recognition of relative words is also a difficult problem.Sometimes a relative word can correspond to multiple categories,which makes the recognition of complex sentences more difficult.At present,there are many methods for complex sentence relationship recognition.Both rule-based and machine learning methods rely too much on the features selected manually,which leads to the situation of sparse features and incomplete semantics.The method based on deep learning does not need to extract features manually,it can automatically mine the hidden features of complex sentences,so as to realize the relationship recognition of complex sentences.At present,there are many methods to recognize the relationship of complex sentences,each of which has its own advantages and disadvantages.Both rule-based and machine learning methods rely too much on manually selected features,resulting in sparse and incomplete features.The method based on deep learning does not need to extract features manually,it can automatically mine the hidden features of complex sentences,so as to realize the relationship recognition of complex sentences.Aiming at the problem that complex sentence relation recognition relies too much on artificial features,this paper proposes a method based on the fusion of DPCNN model and sentence features for complex sentence relation recognition.Although deep learning model can automatically mine the implicit semantic information of sentences,adding artificially selected features can make deep learning model make full use of the knowledge and achievements of linguistic research and make the model more efficient High accuracy,achieve better recognition effect.The research object of this paper is two sentence marked causal complex sentences.The data are from CCCS corpus of central China Normal University and THUCNews corpus of Tsinghua news classification.The main task is to use deep learning model and artificial selection feature fusion method to recognize complex sentence relationship,compare the fusion effect of different features and models,and select the best model.In this paper,we select three artificial features:relational words,part of speech and dependency syntax,and index these features to one-dimensional embedding.First,we use pyltp to label the part of speech and analyze the dependency syntax of the complex sentences in the corpus to get the part of speech and dependency relationship of each word.Then we use gensim to train the complex sentences into pre trained word2vec word embedding representation,and input the word embedding and the feature embedding of relational words,part of speech and dependency relationship into the DPCNN model as a new embedding.In the experiment,relational word feature,part of speech feature and dependency syntactic feature are combined with word embedding respectively,and different feature combinations are also combined with word embedding.The best combination method is obtained,and the F1 value of the experiment is better than the existing methods.The F1 value of the model with features is also increased compared with the model without features.The experimental results show that the model with artificial features in deep learning model performs better.