基于图神经网络和结构稀疏度的极化网络群体意识形态突显与框架建模
Modeling Ideological Salience and Framing in Polarized Online Groups with Graph Neural Networks and Structured Sparsity
摘要：
网络政治话语的日益两极分化需要自动检测和监控社交媒体中意识形态分歧的计算工具。我们引入了一种最低限度的监督方法，利用网络结构的在线论坛，特别是Reddit，检测极化的概念。我们沿着显着性和框架的维度模型极化，借鉴道德心理学的见解。我们的架构将图神经网络与结构化稀疏学习相结合，并产生概念和子reddits的表示，以捕获右翼和左翼激进化等时间意识形态动态。
英文摘要：
The increasing polarization of online political discourse calls for computational tools that automatically detect and monitor ideological divides in social media. We introduce a minimally supervised method that leverages the network structure of online discussion forums, specifically Reddit, to detect polarized concepts. We model polarization along the dimensions of salience and framing, drawing upon insights from moral psychology. Our architecture combines graph neural networks with structured sparsity learning and results in representations for concepts and subreddits that capture temporal ideological dynamics such as right-wing and left-wing radicalization.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
