Research On Corpus Construction And Identification Method Of Chinese Rhetorical Questions

Chinese rhetorical questions recognition refers to judging whether a sentence is a rhetorical question,which belongs to a binary classification task.The key is to explore the similarities and differences between rhetorical questions and non-rhetorical questions.In view of the current research status of Chinese rhetorical questions,this dissertation conducts research from the aspects of corpus construction,rhetorical questions recognition,and implicit rhetorical questions recognition.The main contents of this dissertation are as follows:(1)Research and Construction of Chinese Rhetorical Questions CorpusAiming at the phenomenon that there is no publicly available corpus for Chinese rhetorical questions research,this dissertation proposes a semi-automatic rhetorical questions corpus collection method.Firstly,this dissertation constructs a rhetorical questions recognizer based on the feature dictionary to filter representative rhetorical questions,and then formulates the unlabeled sample selection strategy to train the rhetorical questions active learner for further reducing the manual labeling work,and finally the corpus scale reaches more than 6000 sentences.At the same time,this dissertation also conducts rhetorical questions recognition experiments.The experimental results show that the corpus constructed has certain application value in the research of rhetorical questions recognition.(2)Chinese Implicit Rhetorical Questions Recognition based on Sentiment AnalysisTo address the difficulties encountered in the research process of rhetorical questions recognition,this dissertation proposes an Chinese implicit rhetorical questions recognition task for the first time,and designs an implicit rhetorical questions recognition model which integrates sentiment analysis.Firstly,the model employs the attention mechanism to combine part-of-speech features and sentence representations to obtain semantic information,then captures the emotional changes within sentences to mine the weights of different contextual information.Finally,sentiment analysis task is used to assist in identifying implicit rhetorical questions.The experimental results show that the performance of the model proposed on the implicit rhetorical questions dataset is improved by 1.61.(3)Chinese Implicit Rhetorical Questions Recognition based on Hierarchical Information Fusion Graph Neural NetworksIn order to prevent over-reliance on artificially designed features,this dissertation proposes an implicit rhetorical questions recognition model based on multi-level information fusion graph neural network.Firstly,for the BERT-based encoder,this dissertation designs a representation learning method on contrastive adversarial learning to improve text representation ability.The intra-sentence,inter-sentence and word-sentence subgraphs are designed hierarchically,and finally the heterogeneous graph convolution model is used to integrate information.The experimental results show that the performance of the model proposed on the implicit rhetorical questions dataset is improved by 2.31.This dissertation studies the Chinese rhetorical questions recognition task,and proposes corresponding solutions.The experimental results show that the model designed improves the performance of the task,which has a certain role in promoting the research on the Chinese rhetorical questions recognition.