Culture-Aware Music Recommendation Dataset

<strong>LFM-1b dataset extended by acoustic track features and cultural cues describing users</strong> This dataset is based on the LFM-1b dataset (cf. http://www.cp.jku.at/datasets/LFM-1b/), however, adds acoustic features describing the tracks to the original dataset as well as cultural aspects describing users (taken from Hofstede's six dimension model and the World Happiness Report) on the country-level. For the creation of the dataset, we extract all users for which the original dataset contains country information for. We extract the listening events of these users and match the tracks against the Spotify API to subsequently retrieve the acoustic features of these tracks (cf. [Spotify Audio Feature Description](https://developer.spotify.com/documentation/web-api/reference/object-model/#audio-features-object)). The final dataset contains only events of users with country information and tracks with acoustic features, which can be matched with the country-level data of the World Happiness Report and Hofstede's cultural dimensions to add cultural and socio-economic aspects for users. This new dataset contains 55,190 users 3,471,884 tracks including acoustic features 351,469,333 listening events of those users for tracks we have obtained acoustic features for Hofstede's cultural dimensions for 47 countries World Happiness Report (WHR) data for 164 countries <strong>Files</strong><br> All files are tab-separated, with no quoting of strings. The dataset contains the following files, whose content we describe in more detail in the following parts. * acoustic_features_lfm_id.tsv: acoustic features for all tracks in the dataset, identified by their LFM track identifier<br> * events.tsv: listening events for all users<br> * hofstede.tsv: Hofstede's cultural dimensions<br> * users.tsv: user metadata<br> * world_happiness_report_2018.tsv: World Happiness Report data For further information on the contents of these files, please cf. the Readme file. Please cite the following paper when using the dataset:<br> Zangerle, E., Pichl, M. and Schedl, M., 2020. User Models for Culture-Aware Music Recommendation: Fusing Acoustic and Cultural Cues. <em>Transactions of the International Society for Music Information Retrieval</em>, 3(1), pp.1–16. DOI: http://doi.org/10.5334/tismir.37

**基于音频特征与用户文化线索扩展的LFM-1b数据集**
本数据集以LFM-1b数据集为基础（参见http://www.cp.jku.at/datasets/LFM-1b/），较原始数据集新增了描述音乐曲目的音频特征，以及基于国家层面的用户文化属性数据——这些数据取自霍夫斯泰德文化六维度模型（Hofstede's Six Dimension Model）与《世界幸福报告》（World Happiness Report）。
在数据集构建过程中，我们从原始数据集中提取所有带有国家信息的用户，提取这些用户的听歌行为记录，并将曲目与Spotify API进行匹配，进而获取对应曲目的音频特征（参见[Spotify音频特征说明](https://developer.spotify.com/documentation/web-api/reference/object-model/#audio-features-object)）。
最终数据集仅保留带有国家信息的用户的听歌记录，以及带有音频特征的曲目；通过将这些数据与《世界幸福报告》及霍夫斯泰德文化维度的国家级数据进行匹配，可为用户补充文化与社会经济维度的属性信息。本数据集共涵盖55,190名用户、3,471,884条带音频特征的曲目，以及对应这些曲目的351,469,333条用户听歌记录；其中，霍夫斯泰德文化维度数据覆盖47个国家，《世界幸福报告》（WHR）数据覆盖164个国家。
<strong>文件</strong><br>
所有文件均采用制表符分隔格式，且不对字符串进行引号包裹。本数据集包含以下文件，各文件的详细内容将在后续章节中说明。
* acoustic_features_lfm_id.tsv: 数据集内所有曲目对应的音频特征，以LFM曲目标识符作为唯一标识<br>
* events.tsv: 所有用户的听歌行为记录<br>
* hofstede.tsv: 霍夫斯泰德文化维度数据<br>
* users.tsv: 用户元数据<br>
* world_happiness_report_2018.tsv: 《世界幸福报告》相关数据<br>
若需了解各文件的详细内容，请参阅自述文件（Readme）。
使用本数据集时，请引用以下论文：<br>
Zangerle, E., Pichl, M. 与 Schedl, M., 2020. 《面向文化感知音乐推荐的用户模型：融合音频与文化线索》（<em>User Models for Culture-Aware Music Recommendation: Fusing Acoustic and Cultural Cues</em>）。<em>《国际音乐信息检索学会汇刊》</em>, 3(1), 第1–16页。DOI: http://doi.org/10.5334/tismir.37