Research On Chinese Text Summarization Based On Nuclearity

Automatic summarization is one of the research hotspots in the field of natural language processing.It can help users quickly filter out useful information from massive internet information.However,most of the traditional automatic summarization research focus on the use of text semantic information,and ignore the role of structural information of the text for the summarization task.This defect is particularly prominent in Chinese automatic summarization.Compared with English texts,Chinese texts have more complex and richer chapter structure information.Chapter structure information,especially nuclearity information,plays an important role in locating the core content of the document and summarizing the core ideas of the document.Therefore,this article conducts a study of Chinese text summarization based on nuclearity.The specific research content can be divided into the following three aspects:First of all,in view of the feature that the core sentences in the nuclearity are often the summaries of extractive summarization,this paper proposes a method of extractive summarization based on the nuclearity.This method first obtains the serialization information of the nuclearity of the sentence.Then the neural network is used to enhance information and encode semantic information of the sentences and nuclearity.Finally,summaries are extracted according to the encoding result.Experimental results show that this method can improve the accuracy and stability of the summarization compared with the current mainstream extractive summarization methods.Secondly,in order to solve the problem that abstractive summarizations rarely use text structure information,this paper proposes an abstractive summarization method based on nuclearity.In this method,graph neural network is used to encode nuclearity structure information,and it is integrated with semantic information as text representation.Finally,an end-to-end text generation model is used for summary generation.The experimental results show that the method can better integrate the semantic information and the nuclearity structure information,and greatly improve the quality of the generated summarization.Finally,in view of the problem of missing nuclearity information in most texts,this paper proposes a text summarization method based on nuclearity recognition.This method uses summary extraction as the main task and nuclearity recognition as the auxiliary task.The two methods are joint learning to improve the accuracy of summary extraction.The experimental results show that the method can effectively play the role of the nuclearity information in the summarization task in the absence of the nuclearity,and obtain a high-quality summarization.