Scholarlink

In this work, we propose a novel Transformer-based SeqGAN model tailored for generating Chinese social media text. The study enhances traditional SeqGAN by integrating a Transformer architecture, addressing the unique challenges of Chinese language generation, such as capturing long-range dependencies and maintaining text coherence. Comparative experiments with Seq2Seq and standard SeqGAN models demonstrate the superiority of the Transformer-based model, particularly in improving text diversity and fluency, as evidenced by higher unique n-gram percentages and BLEU scores. Generated text in Wellness category is specifically analyzed afterwards, and the results show that the text generated by this model is superior to that produced by the other two models to a particular degree.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Text Generation for Social Media Based on Generative Adversarial Networks : Focusing on Wellness Content

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Xuan Ge,
  • Peng Wu

摘要

In this work, we propose a novel Transformer-based SeqGAN model tailored for generating Chinese social media text. The study enhances traditional SeqGAN by integrating a Transformer architecture, addressing the unique challenges of Chinese language generation, such as capturing long-range dependencies and maintaining text coherence. Comparative experiments with Seq2Seq and standard SeqGAN models demonstrate the superiority of the Transformer-based model, particularly in improving text diversity and fluency, as evidenced by higher unique n-gram percentages and BLEU scores. Generated text in Wellness category is specifically analyzed afterwards, and the results show that the text generated by this model is superior to that produced by the other two models to a particular degree.