Research On Text Summary Generation Technology Based On Deep Learning

At present,the development of the Internet is growing day by day.With the advent of the information explosion era,the refinement and extraction of information has become an important research topic,and text Abstract extraction is one of the most important links.Existing text summary extraction systems usually use statistical or rule-based methods to process text.There are few methods based on deep learning.Therefore,this paper adopts deep learning method to implement a single text summary generation model based on Seq2Seq+Attention mechanism and Transformer framework,which is done on LCSTS data set and NLPCC data set respectively.Relevant validation analysis was carried out,and ROUGE evaluation index was used to evaluate and compare the results.Firstly,this paper introduces some advantages and disadvantages of traditional abstract extraction methods in detail,and introduces the principles and steps of TF-IDF algorithm and TextRank algorithm,and compares them with the methods based on in-depth learning.Secondly,because the traditional method does not consider the context and semantic information,this paper implements a text summary generation model based on Seq2Seq+Attention mechanism,and uses the Transformer framework in the task of text summary generation for the first time,comparing with the model based on Seq2Seq+Attention mechanism.Experiments show that the text summary generation model based on Transfor mer framework is in the leading position in these aspects by comparing the confusio n,accuracy and Loss values.Finally,two methods of text summary generation based on deep learning model and two traditional methods of text summary extraction are compared on the test set of LCSTS data set and NLPCC data set.Experiments show that,because the training set uses LCSTS data sets,the two text summary generation models based on in-depth learning are much better than the traditional methods,and the model based on Transformer framework is also better than the model based on Seq2Seq+Attention mechanism.On NLPCC data sets,the effect of in-depth learning model is slightly worse than that based on Seq2Seq+Attention mechanism.In the traditional two methods.The reason is that the deep learning model has great limitations.When dealing with unfamiliar types of texts,the effect is not as good as the traditional methods.Only when dealing with texts with the same type of training set,can it show its superiority.In addition,this paper designs and implements a Chinese text information extraction system.The core functions of the system are compared with open source tools.Experime nts show that the system has better practical application value than some open source tools.