Research On Topic-oriented Text Summary Generation

The text summarization technology is to summarize a given single or multiple texts to obtain a text summary.The abstract should not only reflect the main content of the original document,but also keep it as concise and clear as possible.In recent years,due to the rapid development of deep learning technology,text summarization technology has also undergone great changes.It is no longer only the traditional extractive text summarization technology,but has transformed into a text summarization technology that coexists with extractive and generative types.When multiple topics coexist in the original document,most of the current generative summary methods will make a more comprehensive summary and generalization of the content of these multiple topics,and generate text summaries containing these topics as much as possible.Generate a more detailed summary on one of these topics.In response to this problem,this paper proposes a topic-oriented text summary generation method.This method uses the attention mechanism on the decoding side to fuse the topic information of the specific topic and the topic contained in the text,given a specific topic,so that the text that contains the specific topic gets a greater attention weight,and thus obtains Text summaries whose content is more relevant to a specific topic.The main contents of this paper are as follows:(1)Based on the sequence-to-sequence model in the neural network,the model is improved,and the attention mechanism is introduced on the decoder side,so that the decoder can generate a set of attention weights for the input information in a targeted manner,so that different parts of the input text can be calculated.Different semantic content of the text can be used to generate different abstracts,and the text abstracts can be more emotionally reflected.The experimental results on the YELP data set verify that the improvement of the model in this article is indeed effective.(2)Based on the improvement of the sequence-to-sequence model,a topic-oriented text summary generation method is proposed,the topic concept and its vector representation are introduced,and the attention mechanism is used to combine topic information to obtain a topic-oriented text summary generation model.The vector representation of the subject information,encoder and decoder are analyzed in detail,which makes the program operation flow concise and clear,and the data flow is clear,laying the foundation for further experiments.(3)Experiments were performed on the large-scale English text data set YELP.The subject-oriented text summary generation method was compared with the benchmark method.The experimental results verified the feasibility and effectiveness of the method proposed in this article.