面向推荐系统的音乐内涵空间建模研究

Modeling of Music Connotative Space for Recommender System

针对现有音乐推荐系统中音乐情感分析方法很难满足用户情感需求的问题,提出音乐内涵空间方法分析音乐情感。该方法选取音乐力度、速度、音强等表现要素,基于两极尺度的语义,构建音乐内涵空间。通过音乐内涵空间表示音乐情感,减少了音频特征客观水平与主观情感范围之间的差距,解决了情感标签标注音乐中由于用户情感经历不同造成的对音乐情感标注的主观差异性问题。在10 672条音乐评价数据集上,采用Kendall’s tau距离进行有效性验证,与基于情感标签推荐相比,实验结果表明,基于音乐内涵空间推荐音乐能较好地满足用户的情感需求。

Focusing on the music affective, a concept of music connotative space is introduced. The semantic polar scale is examined and the connotative space is constructed by musical strength, speed and intensity for music affective recommendation. The music connotative space bridges the gap between the objective levels of audio frequency characters and subject emotional rang inequality, and deals with the difference of music emotional label due to different listener experiences. According to 10672 music records, the effectiveness of the proposed music connotative space is validated in terms of Kendall＇s tau distance. The music connotation space outperforms emotion label in music classification for music affective recommendation.