Research On Story Ending Generation Technology Based On Deep Learning

With the rapid development of deep learning technology,the research in the field of text generation based on deep learning is getting deeper and deeper.Among them,the task of story ending generation is a popular pure text generation task,which needs to understand the context information effectively and then generate the consistent story ending.There are mainly some shortcomings in the existing models of story ending generation task: 1)Simply splicing the context information into a long sequence as the input of the model,which is easy to lose some important information,resulting in insufficient coding context information;2)Although some models have built information between adjacent sentences,there is no bridge construction for the relationship between sentences,which will greatly hinder the integrity of the overall information when encoding context information;3)The internal information of the sentence has not been fully mined.In order to solve the shortcomings of the story ending generation task,this dissertation constructs a model based on the graph convolution and dependency syntax parsing tree.The graph convolution is used to transfer context information,and the context information is modeled in a dimensional space to capture information across sentences.The dependency syntax parsing tree can capture the intra-sentence information and take the resolved relations as the edges of the graph convolution network.This construction method can make up for the deficiency in the current task of story ending generation.In the published ROCStories dataset,experimental results show that the model based on graph convolution and dependency syntax parsing tree achieves better results then the existing models,which also indicates that the model can effectively improve the defects of the current task.This dissertation describes the research of deep learn-based technology in the task of story ending generation,includingseq2 Seq attention model,transformer framework model,improved progressive transformer model,graph convolution model and graph convolution and dependency syntax parsing,which this dissertation focuses on the graph convolution and dependency syntax parsing tree model.