vancenceho/youtube-features-clean

---
license: cdla-sharing-1.0
language:
- en
tags:
- music
- code
pretty_name: YouTube Audio Features Cleaned
size_categories:
- 10K<n<100K
---

# YouTube Audio Features Cleaned

Tabular **YouTube-side features** for tracks that have been matched to **Spotify** records in the *viral-content-predictor* pipeline. Rows are keyed for alignment with Spotify / audio / lyrics tables; columns combine metadata and numeric signals used for engagement or popularity modeling (exact schema depends on the exporting notebook revision).

## File

| File | Role |
|------|------|
| `youtube_features_cleaned.csv` | One row per matched track; cleaned dtypes and selected columns for baseline / ensemble notebooks. |

## Source

- **Upstream:** YouTube metadata and/or engagement fields collected after resolving **YouTube ↔ Spotify** matches (see notebooks under `notebooks/` for matching, downloads, and feature assembly).
- **Output location:** `data/processed/` — consumed e.g. by `02e_youtube_model_baseline.ipynb` and `explore_youtube_engagement_ensemble.ipynb` as `../data/processed/youtube_features_cleaned.csv`.

## Schema (typical)

Expect on the order of **tens to low hundreds of columns**, including:

- Identifiers and join keys (e.g. `track_id` or aligned IDs) to link to Spotify and other project tables.
- **YouTube**-derived fields (titles, IDs, timestamps, engagement statistics where present).
- Engineered or cleaned numeric/categorical features ready for `scikit-learn` pipelines.

Exact names and counts change with pipeline versions; inspect with:

```python
import pandas as pd
df = pd.read_csv("youtube_features_cleaned.csv", nrows=5)
print(df.columns.tolist())
```

## Modeling notes

- Baseline notebooks may **drop engagement targets** from `X` to avoid leakage when predicting virality from other features — follow the notebook you use.
- Train/validation splits should respect any **group** or **time** structure your project defines.

## Usage

```python
import pandas as pd

df = pd.read_csv("youtube_features_cleaned.csv")
```

## Limitations

- **Snapshot:** reflects the crawl / export date, not live YouTube counts.
- **Coverage:** only tracks with a successful match and feature row in the pipeline.
- **License:** comply with **CDLA-Sharing-1.0** (this card), YouTube’s terms, and third-party dataset licenses you combined to build the file.

## Citation

Cite this repository and the YouTube / Spotify data sources you used, plus the notebook hash or release tag that produced the CSV.

许可证：CDLA-Sharing-1.0
语言：英语
标签：音乐、代码
展示名称：清洗后的YouTube音频特征
规模类别：1万至10万条样本

# 清洗后的YouTube音频特征

本数据集为**YouTube侧特征（YouTube-side features）**的表格型数据，对应在*病毒式内容预测器*（viral-content-predictor）流程中与**Spotify**记录完成匹配的曲目。数据行通过键值对齐Spotify、音频及歌词数据表；列则整合了用于互动度或流行度建模的元数据与数值特征，具体表结构取决于导出笔记本的修订版本。

## 文件

| 文件 | 用途 |
|------|------|
| `youtube_features_cleaned.csv` | 每条匹配曲目对应一行数据；已完成数据类型清洗并筛选出适用于基线模型/集成模型笔记本的列。 |

## 数据源

- **上游数据源**：在完成**YouTube ↔ Spotify**匹配后收集的YouTube元数据及/或互动度字段，有关匹配、下载与特征组装的流程，请参阅`notebooks/`目录下的Jupyter笔记本。
- **输出路径**：存储于`data/processed/`目录，例如`02e_youtube_model_baseline.ipynb`与`explore_youtube_engagement_ensemble.ipynb`会通过`../data/processed/youtube_features_cleaned.csv`读取该文件。

## 典型表结构

典型情况下，数据集包含数十至数百个列，具体包括：
- 用于关联Spotify及其他项目数据表的标识符与连接键（例如`track_id`或对齐后的ID）；
- **YouTube**衍生字段，包括标题、ID、时间戳以及可用的互动度统计数据；
- 已完成特征工程或数据清洗、可直接用于`scikit-learn`流程的数值/分类特征。

列的具体名称与数量会随流程版本变更，可通过以下代码查看：
python
import pandas as pd
df = pd.read_csv("youtube_features_cleaned.csv", nrows=5)
print(df.columns.tolist())

## 建模注意事项

- 基线模型笔记本可能会从特征矩阵`X`中移除互动度目标变量，以避免在通过其他特征预测病毒式传播性时出现数据泄露问题，请遵循你所使用的笔记本的流程。
- 训练集/验证集的划分需遵循你项目中定义的任何分组或时间结构。

## 使用方法

python
import pandas as pd

df = pd.read_csv("youtube_features_cleaned.csv")

## 局限性

- **快照特性**：数据集反映的是爬虫/导出当日的统计数据，而非YouTube实时平台数据；
- **覆盖范围**：仅包含流程中完成成功匹配并生成特征行的曲目；
- **许可证要求**：需遵守**CDLA-Sharing-1.0**（本数据集卡片）、YouTube服务条款以及你用于构建该文件的第三方数据集许可证。

## 引用规范

引用时需注明本仓库、你所使用的YouTube/Spotify数据源，以及生成该CSV文件的笔记本哈希值或发布标签。