Scholarlink

<p>Chinese sequence labeling is an essential aspect of NLP applications like machine translation, sentiment analysis, and information retrieval. However, the absence of word boundaries in Chinese is a major hindrance, especially in academic settings where linguistic knowledge is paramount. Character segmentation and contextual interpretation challenges still affect task performance in applications like named entity recognition and pos tagging despite the dominating efforts. To bridge these loopholes, this paper introduces a deep learning model that improves sequence labeling performance with the application of transformers and graph convolutional networks (GCNs). Our method utilizes transformers’ multi-head attention to represent fine-grained contextual relationships and graph-based models to represent interactions between characters. Our models outperform typical baselines, as evidenced by experimental performance on benchmarking datasets, with efficient improvements in precision and recall. With these state-of-the-art methods, our study not only provides a technical innovation to Chinese sequence labeling but also enhances students’ reading literacy and linguistic analytical ability, and thus enables more efficient Chinese language instruction. Our study demonstrates deep learning can be used to tackle long-standing NLP challenges and enhance creativity in language instruction.</p>

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

A hybrid model combining GCN transformer and Word2Vec for Chinese sequence labeling with deep linguistic knowledge

  • 期刊论文

  • 发表日期：

    2025年8月27日

查看全文
  • Yang YaPing,
  • Zain Ul Abideen,
  • Amir Ali,
  • Muhammad Aoun,
  • Tehseen Mazhar,
  • Tariq Shahzad

摘要

Chinese sequence labeling is an essential aspect of NLP applications like machine translation, sentiment analysis, and information retrieval. However, the absence of word boundaries in Chinese is a major hindrance, especially in academic settings where linguistic knowledge is paramount. Character segmentation and contextual interpretation challenges still affect task performance in applications like named entity recognition and pos tagging despite the dominating efforts. To bridge these loopholes, this paper introduces a deep learning model that improves sequence labeling performance with the application of transformers and graph convolutional networks (GCNs). Our method utilizes transformers’ multi-head attention to represent fine-grained contextual relationships and graph-based models to represent interactions between characters. Our models outperform typical baselines, as evidenced by experimental performance on benchmarking datasets, with efficient improvements in precision and recall. With these state-of-the-art methods, our study not only provides a technical innovation to Chinese sequence labeling but also enhances students’ reading literacy and linguistic analytical ability, and thus enables more efficient Chinese language instruction. Our study demonstrates deep learning can be used to tackle long-standing NLP challenges and enhance creativity in language instruction.