📚 Who Controls the Cult? A Graph Theory Analysis | 谁在控制邪教？图论分析
Understanding the control structure of a secretive group such as a cult is not only a sociological or psychological challenge, but also a mathematical one. By modelling interpersonal relationships as a network and applying graph theory metrics, we can objectively identify the individuals who exert the most influence, broker information, and hold the group together. This article explores how centrality measures, clustering coefficients, and basic game theory can help answer the question: Who truly controls the cult?
理解像邪教这样隐秘团体的控制结构，不仅是社会学或心理学的难题，也是一个数学问题。通过将人际关系建模为网络并应用图论度量，我们可以客观地识别出那些施加最大影响力、居中传递信息并维系整个团体的个人。本文将探讨如何利用中心性测度、聚类系数以及基本的博弈论来回答一个问题：究竟谁在控制邪教？
1. Modelling a Cult as a Network | 将邪教建模为网络
sent members and edges E represent communication, trust, or obedience links. In the simplest case, an edge exists between two members if they interact regularly or share a direct reporting relationship. This abstraction allows us to apply rigorous mathematical tools to dissect the power dynamics without subjective bias.
在网络理论中，邪教可以表示为一个无向或有向图 G = (V, E)，其中顶点 V 代表成员，边 E 代表沟通、信任或服从关系。在最简单的情形下，如果两名成员定期互动或存在直接汇报关系，则他们之间存在一条边。这种抽象使我们能够运用严格的数学工具来剖析权力动态，而避免主观偏见。
For a small cult of n = 10 members, we might label the vertices A, B, C, ..., J. The graph's adjacency matrix A is an n × n matrix where aᵢⱼ erwise. Once this matrix is constructed, we are ready to compute several standard metrics.
对于一个 n = 10 人的小型邪教，我们可以将顶点标记为 A, B, C, …, J。图的邻接矩阵 A 是一个 n × n 矩阵，若成员 i 与成员 j 之间存在连接则 aᵢⱼ = 1，否则为 0。一旦构造出该矩阵，我们就可以开始计算多个标准度量。
2. Data and Assumptions | 数据与假设
To illustrate the methodology, let us assume the following connection structure: Leader A is directly connected to everyone; lieutenants B and C are connected to all other members except each other; ordinary members D, E, F, G, H, I, J are connected only to A, B, and C, forming a star-like core with two intermediaries. This toy model reflects a typical authoritarian cult where power concentrates in a small elite, but the leader needs trusted aides to manage the rank and file.
为了演示该方法，我们假设如下的连接结构：领袖 A 与所有人直接相连；副手 B 和 C 与除彼此之外的所有成员相连；普通成员 D、E、F、G、H、I、J 仅与 A、B 和 C 相连，形成一个带有两个中间层的类星型核心。这个简单模型反映了一个典型的威权邪教，权力集中在一小群精英手中，但领袖需要信任的助手来管理普通成员。
The adjacency list and visual inspection already suggest A is central, but mathematics can quantify precisely how central each member is, and more subtly, reveal whether B or C might actually control the flow of information more than A. We will use three classic centrality measures: degree, betweenness, and eigenvector centrality.
邻接列表和目视检查已经暗示 A 是中心的，但数学可以精确量化每个成员的中心程度，并且更巧妙地揭示出 B 或 C 是否可能实际上比 A 更能控制信息流动。我们将使用三个经典的中心性测度：度中心性、介数中心性和特征向量中心性。
3. Degree Centrality: Who Has the Most Connections? | 度中心性：谁拥有最多连接？
Degree centrality is the simplest measure. For an undirected graph, the degree deg(v) is the number of edges incident to vertex v. The normalised degree centrality is CD(v) = deg(v) / (n − 1), which ranges from 0 to 1. In our network, A has degree 9, so CD(A) = 9/9 = 1. B and C each have degree 8, giving CD(B) = CD(C) = 8/9 ≈ 0.889. Ordinary members have degree 3, yielding CD ≈ 0.333. Degree centrality clearly flags Leader A as the most connected individual.
度中心性是最简单的度量。对于无向图，度 deg(v) 是顶点 v 所关联的边的数量。标准化的度中心性为 CD(v) = deg(v) / (n − 1)，取值范围为 0 到 1。在我们的网络中，A 的度为 9，因此 CD(A) = 9/9 = 1。B 和 C 的度各为 8，得到 CD(B) = CD(C) = 8/9 ≈ 0.889。普通成员的度为 3，得到 CD ≈ 0.333。度中心性明确地将领袖 A 标记为连接最广的个体。
However, degree does not capture the control of information flow. A member who is highly connected but does not lie on the shortest paths between others may be easily bypassed. To identify who brokers information, we need betweenness centrality.
然而，度并不反映对信息流的控制。一个连接广泛但不在他人之间最短路径上的成员，可能很容易被绕过。要识别出谁在居中传递信息，我们需要介数中心性。
4. Betweenness Centrality: Controlling Information Flow | 介数中心性：控制信息流
Betweenness centrality quantifies how often a vertex acts as a bridge along the shortest path between two other vertices. It is defined as CB(v) = Σs≠v≠t σst(v) / σst, where σst is the total number of shortest paths from s to t, and σst(v) is the number of those paths that pass through v. For an n-vertex graph, this raw score is often normalised by dividing by (n−1)(n−2)/2 for undirected graphs.
介数中心性量化一个顶点在其他两个顶点之间最短路径上充当桥梁的频繁程度。其定义为 CB(v) = Σs≠v≠t σst(v) / σst，其中 σst 是从 s 到 t 的最短路径总数，σst(v) 是通过 v 的那些路径的数目。对于 n 个顶点的无向图，通常用 (n−1)(n−2)/2 进行归一化。
In our network, all shortest paths between ordinary members (e.g., D to E) must pass through A, B, or C. However, because A is connected to everyone, many shortest paths go directly through A. B and C, lacking direct connection to each other, see less betweenness than A. Computing exact values: A lies on every path that does not already find a shorter alternative, giving a high normalised betweenness close to 0.56. B and C score around 0.22. This suggests A is a major bottleneck, but interestingly, if B and C were to form a hidden back channel between themselves, betweenness centrality would shift dramatically.
在我们的网络中，普通成员（如 D 到 E）之间的所有最短路径都必须经过 A、B 或 C。然而，由于 A 与所有人相连，许多最短路径直接经过 A。B 和 C 由于彼此没有直接连接，其介数低于 A。计算精确数值：A 几乎处于所有无法找到更短替代方案的路径上，归一化介数接近 0.56。B 和 C 的得分约为 0.22。这表明 A 是一个主要瓶颈，但有趣的是，如果 B 和 C 之间形成了一个隐秘的后门通道，介数中心性将发生剧烈变化。
5. Eigenvector Centrality: Influence by Association | 特征向量中心性：关联影响力
Eigenvector centrality extends degree centrality by accounting for the importance of one' quation is Ax = λx, where x is the eigenvector and its entries give the centrality scores.
特征向量中心性对度中心性进行了扩展，考虑了邻居的重要性。与有权势者相连的人，其得分高于与无足轻重者相连的人。它通过求邻接矩阵 A 的最大特征值所对应的特征向量来计算。定义方程为 Ax = λx，其中 x 为特征向量，其分量即为各节点的中心性得分。
In our model, A is connected to everyone, including B and C who themselves are connected to many others. Therefore A's eigenvector centrality is highest, followed by B and C with nearly equal scores. The ordinary members receive low scores because their only connections are to the already central figures. The eigenvector measure confirms that the leader benefits not only from direct ties but also from the reflected status of their lieutenants.
在我们的模型中，A 与所有人相连，包括 B 和 C，而 B 和 C 本身又与许多人相连。因此 A 的特征向量中心性最高，B 和 C 的得分几乎相等，紧随其后。普通成员得分较低，因为他们仅与那些已然处于中心的人物相连。特征向量度量证实，领袖不仅从直接联系中获益，还从其副手的反射地位中获益。
6. Clustering and Cliques | 聚类与派系
Clustering coefficients measure the extent to which a vertex's neighbours are connected to each other. A high clustering coefficient around a leader may indicate a tight-knit inner circle that reinforces control. In our cult network, the neighbours of A (everyone) have many missing edges among themselves except those involving B and C. The local clustering coefficient for A is thus relatively low, while B and C show higher clustering because their neighbour sets include A and many common members.
聚类系数衡量一个顶点的邻居之间相互连接的程度。领袖周围的高聚类系数可能意味着一个紧密团结的核心圈子，从而巩固控制。在我们的邪教网络中，A 的邻居（所有人）之间除了涉及 B 和 C 的边以外，存在许多缺失边。因此 A 的局部聚类系数相对较低，而 B 和 C 的聚类系数较高，因为他们的邻居集包含 A 和许多共同成员。
Detecting cliques (complete subgraphs) can reveal cells or factions. The largest clique in this toy network is of size 3 (e.g., A-B-D, A-C-F), but there is no clique larger than that. This suggests a hierarchical rather than a clustered power structure. In real cults, such analysis might help identify hidden subgroups that could challenge the leader's authority.
检测派系（完全子图）可以揭示小组或派别。在这个玩具网络中，最大派系的规模为 3（例如 A-B-D，A-C-F），但没有更大的派系。这表明了一个层级式的而非聚团式的权力结构。在真实的邪教中，此类分析可能有助于识别可能挑战领袖权威的隐藏亚群体。
7. The Control Index: A Composite Measure | 控制指数：综合度量
a weighted sum of normalised degree, betweenness, and eigenvector centrality: CCI(v) = α·CD(v) + β·CB(v) + γ·CE hts depends on context: if information brokering is deemed critical, β gets a higher weight.
单一中心性度量无法提供控制力的全貌。我们可以定义一个综合控制指数 CCI(v) = α·CD(v) + β·CB(v) + γ·CE(v)，其中 α+β+γ = 1。权重的选择取决于具体情境：如果信息中转被视为至关重要，则 β 会被赋予更高的权重。
, if we dramatically increase the betweenness weight to β = 0.6, the gap narrows because A' ematically identifiable, the nuanced control depends on what we value most in the network.
使用等权重（各 1/3），领袖 A 的 CCI 约为 0.85，B 和 C 的得分在 0.68 左右。即使权重略有变化，A 依然高居榜首。但如果我们将介数权重大幅提升至 β = 0.6，差距就会缩小，因为 A 在介数上的优势并不像在度上那样具压倒性。这种敏感性分析表明，虽然明显的领袖在数学上可被识别，但微妙层面的控制力取决于我们在网络中最看重什么。
8. Dynamic Analysis: A Game Theory Perspective | 动态分析：博弈论视角
Static network measures assume a fixed structure, but control can shift when members strategically form or sever ties. Simple game theory can model whether a lieutenant benefits from bypassing the leader to create a direct link with another lieutenant. Consider a coordination game where each pair of members receives a payoff proportional to the information they can exchange independently of the leader. The Nash equilibrium of such a game may predict the emergence of a shadow network that dilutes the leader's control.
静态网络度量假设结构是固定不变的，但当成员策略性地建立或切断联系时，控制力可能会转移。简单的博弈论可以模拟副手是否会从绕过领袖、与其他副手建立直接联系中获益。设想一个协调博弈，每一对成员获得的收益与其能够独立于领袖进行交换的信息量成正比。该博弈的纳什均衡可能预测出一个影子网络的出现，从而削弱领袖的控制力。
In a repeated game setting, a lieutenant who invests in a hidden channel with another lieutenant can accumulate power over time, as measured by a dynamic centrality metric. This highlights that control is not static; it is continuously negotiated and can be captured by looking at network evolution.
在重复博弈设定中，一个副手若投资于与另一副手之间的隐秘通道，其权力会随着时间累积，这可以通过动态中心性度量来衡量。这突显出控制不是静态的；它是在持续协商中形成的，并且可以通过观察网络演化来捕捉。
9. Limitations and Ethical Considerations | 局限性与伦理考量
Network models are simplifications. In real cults, edges may represent different types of influence (financial, emotional, ideological) that cannot be simply aggregated into a single adjacency matrix. Missing data, especially about covert ties, can bias the metrics. Moreover, mathematical identification of a "controller" must be interpreted with care, as power can be diffused or based on non-network factors like charisma or coercion.
网络模型是简化处理。在真实的邪教中，边可能代表不同类型的影响（经济、情感、意识形态），无法简单地汇总为单一的邻接矩阵。数据缺失，特别是关于隐蔽联系的缺失，会使度量产生偏差。此外，对“控制者”的数学识别必须谨慎解读，因为权力可能是分散的，或基于非网络因素，如个人魅力或胁迫。
Ethically, using these techniques on living groups raises privacy concerns. Our analysis is intended purely as an educational exercise in applied mathematics and should not be used to profile real individuals without consent. Nevertheless, understanding these tools can aid authorities and researchers in deconstructing harmful power structures.
从伦理角度看，将这些技术用于在世的团体涉及隐私问题。我们的分析纯粹作为应用数学的教学练习，不应在未获同意的情况下用于对真实个人进行画像。然而，理解这些工具可以帮助当局和研究人员解构有害的权力结构。
10. Conclusion: Identifying the Puppeteer | 结论：识别幕后操纵者
So, who controls the cult? Mathematically, it depends on your definition of control. By degree and eigenvector centrality, the obvious leader dominates. By betweenness, the leader still leads, but trusted lieutenants are indispensable conduits. A composite index can synthesise these dimensions, and sensitivity analysis reveals that under certain value systems, the perceived controller might shift. Ultimately, network science does not give a single answer but provides a rigorous framework for asking smarter questions. In a cult, the visible leader may not be the only one pulling the strings.
那么，谁在控制邪教？从数学上讲，这取决于你对控制的定义。就度中心性和特征向量中心性而言，显而易见的领袖占据主导。就介数而言，领袖依然领先，但信赖的副手是不可或缺的管道。综合指数可以综合这些维度，而敏感性分析揭示，在某些价值体系下，认知中的控制者可能会发生转移。归根结底，网络科学并不提供单一答案，而是提供了一个严格的框架来提出更聪明的问题。在邪教中，台前领袖或许并非唯一的操线者。
Published by TutorHao | IB Mathematics Revision Series | aleveler.com
更多咨询请联系16621398022（同微信）
或截图后用「扫一扫 → 相册」选择本图