Proposal of Context-Aware Music Recommender System Using Negative Sampling
Proposal of Context-Aware Music Recommender System Using Negative Sampling
复制标题
DOI：
10.1007/978-3-030-39878-1_11
复制
发表时间：
2019-06
期刊：
影响因子：
--
通讯作者：
Jin-cheng Zhang;Y. Takama
中科院分区：
文献类型：
--
作者：
Jin-cheng Zhang;Y. Takama
这是对JSAI2019中选定论文的扩展。本文提出了一种考虑听者语境信息的音乐项目推荐方法。最近，随着Spotify等在线音乐服务的发展，用户可以随时随地轻松地欣赏音乐。然而，我们很难从庞大的资源中找到合适的音乐项目。另一方面，由于音乐项目的聆听风格和特点，音乐项目通常没有明确的评级。因此，游戏次数等隐式反馈被广泛用于构建推荐系统。作为附加信息，本文考虑了听者的语境。提出的方法采用FMs (Factorization Machines)，其中上下文信息被视为因素。负采样用于减少负采样（用户尚未听过的音乐项目）的数量。通过离线实验验证了该方法的有效性和负采样的效果。在nowplaying-rs数据集上的实验结果表明，该方法优于加权交替最小二乘方法。此外，还研究了不同的负抽样方法，如基于人气的抽样和不同时间窗大小的抽样。
This is an extension from a selected paper from JSAI2019. This paper proposes a method for recommending music items considering listeners’ context information. Recently, users can enjoy music easily regardless of time and a place due to evolution of online music services such as Spotify. However, it is difficult for us to find appropriate music items from enormous resources. On the other hand, because of listening style and characteristic of music items, music items do not usually have explicit rating. Therefore, implicit feedback such as playing count has been widely used to construct recommender systems. As additional information, this paper considers listeners’ context. The proposed method employs FMs (Factorization Machines), in which the context information is treated as factors. Negative sampling is applied to reduce the number of negative samples (music items a user has yet to be listened). The effectiveness of the proposed method and the effect of negative sampling are evaluated with an offline experiment. The experimental result on nowplaying-rs dataset shows that the proposed method outperforms wALS (weighted Alternating Least Squares) method. Furthermore, different negative sampling methods such as popularity-based one and sampling with different time window size are also investigated.