Hybrid Machine-Crowd Approach for Fake News Detection
Hybrid Machine-Crowd Approach for Fake News Detection
复制标题
用于假新闻检测的混合机器-人群方法
DOI：
10.1109/cic.2018.00048
复制
发表时间：
2018
期刊：
影响因子：
--
通讯作者：
M. Sokhn
中科院分区：
文献类型：
--
作者：
Shaban Shabani;M. Sokhn
假新闻的快速增长，特别是在社交媒体上，已成为一个具有挑战性的问题，在全球范围内产生负面社会影响。与意图欺骗和操纵读者的假新闻相反，讽刺故事旨在通过抨击或批评社会人物来娱乐读者。由于误导性信息的严重威胁，研究人员、政府、记者和事实核查志愿者正在共同努力解决假新闻问题，并加强数字媒体的问责制。自动假新闻检测系统能够识别欺骗性新闻。低精度仍然是这些系统的主要缺点。仅使用新闻内容进行自动检测是一项具有技术挑战性的任务，因为这些文章中使用的语言是为了绕过假新闻检测器。当任务是区分讽刺故事和假新闻时，这变得更加复杂。另一方面，人类的认知技能在执行此类任务时表现优于基于机器的系统。在本文中，我们通过提出一种使用混合机器人群方法检测潜在欺骗性新闻的方法来解决假新闻和讽刺检测问题。该系统将人的因素与机器学习方法和决策模型相结合，该模型估计算法的分类置信度并决定任务是否需要人工输入。与报告的基线结果相比，我们的方法实现了合理的更高准确性，以换取使用众包服务的成本和延迟。
The rapid growth of fake news, especially in social media has become a challenging problem that has negative social impacts on a global scale. In contrast to fake news which intend to deceive and manipulate the reader, satirical stories are designed to entertain the reader by ridiculing or criticizing a social figure. Due to its serious threats of misleading information, researchers, governments, journalists and fact-checking volunteers are working together to address the fake news issue and increase the accountability of digital media. The automatic fake news detection systems enable identification of deceptive news. Low accuracy remains the main drawback of these systems. The automatic detection using only news' content is a technically challenging task as the language used in these articles is made to bypass the fake news detectors. This becomes even more complicated when the task is to differentiate the satirical stories from fake news. On the other side, human cognitive skills have shown to overperform machine-based systems when it comes to such tasks. In this paper, we address the fake news and satire detection by proposing a method that uses a hybrid machine-crowd approach for detection of potentially deceptive news. This system combines the human factor with the machine learning approach and a decision-making model that estimates the classification confidence of algorithms and decides whether the task needs human input or not. Our approach achieves reasonably higher accuracy compared to the reported baseline results, in exchange of cost and latency of using the crowdsourcing service.