微信健康类谣言特征分析与早期检测模型研究
Feature Analysis and Study on Early Detection Model of WeChat Health-related Rumors
王晓艳 1林木辉2
作者信息
- 1. 福建师范大学协和学院,福州 350117
- 2. 福建师范大学教育学院,福州 350117
- 折叠
摘要
针对微信平台健康类谣言肆虐问题,揭示微信健康类谣言的主要特征,构建可在信息发布初期识别谣言的检测模型.在源自微博谣言检测的10个基础特征之外,发掘了微信健康类谣言在文本修辞、发文动机、出版规范、编辑排版四个方面的18个新特征,并将利用深度神经网络模型获得的预分类结果作为深层语义特征表示,对人工构建的特征集进行扩展,在此基础上利用机器学习模型完成谣言检测.结果显示,构建的特征集和检测模型在谣言识别方面具有先进性,将检测准确度提升了 10％左右,且识别能力最强的6个特征中,有3个为本文提出.该方法能在信息发布初期有效识别微信健康类谣言,从而为微信信息生态治理提供理论参考和方法依据.
Abstract
Aiming at the problem that health-related rumors are rampant on WeChat platform,this paper reveals the main characteristics of health-related rumors on WeChat,and constructs a detection model capable of identifying rumors at the early stage of information dissemination.In addition to the 10 basic features derived from Weibo rumor detection,18 new features of health-related rumor on WeChat are explored,involving four aspects:text rhetoric,posting motivation,publishing norms and editing & typesetting.And the pre-classification results obtained by using the deep neural network model are used as the deep semantic feature representation to expand the artificially constructed feature set.On this basis,machine learning model is used to complete rumor detection.The results show that the feature set and detection model is advanced in rumor detecition,and the accuracy of the detection is improved by about 10％.This method can effectively identify WeChat health-related rumors at the initial stage of information dissemination,thus providing theoretical reference and methodological foundations basis for governing the information ecosystem on WeChat.
关键词
谣言检测/微信谣言/健康信息/虚假信息识别/检测特征Key words
rumor detection/WeChat rumor/health-related information/identification of false information/detection features引用本文复制引用
出版年
2024
NETL
NSTL
万方数据