Research And Implementation Of Text Automatic Generative System Based On Pca Dimension Reduction

In the rapid development of this era,with the explosion of text information,people need to obtain information more conveniently to meet the needs of daily life.The acquisition of text summarization’s technology naturally becomes important in the field of text processing.With the natural language processing’s technology widely used in daily life,efficient text summarization processing technology emerges as the times require.At present,in the field of text generation,the sequence-to-sequence structure is widely be used.However,this constructure consumes too much time to generate abstract,while the generative summary has the problem of poor effect.To solve the above problems,this paper proposes a way to combine extractive abstract with generative abstract.At the same time,the insecurity of text content has also caused great loss to the personal rights.In order to effectively combat all kinds of telecommunications fraud activities,this article adds the security test of text information to effectively screen illegal content.The main contents of this paper are as follows.1)This paper firstly introduces some common detection methods of text content security detection in detail.After summarizing the shortcomings of these methods,this paper proposes a method of using Bert,Nezha and other advanced nlp models to test unsafe text.2)Next,this paper introduces the dataset,models and corresponding evaluation indicators commonly used in the previous text summary field,so that MT5 is this paper’s focus.This topic realizes the generation of sentence vector based on the extracted weight from nezha model,designs and implements an expansive gate convolution neural network that can better capture the article’s information,and applies PC A dimensioin reduction to get reducive sentence vector.3)Design and improve the generative model.At present,the most commonly used generative model is the MT5.By using the key and value calculated before to avoid repeative calculation,the generative speed of the model is accelerated.Some of the research results of this paper have been published before,aiming to provide the theoretical support and experimental basis for Chinese text generation.