Citation Network of Section 230 of the Communications Decency Act: A Validated Corpus of Published Federal Circuit Appellate Holdings, 1997–2025

This deposit provides the data and code for a computational legal study of Section 230 of the Communications Decency Act (47 U.S.C. § 230). It includes a validated corpus of 63 published federal circuit and Supreme Court opinions that substantively applied § 230 immunity doctrine between 1997 and 2025, the citation network constructed from those opinions (63 nodes, 320 directed edges, confirmed directed acyclic graph), all derived network metrics, and the complete analysis pipeline. The corpus was constructed by retrieving 211 candidate opinions from CourtListener (courtlistener.com) and subjecting each to systematic human review against a documented operative inclusion criterion and a ten-pattern exclusion taxonomy. The false positive rate in automated retrieval was 70.1% ((211 − 63) / 211 = 148 / 211), a methodological finding with implications for corpus construction in empirical legal studies. Corpus recall was 1.0 (95% Wilson CI: 0.942–1.000) based on a stratified sample of 20 cases across five temporal strata. Key findings: Citation authority concentrates in Zeran v. America Online, Inc., 129 F.3d 327 (4th Cir. 1997), which holds PageRank 0.200 (α = 0.85), accounting for 20.0% of total network authority and ranking first in every annual snapshot from 1997 through 2025. The Leiden algorithm identifies 7 communities (mean NMI stability = 0.621 across 100 seeds, SD = 0.121). Mean pairwise Jensen-Shannon divergence between circuits is 0.660 bits (base 2; range 0.471–0.851), increasing from 0.519 bits in 2003 to 0.660 bits in 2025 with no convergence event. Mutual information between community membership and immunity outcome is 0.155 bits (NMI = 0.119, permutation test p = 0.023, n = 10,000 permutations, seed 42). Platform win rate declined from 80% (2000–2004, n=5) to 38% (2020–2024, n=16), consistent with increasing judicial skepticism toward broad immunity claims. DAG hierarchical level correlates with PageRank at Spearman ρ = −0.838 among cases with at least one inbound citation (n=41, p < 0.0001); full corpus ρ = −0.964 is inflated by 22 dangling nodes sharing identical PageRank floor values. Fair Housing Council v. Roommates.com, LLC, 521 F.3d 1157 (9th Cir. 2008), holds the highest betweenness centrality in the corpus (BC = 0.021) while ranking fifth by PageRank (PR = 0.043), reflecting its structural role as a doctrinal bridge between early immunity expansion cases and later platform-liability litigation. All data and code are provided to support full reproduction of all analyses. See ZENODO_README.md for the complete file inventory and replication_guide.md for step-by-step reproduction instructions.本数据集配套提供针对《通信规范法案》第230条（47 U.S.C. § 230）开展计算法学研究的全部数据与代码。本数据集包含经过验证的判例语料库：涵盖1997年至2025年间实质适用第230条豁免法理的63篇已公开联邦巡回法院及最高法院判例；同时包含由这些判例构建的引用网络（含63个节点、320条有向边，经验证为有向无环图）、全部衍生网络度量指标，以及完整的分析流程管线。 该语料库的构建流程为：从CourtListener（courtlistener.com）检索得到211份候选判例，随后依据已公开的有效纳入标准与十类排除分类体系，对每份候选判例开展系统性人工评审。自动检索阶段的假阳性率达70.1%（(211−63)/211=148/211），该方法学发现对实证法学研究中的语料库构建具有重要参考价值。基于覆盖5个时间层的20个判例分层抽样结果显示，语料库的查全率为1.0（95%威尔逊置信区间：0.942–1.000）。 核心研究发现如下： 引用权威度集中于"Zeran v. America Online, Inc."、129 F.3d 327 (4th Cir. 1997)一案，该案的页面排名（PageRank）值为0.200（α=0.85），占网络总权威度的20.0%，且在1997年至2025年的每年度快照中均位列第一。莱顿算法（Leiden algorithm）共识别出7个社区（100次随机种子实验下的标准化互信息（NMI）平均稳定性为0.621，标准差为0.121）。各巡回法院间的平均两两詹森-香农散度为0.660比特（以2为底；取值范围0.471–0.851），从2003年的0.519比特持续升至2025年的0.660比特，未出现收敛现象。社区归属与豁免结果之间的互信息为0.155比特（标准化互信息NMI=0.119，置换检验p值=0.023，共开展10000次置换实验，随机种子为42）。平台胜诉率从2000–2004年的80%（n=5）降至2020–2024年的38%（n=16），这与司法机关对宽泛豁免主张的质疑态度日益增强的趋势相符。在至少拥有1条入向引用的判例（n=41，p<0.0001）中，有向无环图（DAG）层级与页面排名（PageRank）的斯皮尔曼秩相关系数ρ为-0.838；全语料库的相关系数ρ为-0.964，该结果因22个悬挂节点共享相同的页面排名下限值而被高估。"Fair Housing Council v. Roommates.com, LLC"、521 F.3d 1157 (9th Cir. 2008)一案在语料库中拥有最高的介数中心性（Betweenness Centrality, BC=0.021），但按页面排名仅位列第五（PR=0.043），这体现了其作为早期豁免扩张判例与后期平台责任诉讼之间的法理桥梁的结构性作用。 本数据集提供全部数据与代码，以支持所有分析的完整复现。完整文件清单请参阅ZENODO_README.md，分步复现指南请参阅replication_guide.md。