ZHIHU · OCT 2015
Who gets heard? How close are the people at the top? An exploration of a 2015 Zhihu sample, through four questions.谁被看见？大V之间有多近？从四个问题，探索 2015 年知乎的一份样本。
2015 study · 2016 essays · 2026 interactive edition2015 年研究 · 2016 年文章 · 2026 年交互呈现
375 people人 / 27,324 following links条关注Reciprocal view · explore the network互相关注视角 · 进入网络 ↗Start with the connections. Switch scales, inspect a user’s neighbours, or zoom out to the relationships between structural groups.先看连接本身。切换网络规模，走进一个节点的邻居，或者退一步看不同分组怎样相连。
Real 2015 following links, rebuilt from the verified archive. Node size shows incoming follows within this network. Colours identify computed structural groups, not professions or topics. Layout distance is not social distance. Only names already published in the original articles are labelled; other accounts stay anonymous. User numbers are ordered by incoming follows within each network.由核验过的 2015 年归档重建。点越大，在这张网络内被关注越多。颜色区分算法计算的结构分组，不代表职业或话题；图上的距离不等于社交距离。仅标出原文已发表的姓名，其余账号匿名；编号按网内被关注数排序。
Net10k recount contains 1,896 nodes, one more than the 1,895 in the published table; both have 231,416 links. Historical statistics below keep the original numbers. Background ink is sampled when needed; selecting a node always shows all its neighbours.Net10k 从归档复计为 1,896 个节点，比原文的 1,895 多 1 个；连接数同为 231,416。下方历史统计保留原文数字。概览背景按需抽取连线；选中节点后，会显示它的全部邻居。
¹ Path lengths are calculated only within each network’s largest strongly connected component. Density is reported at the original precision.¹ 路径长度仅在各网络的最大强连通分量内计算。密度沿用原文精度。
A → B → C → D: 3 following links.A → B → C → D：经过 3 条关注连接。
These are overlapping groups selected by upvotes, with different sizes. Their density difference describes this sample; it does not by itself prove a preference for forming elite circles. A strongly connected component means that directed paths exist both ways between every pair of its users.这两个群体按赞同数筛选，彼此重叠、规模也不同。密度差异描述了这个样本，不能单凭它证明大V偏好抱团。强连通分量意味着其中任意两人之间，都有沿关注方向彼此到达的路径。
Read the network analysis阅读关注网络分析A few very large counts pull the mean away from the middle. Select a metric to see the gap.少数极大的数值，把均值拉离了中间位置。选一个指标，看看这个差距。
40.2×The mean is this many times the median.均值是中位数的这么多倍。
Recounted from the recovered 2015 archive on 1 Oct 2026. Bar length shows users per interval, on a linear scale. Intervals have different widths; this is not a probability-density plot.2026-10-01 从恢复的 2015 年归档重新计数。柱长线性表示各区间的用户数；区间宽度不同，这不是概率密度图。
32.3×The mean is this many times the median.均值是中位数的这么多倍。
Recounted from the recovered 2015 archive on 1 Oct 2026. Bar length shows users per interval, on a linear scale. Intervals have different widths; this is not a probability-density plot.2026-10-01 从恢复的 2015 年归档重新计数。柱长线性表示各区间的用户数；区间宽度不同，这不是概率密度图。
2.6×The mean is this many times the median.均值是中位数的这么多倍。
Recounted from the recovered 2015 archive on 1 Oct 2026. Bar length shows users per interval, on a linear scale. Intervals have different widths; this is not a probability-density plot.2026-10-01 从恢复的 2015 年归档重新计数。柱长线性表示各区间的用户数；区间宽度不同，这不是概率密度图。
4.1×The mean is this many times the median.均值是中位数的这么多倍。
Recounted from the recovered 2015 archive on 1 Oct 2026. Bar length shows users per interval, on a linear scale. Intervals have different widths; this is not a probability-density plot.2026-10-01 从恢复的 2015 年归档重新计数。柱长线性表示各区间的用户数；区间宽度不同，这不是概率密度图。
30.9×The mean is this many times the median.均值是中位数的这么多倍。
Recounted from the recovered 2015 archive on 1 Oct 2026. Bar length shows users per interval, on a linear scale. Intervals have different widths; this is not a probability-density plot.2026-10-01 从恢复的 2015 年归档重新计数。柱长线性表示各区间的用户数；区间宽度不同，这不是概率密度图。
Bars share a zero baseline within each metric. A long tail alone does not establish a power-law distribution.同一指标的两条柱从零起算。长尾现象本身不足以证明幂律分布。
| Metric指标 | Mean均值 | Median中位数 | Std. deviation标准差 |
|---|---|---|---|
| Upvotes赞同 | 3,858.4 | 96 | 21,951.4 |
| Followers粉丝 | 3,620.0 | 112 | 22,978.5 |
| Following关注 | 176.3 | 67 | 565.9 |
| Answers回答 | 68.9 | 17 | 225.9 |
| Thanks感谢 | 865.3 | 28 | 4,627.6 |
The answer changes with the question. Compare the published leaders under three network measures.衡量方式改变，答案也会改变。对照三种网络指标下，原文发表的前五名。
Attention passed through the network沿关注网络传递的关注权重
Followed by high-scoring hubs受到高分枢纽关注的节点
Following high-scoring authorities关注高分权威节点的节点
Attention passed through the network沿关注网络传递的关注权重
Followed by high-scoring hubs受到高分枢纽关注的节点
Following high-scoring authorities关注高分权威节点的节点
Scores belong to different algorithms and networks: compare positions, not score magnitudes across lists. An absent name means “not in the published Top 5”, not a zero score. Names are as published in 2016.不同算法、不同网络的分数不在同一尺度上，应对照名次。名字未出现表示“未列入原文前五”，不代表分数为零。姓名沿用 2016 年原文。
These are measures of position in a following network. They are not direct measures of expertise or answer quality. PageRank and HITS answer related but distinct questions; PageRank is not the product of the two HITS scores.这里衡量的是用户在关注网络中的位置，并不直接衡量专业水平或回答质量。PageRank 与 HITS 回答的是相关但不同的问题；PageRank 并不是两种 HITS 分数的乘积。
Read the original rankings阅读原文榜单Two lists, one shared scale. Explore the overlap and the differences in the questions these selected users answered.两份列表，共用一个刻度。看看这些选定用户所答问题的话题，有哪些交集与不同。
Counts of question tags in answers by a selected dominating set. The two samples have different sizes. Unreported is not zero.统计各网络支配集用户所答问题的话题标签次数。两组样本规模不同。“未列出”不等于零。
| Topic话题 | Net10k | Net50k |
|---|---|---|
| Survey questions调查类问题 | 3,792 | 1,365 |
| Life生活 | 3,096 | 1,435 |
| History历史 | 1,713 | 1,204 |
| Dating恋爱 | 1,464 | 717 |
| Psychology心理学 | 1,432 | 634 |
| Film电影 | 1,419 | 1,084 |
| Social relationships人际交往 | 1,404 | 640 |
| Society社会 | 1,332 | 984 |
| Internet互联网 | 1,214 | 595 |
| Emotions情感 | 1,197 | Unreported未列出 |
| Politics政治 | 1,028 | 1,285 |
| Gender relations两性关系 | 994 | 688 |
| Education教育 | 897 | Unreported未列出 |
| China中国 | 823 | 695 |
| Life choices人生 | 815 | Unreported未列出 |
| Games游戏 | 805 | Unreported未列出 |
| Literature文学 | 772 | Unreported未列出 |
| Zhihu知乎 | 772 | Unreported未列出 |
| Law法律 | 750 | 587 |
| Music音乐 | 738 | Unreported未列出 |
| Love爱情 | 699 | Unreported未列出 |
| Culture文化 | 659 | Unreported未列出 |
| Entrepreneurship创业 | 628 | Unreported未列出 |
| University大学 | 621 | Unreported未列出 |
| Programmers程序员 | 619 | Unreported未列出 |
| Mental life心理 | 617 | Unreported未列出 |
| How would you rate X?你如何评价 X | 609 | Unreported未列出 |
| Women女性 | 604 | Unreported未列出 |
| Programming编程 | 585 | 511 |
| What is X like?X 是种怎样的体验 | 582 | Unreported未列出 |
| Health健康 | Unreported未列出 | 996 |
| Medicine医学 | Unreported未列出 | 941 |
| English英语 | Unreported未列出 | 678 |
| Microsoft微软（Microsoft） | Unreported未列出 | 555 |
| United States美国 | Unreported未列出 | 552 |
| Fitness健身 | Unreported未列出 | 538 |
The article labels both lists “Top 20”, but prints 30 entries for Net10k and 20 for Net50k. All are retained here. These are raw tag counts, not percentages of users or a platform-wide topic survey.原文两组标题都写“Top 20”，但实际列出 Net10k 30 项、Net50k 20 项，这里全部保留。这些是标签次数，并非用户占比或全站话题调查。
The crawl began with Zhao Che’s account and followed outgoing links for two steps. The seed and both layers total three levels. This was not a random sample of Zhihu, and external follow targets are not all complete user profiles.爬取从 Zhao Che 的账号出发，沿关注关系扩展两层，连同种子共三层。这不是知乎的随机样本；向外关注连接的终点，也并非都有完整用户资料。
The tables and rankings reproduce the original published results. Profile distributions were recounted from the recovered archive in 2026, and all five means and medians were checked against the original. Anonymous network views preserve the real archived following links; their grouping and layout were prepared in 2026. Later exploratory experiments are not presented as confirmed findings.统计表和榜单沿用原研究已发表的结果。用户分布于 2026 年从恢复的归档重新计数，五项均值、中位数也与原文核对一致。匿名网络图保留归档中的真实关注连接，分组与布局于 2026 年重新计算。后来的探索性实验没有被当作已确认结论。
Long tails need distribution tests; denser groups need comparison baselines; low betweenness alone does not establish resilience to removing users. The sample does not describe today’s Zhihu.长尾需要分布检验；密集群体需要比较基线；低介性本身不能证明删除用户后的网络韧性。这份样本也不能描述今天的知乎。
A 2015 Social Computing course project at CUHK by Gu Zhijing, Huang Xianghai, Lyu Zishen and Zhao Che. Zhao Che worked on the database, basic statistics, network analysis, proposal, presentation and report.2015 年香港中文大学 Social Computing 课程小组项目，由 Gu Zhijing、Huang Xianghai、Lyu Zishen 和 Zhao Che 共同完成。Zhao Che 参与数据库、基本统计、网络分析、提案、演示及报告。