基于深度学习的虚假健康信息识别

随着互联网的迅猛发展，网上健康信息以几何速度增长，其中大量虚假健康信息给人们的生活带来了很大影响，但目前对虚假健康信息文本识别的研究非常缺乏，以往研究主要集中在识别微博上的谣言、伪造商品评论、垃圾邮件及虚假新闻等方面。鉴于此，采用基于词向量的深度神经网络模型和基于双向编码的语言表征模型，对互联网上流传广泛的健康信息文本进行自动分类，识别其中的虚假健康信息。实验中，深度网络模型比传统机器学习模型性能提高10%，融合Word2vec的深度神经网络模型比单独的CNN或Att-BiLSTM模型在分类性能上提高近7%。BERT模型表现最好，准确率高达88.1%。实验结果表明，深度学习可以有效识别虚假健康信息，并且通过大规模语料预训练获得的语言表征模型比基于词向量的深度神经网络模型性能更好。

False Health Information Recognition Based on Deep Learning

With the rapid development of the Internet， online health information has been growing exponentially. A lot of fake health information has a great effect on people’s daily life. However， there is a lack of research on text recognition of fake health information recognition. Existing research mainly focus on rumors on microblogs， fabricated product reviews， spam and fake news， etc. This paper utilizes a deep neural network based on word vector and a language presentation model based on bidirectional encoder to classify health information automatically， so that the fake health information can be recognized. In this experiment， the performance of the deep network model is 10% higher than the traditional machine learning model. The deep neural network model integrated with Word2vec improves the classification performance by nearly 7% compared with the CNN or Att-BiLSTM model alone. The BERT model performs best， with an accuracy rate of 88.1%. The experimental results show that the deep learning techniques can recognize fake health information effectively， and the language representation model pretrained with large-scale corpus performs better than the deep neural network model based on word vector.