A Text Generation Algorithm Based On Layered Text Metrics Generative Adversarial Networks

With the rapid development of internet social media in recent years,massive amounts of text data are generated and disseminated daily.With the increase in people's cost requirements,automated extraction of key information from text information on the internet and the generation of corresponding texts are becoming more and more important.Through the text generation algorithm to output the semantic coherence,the simulative text with smooth order can save labor cost.The main work of this paper is to study the text generation model based on the generative adversarial network model to achieve the text generation task,and to explore the shortcomings in the text generation process and the traditional text generation model and propose corresponding improvements.The research work of this paper mainly includes the following aspects:1)In response to the common problem of discrete data features and corresponding feature information in text generation models,this article studies and analyzes the traditional sequence-to-sequence model in feature extraction according to the different feature extraction processes in the text generation process Lack,improve the existing feature acquisition process,propose a supplementary mechanism based on the feature space of the original data set,and construct a feature-enhanced text generation model.By optimizing the traditional feature extraction process,the feature space integrity is improved.Experiments show that the feature enhancement model can effectively use text features and more effectively enhance text generation effects through features.2)Aiming at the word-level word order in the process of text generation,this paper proposes a symmetric training model based on the generational confrontation network,inspired by mirror symmetry,according to the requirements of the original text information processing in the text generation process.Generative adversarial networks areused to ensure the effectiveness of the generation process,and the symmetric training mechanism is used to improve the performance requirements on word-level word order in the text generation process.The text generation tasks are divided into three steps:code generation,symmetric training,and authenticity judgment.The experimental results prove that the symmetrical adversarial training network can effectively improve the word-level word order optimization problem in the generation process and improve the overall text generation effect.3)Aiming at the problem of the overall semantics and word-level word order of generated text,this paper improves the text generation model generation process by learning the overall semantically dominant word-level word order during natural language formation,and proposes a hierarchical text generation model.Based on the symmetrical confrontation text generation model,a two-layer codec model is used to process important features in semantics and word order,and a hierarchical symmetric confrontation training model is constructed.Experiments show that the readability of the hierarchical symmetric adversarial training model in generating text is better than the traditional text generating model,and it effectively reduces the problem of word order and semantic intersection and improves the text generation effect.