摇滚中国图志：方法论

Rock China Atlas: Methodology

中文

项目定位

摇滚中国图志（Rock China Atlas）是一个基于地理信息系统的中国摇滚音乐数字档案馆，旨在通过空间可视化手段，呈现中国摇滚音乐的地理分布格局、人际协作网络、时间演化轨迹与风格流派生态。

项目面向音乐研究者、文化地理学者、数字人文工作者及公众，提供可交互的探索式界面，将离散的乐队信息转化为可感知的空间叙事。

一、数据采集与构建

1.1 数据源

核心数据来源于 yaogun.com（摇滚）——目前收录最全面的中国摇滚乐队日语维基数据库。该数据库由日本音乐研究者长期维护，涵盖中国大陆摇滚乐队的基础信息。

数据规模: 45,000+ 乐队，覆盖全国 31 个省级行政区。

1.2 数据清洗

• 地名标准化: 日语地名→中文行政区划，建立 248 条日中城市名映射表（JP_TO_CN）和 371 条中英映射表（CN_TO_EN）
• 成员名规范化: 日文汉字→简体中文转换（JA_TO_ZH 映射表），解决同一人在不同乐队中的姓名变体问题
• 风格标签清洗: Wikipedia Infobox 原始标签存在大量解析错误（混入 |、occupation 等非风格字段），通过正则过滤和人工校验清理；建立 482 条日语→中文风格翻译表
• 地理编码: 基于行政区划数据将地名转换为经纬度坐标

1.3 知名度指标

采用维基百科页面字数（weight）作为乐队知名度的代理变量。该指标的假设基础是：知名度更高的乐队通常拥有更详尽的维基百科条目。尽管存在编辑偏差，但在大样本下提供了可用的排序依据。

二、关系网络分析

2.1 边的定义

两个乐队之间存在"边"（关系），当且仅当它们共享至少一名成员（曾经或现在）。这反映了中国摇滚圈的一个重要特征：音乐人频繁在不同乐队之间流动，形成紧密的协作网络。

2.2 图计算

使用 Graphology 图计算库，在 Web Worker 中执行以下分析：

• 度数（Degree）: 每个乐队与其他乐队的连接数量，反映其在网络中的枢纽地位
• Louvain 社区检测: 自动识别紧密协作的乐队社群（场景），揭示地域性或风格性的音乐圈子
• 连通分量: 识别孤立乐队与连通子网络

2.3 可视化

成员关系以三维弧线呈现（Three.js），弧线高度与距离成正比。悬停某成员时，其参与的所有乐队间自动连线并高亮，非相关乐队淡化处理。

三、风格流派分类体系

3.1 分类架构

采用两级分类体系：9 个大类 → 217 个子分类。

3.2 配色系统：国画颜料色谱

为避免西方音乐可视化中常见的任意配色方案，本项目建立了一套基于中国传统国画颜料的语义化色彩映射系统，将音乐风格的感知特征与色彩心理学关联。

设计原则：色相即语义——红色系对应攻击性风格，绿色系对应有机自然，蓝紫系对应沉思与复杂。

19 个国画颜料色构成锚点色带，9 个大类各对应一个锚点，217 个子分类在所属大类的色相邻域内进行语义微偏移（色相 ±8°、饱和度 ±8%、明度 ±8%），形成"点彩"效果——既保证大类可辨识，又让子分类在视觉上有所区分。

3.3 未分类处理

约 40% 的乐队缺少有效风格标签。这些乐队以黑色小点表示（#1a1a1a，半径 2px，透明度 0.7），在视觉上退居背景，避免对有标签乐队的色彩叙事造成干扰。

四、城市音乐气质模型

4.1 研究问题

不同城市是否存在可量化的音乐风格偏好？如何衡量一个城市在某种音乐风格上的"特色程度"？

4.2 区位商模型（Location Quotient）

借鉴经济地理学中的区位商概念，定义城市 c 在风格 g 上的音乐气质指数：

LQ(c, g) = (该城市风格 g 的加权乐队数 / 该城市全部加权乐队数)
          ÷ (全国风格 g 的加权乐队数 / 全国全部加权乐队数)

• LQ > 1: 该城市在此风格上具有相对集聚优势
• LQ = 1: 与全国平均水平持平
• LQ < 1: 该城市在此风格上相对不足

4.3 方法改进

知名度加权: 单纯计数会导致拥有大量小乐队的城市（如北京）在所有风格上都趋近全国均值。引入知名度分档权重后，知名乐队对城市气质的贡献更大，更符合直觉认知。

Leave-one-out 修正: 计算某城市的 LQ 时，从全国基线中排除该城市自身的数据。这解决了北京、上海等大城市因体量过大导致 LQ 始终趋近 1 的问题，使其真正的风格特色得以显现。

子分类粒度: 在 217 个子分类级别而非 9 个大类级别计算 LQ，保留更细致的风格差异。

4.4 可视化呈现

• 省级（zoom 0–5）: 省级聚合气质点，展示各省最具特色的风格
• 市级（zoom 5–8）: 城市气质点，弹窗展示 Top 子分类及其 LQ 偏离度
• 乐队级（zoom 8+）: 回到单个乐队点

气质弹窗中，每个子分类显示其 LQ 偏离度（如 +320%），并可展开查看该城市该风格下的代表乐队（附知名度数值）。

五、多语言支持

项目支持中文、英语、日语三种界面语言，覆盖：

• 乐队名（原名/中文/拼音）
• 城市名（248 条日中映射 + 371 条中英映射）
• 风格标签（217 子分类 + 9 大类的三语翻译）
• 界面文案（按钮、弹窗、说明文字）

六、外部资源整合

七、局限性与已知偏差

1. 数据源单一: 核心数据依赖 yaogun.com 单一来源，该数据库由日本研究者维护，可能存在对日语圈更熟悉的乐队的收录偏向
2. 风格标签噪声: 原始 Wikipedia Infobox 标签存在解析错误和分类不一致，虽经清洗仍有残余噪声
3. 地理精度: 乐队活动地点精确到区级，部分仅到市级或省级；流动性强的乐队仅记录主要活动城市
4. 时间维度: 仅记录成立年份，缺少解散/暂停/重组等生命周期事件
5. 知名度代理: 维基百科页面字数作为知名度指标存在编辑者偏差，且不反映实际演出频次或唱片销量
6. 地下场景盲区: 数据库更偏向有一定网络痕迹的乐队，纯粹的地下/DIY 场景覆盖不足

八、引用与致谢

本项目数据来源于公开平台，用于非商业性的学术研究与文化传播目的。

如需引用本项目：

摇滚中国图志 / Rock China Atlas. (2025). https://rockchinaatlas.com

致谢：

• yaogun.com —— 数据基础
• MapLibre / MapTiler —— 地图引擎与底图
• Apple Music —— 音乐试听
• 豆瓣 —— 乐队主页链接

English

Project Overview

Rock China Atlas is a GIS-based digital archive of China's rock music, designed to reveal the geographic distribution, interpersonal collaboration networks, temporal evolution, and genre ecology of the Chinese rock scene through interactive spatial visualization.

The project serves music researchers, cultural geographers, digital humanities scholars, and the general public, transforming discrete band information into perceivable spatial narratives.

1. Data Collection and Construction

1.1 Data Source

Core data is sourced from yaogun.com — the most comprehensive Japanese-language wiki database of Chinese rock bands, maintained by Japanese music researchers over many years.

Scale: 45,000+ bands across all 31 provincial-level regions of mainland China.

1.2 Data Cleaning

• Place name normalization: Japanese → Chinese administrative divisions (248 JP→CN city mappings + 371 CN→EN mappings)
• Member name normalization: Japanese kanji → Simplified Chinese conversion to resolve name variants across bands
• Genre tag cleaning: Raw Wikipedia Infobox tags contain parsing errors (mixed with |, occupation, and other non-genre fields); cleaned via regex filtering and manual review; 482 Japanese→Chinese genre translations established
• Geocoding: Administrative division names converted to latitude/longitude coordinates

1.3 Notability Metric

Wikipedia page length (weight) serves as a proxy variable for band notability. While editorial bias exists, this metric provides a usable ranking basis across the large sample.

2. Network Analysis

2.1 Edge Definition

An edge exists between two bands if and only if they share at least one member (past or present). This reflects a key characteristic of China's rock scene: musicians frequently move between bands, forming dense collaboration networks.

2.2 Graph Computation

Using the Graphology library in a Web Worker:

• Degree: Number of connections per band, indicating hub status
• Louvain community detection: Automatic identification of tightly collaborating band communities (scenes)
• Connected components: Identifying isolated bands and connected subnetworks

2.3 Visualization

Member relationships are rendered as 3D arcs (Three.js), with arc height proportional to geographic distance. Hovering over a member highlights all their bands with connecting arcs while fading unrelated bands.

3. Genre Classification System

3.1 Taxonomy

A two-tier system: 9 families → 217 sub-genres.

3.2 Color System: Chinese Ink-Painting Pigment Spectrum

To avoid arbitrary color schemes common in Western music visualization, this project establishes a semantic color mapping system based on traditional Chinese painting pigments, linking perceived characteristics of musical styles with color psychology.

Design principle: hue as semantics — reds for aggressive styles, greens for organic and natural, blue-purples for contemplative and complex.

19 traditional pigment colors form anchor points. Each of the 9 families maps to one anchor, and 217 sub-genres undergo semantic micro-displacement within their family's hue neighborhood (hue ±8°, saturation ±8%, lightness ±8%), creating a "pointillist" effect — maintaining family-level recognizability while visually distinguishing sub-genres.

4. City Music Character Model

4.1 Research Question

Do different cities exhibit quantifiable musical style preferences? How can we measure a city's degree of specialization in a given genre?

4.2 Location Quotient (LQ) Model

Borrowing from economic geography, we define a city c's music character index for genre g:

LQ(c, g) = (City c's weighted band count in genre g / City c's total weighted band count)
          ÷ (National weighted band count in genre g / National total weighted band count)

• LQ > 1: The city has a relative concentration advantage in this genre
• LQ = 1: On par with the national average
• LQ < 1: The city is relatively underrepresented in this genre

4.3 Methodological Improvements

Notability weighting: Simple counting causes cities with many small bands (e.g., Beijing) to converge toward the national mean across all genres. Tiered notability weighting ensures that prominent bands contribute more to a city's character profile.

Leave-one-out correction: When computing a city's LQ, the city's own data is excluded from the national baseline. This addresses the problem of large cities like Beijing and Shanghai always having LQ ≈ 1 due to their sheer volume, allowing their true style specializations to emerge.

Sub-genre granularity: LQ is computed at the 217 sub-genre level rather than the 9 family level, preserving finer stylistic distinctions.

4.4 Visualization

• Provincial (zoom 0–5): Province-level aggregated character points showing top distinctive genres
• City (zoom 5–8): City character points with popups showing top sub-genres and LQ deviation
• Band (zoom 8+): Individual band points

Character popups display each sub-genre's LQ deviation (e.g., +320%) and can expand to show representative bands with their notability scores.

5. Multilingual Support

The project supports Chinese, English, and Japanese across:

• Band names (original / Chinese / Pinyin)
• City names (248 JP→CN + 371 CN→EN mappings)
• Genre labels (217 sub-genres + 9 families, trilingual)
• UI text (buttons, popups, descriptions)

6. External Resource Integration

7. Limitations and Known Biases

1. Single data source: Core data relies on yaogun.com alone, maintained by Japanese researchers, potentially biasing toward bands more visible in Japanese-language circles
2. Genre tag noise: Raw Wikipedia Infobox tags contain parsing errors and classification inconsistencies; residual noise remains despite cleaning
3. Geographic precision: Band locations are accurate to the district level at best; some are only city- or province-level; mobile bands are recorded at their primary city only
4. Temporal dimension: Only founding year is recorded; dissolution, hiatus, and reunion events are absent
5. Notability proxy: Wikipedia page length as a notability indicator carries editorial bias and does not reflect actual performance frequency or record sales
6. Underground blind spot: The database skews toward bands with some online presence; purely underground/DIY scenes are underrepresented

8. Citation and Acknowledgments

Data sourced from public platforms for non-commercial academic research and cultural dissemination purposes.

To cite this project:

Rock China Atlas. (2025). https://rockchinaatlas.com

Acknowledgments:

• yaogun.com — Data foundation
• MapLibre / MapTiler — Map engine and tiles
• Apple Music — Music previews
• Douban — Band page links