摘要
随着数字音乐的蓬勃发展,为用户提供自动生成的连续播放列表已成为大型音乐平台的重要服务方式之一。传统的基于马尔可夫链的音乐推荐模型在音乐空间向量嵌入的过程中忽略了很多有价值的信息,比如音乐的特征值、用户的偏好值等。考虑到用户在某一时间段或某一情景会话听同一种类型的音乐的可能性极大,故在马尔可夫链的音乐推荐模型的基础上考虑了音乐特征值对音乐嵌入位置的影响,提出了一种基于特征值嵌入的音乐播放列表推荐模型。该模型根据用户真实的历史播放列表进行训练,生成最符合用户当前会话场景的音乐播放列表。通过实验验证了基于特征值嵌入的音乐播放列表推荐模型相较于传统的基于马尔可夫链的音乐推荐模型所推荐的音乐具有更高的相似性,即更符合用户当前场景需要。
With the rapid development of digital music,providing users with automatically generated continuous playlist has become one of the important services for large music platforms.The traditional Markov chain-based music recommendation model ignores a lot of valuable information in the process of music space vector embedding,such as music feature value,user preference value and so on.Considering that users are likely to listen to the same type of music in a certain period of time or in a certain situation,we consider the influence of music feature value on the music embedding position based on the music recommendation model of Markov chain and propose a music playlist recommendation model based on feature value embedding.The model is trained according to the user’s real historical playlist to generate a music playlist that best matches the user’s current session scene.Finally,it is verified that the music playlist recommendation model based on feature value embedding has higher similarity than the music recommended by the traditional Markov chain-based music recommendation model,which is more in line with the user’s current scene needs.
作者
何丽
于洋
HE Li;YU Yang(School of Computer,North China University of Technology,Beijing 100144,China)
出处
《计算机技术与发展》
2019年第11期144-148,共5页
Computer Technology and Development
基金
国家自然科学基金(61371143)
北京市教委科研计划面上项目(KM201510009008)
关键词
音乐播放列表
音乐特征值
推荐系统
空间嵌入模型
music playlist
music feature value
recommendation system
space embedding model
作者简介
何丽(1977-),女,硕导,研究方向为数据库;于洋(1993-),女,研究生,研究方向为推荐系统。
