基于语义增强的虚假新闻检测

FALSE NEWS DETECTION BASED ON SEMANTIC ENHANCEMENT

  • 摘要: 针对将虚假新闻检测抽象为文本分类任务时,可能会忽略新闻文本语义信息的问题,提出一种语义增强的虚假新闻检测方法。使用TextRank算法提取真假新闻的关键词,并融入原始文本中进行信息增强,利用ERNIE模型学习知识增强的语义表示,通过CNN模型提取新闻文本局部特征,并输入到BiGRU学习序列特征,同时引入注意力机制突出关键特征词,在分类前将特征向量与知识增强的语义表示进行融合,实现虚假新闻检测。实验结果表明,该方法能够有效分类真假新闻,在虚假新闻检测任务中比常用方法准确率有显著提升。

     

    Abstract: Aimed at the problem that when false news detection is abstracted as a text classification task, the semantic information of news text may be ignored, a semantically enhanced false news detection method is proposed. The TextRank algorithm was used to extract the keywords of true and false news, and we integrated them into the original text for information enhancement. The ERNIE model was used to learn the semantic representation of knowledge enhancement, and extracted the local features of the news text through the CNN model, and input them to BiGRU to learn the sequence features. At the same time, the attention mechanism was introduced to highlight key feature words, and the feature vector was integrated with the semantic representation of knowledge enhancement before classification to realize false news detection. Experimental results show that the proposed method can effectively classify true and false news, and has a significant improvement in accuracy compared with common methods in false news detection tasks.

     

/

返回文章
返回