Online polarization research currently focuses on studying single-issue opinion distributions or computing distance metrics of interaction network structures. Limited data availability often restricts studies to positive interaction data, which can misrepresent the reality of a discussion. We introduce a novel framework that aims at combining these three aspects, content and interactions, as well as their nature (positive or negative), while challenging the prevailing notion of polarization as an umbrella term for all forms of online conflict or opposing opinions. In our approach, built on the concepts of cleavage structures and structural balance of signed social networks, we factorize polarization into two distinct metrics: Antagonism and Alignment. Antagonism quantifies hostility in online discussions, based on the reactions of users to content. Alignment uses signed structural information encoded in long-term user-user relations on the platform to describe how well user interactions fit the global and/or traditional sides of discussion. We can analyse the change of these metrics through time, localizing both relevant trends but also sudden changes that can be mapped to specific contexts or events. We apply our methods to two distinct platforms: Birdwatch, a US crowd-based fact-checking extension of Twitter, and DerStandard, an Austrian online newspaper with discussion forums. In these two use cases, we find that our framework is capable of describing the global status of the groups of users (identification of cleavages) while also providing relevant findings on specific issues or in specific time frames. Furthermore, we show that our four metrics describe distinct phenomena, emphasizing their independent consideration for unpacking polarization complexities.

翻译：中文翻译摘要：当前在线极化研究主要聚焦于单一议题的观点分布或交互网络结构的距离度量。受数据可用性限制，研究常局限于正向交互数据，这可能扭曲讨论的真实面貌。本文提出一种融合内容、交互及其性质（正向或负向）三个维度的新框架，同时挑战将极化作为所有在线冲突或对立意见统称的主流观点。基于分裂结构理论与符号社交网络的结构平衡概念，我们将极化分解为两个独立度量指标：对抗性（Antagonism）与对齐性（Alignment）。对抗性通过用户对内容的反应量化在线讨论中的敌对程度；对齐性则利用平台中长期用户-用户关系编码的符号结构信息，描述用户交互与讨论全局立场/传统立场的契合程度。我们可追踪这些指标随时间的变化，既识别相关趋势，也能定位可映射至具体语境或事件的突变。该方法在两大平台得到验证：美国众包事实核查平台Birdwatch（Twitter扩展应用）与奥地利在线报纸讨论论坛DerStandard。在这两个案例中，我们的框架既能描述用户群体的全局状态（识别分裂结构），也能在特定议题或时间窗口内提供有效发现。此外，四个度量指标揭示了截然不同的现象，强调需独立考量以解析极化复杂性。