Research On Automatic Text Summarization Technology Based On Deep Learnin

With the rapid popularization of China’s new generation of 5G communication technology,data such as digital media information on the Internet has exploded.As a primitive and conventional form of information,text text has many limiting factors such as lengthy space and long reading time.How to accurately and effectively obtain the required content required by users from a wide range of text data information,so as to save readers’ reading time for irrelevant text information,and the ability to significantly improve the efficiency of users’ access to information is an urgent problem to be solved.As a research hotspot in the field of natural language processing,this technology on automatic text summarization algorithm refers to quickly condensing the source text while ensuring that the key information of the original text is retained,and generating short text summaries based on its main content.However,the abstract text obtained by the traditional extractive text summarization algorithm has problems such as poor effect and insufficient semantic understanding.In response to the above issues,this article conducted the following research:First,the article proposes to build a text summary generation model based on the ALBERT-Uni LM model.The model combines the pre-training language model ALBERT with the Uni LM model,and realizes summary generation through two stages,namely the word vector parameter acquisition stage based on the pre-training model ALBERT and the summary generation stage based on the Uni LM model.The principle is to first encode the vector parameters of the input text through the ALl BERT model to obtain the text input sequence representation;then input the obtained input sequence into the Seq2 Seq LM of the Uni LM model,and perform task fine-tuning with the idea of migration learning to obtain the text summary.Secondly,the article proposes to build a Uni LM-PGN text summary generation model based on the fusion topic word attention mechanism.Fully integrate the key information of topic words in the source text sequence,so that the summary model uses the topic words in the source text sequence as prior knowledge to guide the generation of text summaries;in addition,this article introduces the topic word information of the article into the pointer generation through the attention mechanism In the network model,the Uni LM-PGN model can make the language model Uni LM-PGN more fully understand the sequence semantic information reflected by the text subject words to realize the generation of the summary text,so that the Uni LM-PGN model can significantly improve the ability to use the text subject word information,and generate a more comprehensive and appropriate generalization effect Abstract text to improve the quality of the abstract text.Finally,the article evaluates the effect of the above model on the NLPCC-2018 Chinese public data set,and the experiments have proved the effectiveness of the language model proposed in this article.