中文

基于嵌入的偏激新闻检测方法

机器学习 2025-07-09 v3 计算与语言

摘要

本报告描述了用于确定给定新闻文章是否可能被视为偏激新闻(hyperpartisan)的系统。偏激新闻以极度两极化的政治立场为特点,旨在创建公众的政治分裂。 Several方法,包括n-gram、情感分析,以及使用预训练ELMo模型进行句子和文档表示,均被采用。最佳系统使用LLMs进行嵌入生成,准确率约为92%,优于之前最好的系统(使用预训练ELMo和双向LSTM),后者通过10折交叉验证实现了约83%的准确率。

关键词

引用

@article{arxiv.2501.01370,
  title  = {Embedding-Based Approaches to Hyperpartisan News Detection},
  author = {Karthik Mohan},
  journal= {arXiv preprint arXiv:2501.01370},
  year   = {2025}
}

备注

Updated version reflecting sole authorship. All coauthor contributions have been removed. Experimental corrections and analysis updates were introduced in the original version and are retained here as part of the submitter's independent work, along with expanded experiments by the submitter