vancenceho/spotify-lyrics-clean

---
license: cdla-sharing-1.0
language:
- en
tags:
- music
- code
pretty_name: Spotify Million Song Lyrics Cleaned
size_categories:
- 10K<n<100K
---

# Spotify Million Song Lyrics Cleaned

CSV of **deduplicated, normalized lyrics** aligned to the Million Song–style raw lyrics dump used in the *viral-content-predictor* project. Each row is one **(artist, song)** identity after normalization; lyrics are cleaned text suitable for TF‑IDF, retrieval, or joining to Spotify metadata via `artist_norm` / `title_norm`.

## File

| File | Role |
|------|------|
| `lyrics_cleaned.csv` | One row per normalized `(artist_norm, title_norm)`; includes raw display columns and lyric text. |

## Source

- **Upstream:** Raw lyrics table (e.g. `spotify_millsongdata.csv` in `data/raw/`), derived from the **Spotify Million Song Dataset**–style corpus.
- **Pipeline:** `notebooks/exploratory/explore_lyrics_clean.ipynb` — normalization, lyric text cleanup, **deduplication** (keeping the longest lyric per normalized artist–title pair where applicable).

## Schema (typical columns)

| Column | Description |
|--------|-------------|
| `artist` | Original artist string. |
| `song` | Original track title string. |
| `artist_norm` | Normalized artist key for joins. |
| `title_norm` | Normalized title key for joins. |
| `lyrics_char_len` | Character length of the lyric text (or similar size signal). |
| `lyrics` | Full lyric text (cleaned). |

Exact column order and any extra fields follow the exporting notebook revision.

## Usage

```python
import pandas as pd

lyr = pd.read_csv("lyrics_cleaned.csv")
# Join to Spotify-side tables on artist_norm / title_norm (and track_id after mapping)
```

## Size note

The **full** cleaned file can be **large** (multi‑million characters / many rows). If your snapshot is bigger than `10K<n<100K`, update the Hub **`size_categories`** in the YAML (e.g. `1M<n<10M` or `n>10M`) so the dataset card matches reality.

## Limitations

- **Language:** Primarily **English** lyrics in practice; not language-tagged per row in this export.
- **Rights:** Lyrics are third-party content; redistribution must respect **CDLA-Sharing-1.0** (this card), the original dataset license, and applicable copyright rules in your jurisdiction.

## Citation

Cite the original **Million Song Dataset** / lyrics source you used, plus your repository or notebook revision for the cleaning steps.

许可证: cdla-sharing-1.0
语言:
- 英语
标签:
- 音乐
- 代码
展示名称: 清理版Spotify百万歌曲歌词
规模分类:
- 10K<n<100K

# 清理版Spotify百万歌曲歌词

本数据集为经**去重（deduplicated）、归一化（normalized）**处理的歌词CSV文件，与*viral-content-predictor*项目中使用的百万歌曲风格原始歌词转储文件对齐。每一行代表归一化后的一组**（艺术家，歌曲）**标识；歌词为经清洗后的文本，适用于TF-IDF、检索任务，或通过`artist_norm`/`title_norm`字段与Spotify元数据进行关联。

## 文件

| 文件名 | 作用 |
|--------|------|
| `lyrics_cleaned.csv` | 以归一化后的`(artist_norm, title_norm)`为单位生成一行数据，包含原始展示列与歌词文本。 |

## 数据源

- **上游数据源**：原始歌词表（例如`data/raw/`目录下的`spotify_millsongdata.csv`），源自**Spotify百万歌曲数据集（Spotify Million Song Dataset）**风格的语料库。
- **处理流水线**：`notebooks/exploratory/explore_lyrics_clean.ipynb` — 包含归一化、歌词文本清洗、**去重（deduplication）**（针对每组归一化后的艺术家-歌曲对，保留最长的歌词内容，若适用）。

## 字段架构（典型列）

| 列名 | 说明 |
|--------|-------------|
| `artist` | 原始艺术家名称字符串。 |
| `song` | 原始歌曲标题字符串。 |
| `artist_norm` | 用于关联操作的归一化艺术家键。 |
| `title_norm` | 用于关联操作的归一化歌曲标题键。 |
| `lyrics_char_len` | 歌词文本的字符长度（或类似的规模指标）。 |
| `lyrics` | 完整的清洗后歌词文本。 |

列的精确顺序与额外字段均遵循导出所用Notebook的版本规范。

## 使用示例

python
import pandas as pd

lyr = pd.read_csv("lyrics_cleaned.csv")
# 通过`artist_norm`/`title_norm`（映射后可额外关联track_id）与Spotify侧数据表进行关联

## 规模说明

**完整**的清洗后文件体积可能较大（字符数达数百万/行数众多）。若您获取的数据集快照规模超出`10K<n<100K`范围，请更新Hugging Face Hub端**`size_categories`**字段（例如`1M<n<10M`或`n>10M`），以确保数据集卡片与实际规模一致。

## 限制说明

- **语言限制**：实际以**英语（English）**歌词为主；本次导出未对每一行歌词进行语言标注。
- **版权限制**：歌词属于第三方内容；重新分发时需遵守**CDLA-Sharing-1.0**（本数据集卡片）、原始数据集许可证以及您所在司法辖区的适用版权法规。

## 引用要求

请引用您所使用的原始**百万歌曲数据集（Million Song Dataset）**/歌词数据源，以及您用于清洗步骤的代码仓库或Notebook版本。