One Comment from One Perspective: An Effective Strategy for Enhancing Automatic Music Comment

Proceedings of the 28th International Conference on Computational Linguistics, pages 2889–2899 Barcelona, Spain (Online), December 8-13, 2020 2889 One Comment from One Perspective: An Effective Strategy for Enhancing Automatic Music Comment Tengfei Huo†∗, Zhiqiang Liu‡, Jinchao Zhang‡ and Jie Zhou‡ †CAS Key Lab of Network Data Science and Technology †Institute of Computing Technology, Chinese Academy of Sciences †University of Chinese Academy of Sciences, Beijing, China ‡Pattern Recognition Center, WeChat AI, Tencent Inc, China huotengfei19s@ict.ac.cn {zhiqliu,dayerzhang,withtomzhou}@tencent.com Abstract The automatic generation of music comments is of great significance for increasing the popu- larity of music and the music platform’s activity. In human music comments, there exists high distinction and diverse perspectives for the same song. In other words, for a song, different com- ments stem from different musical perspectives. However, to date, this characteristic has not been considered well in research on automatic comment generation. The existing methods tend to generate common and meaningless comments. In this paper, we propose an effective multi- perspective strategy to enhance the diversity of the generated comments. The experiment results on two music comment datasets show that our proposed model can effectively generate a series of diverse music comments based on different perspectives, which outperforms state-of-the-art baselines by a substantial margin.1 1 Introduction In recent years, neural networks have achieved great success in natural language generation (NLG), which can be applied in many real-world scenarios, such as poetry generation, dialogue generation, com- ment generation, and so on. Music comment generation is a sub-task of NLG. High-quality comments can effectively increase the popularity of music and the activity of the platform (Zeng et al., 2019). Title:�⼀剪梅�(A�Spray�of�Plum�Blossoms)������������������������������������Singer:�费⽟清�(Fei�Yuqing)��� Lyrics:�雪花飘飘北风萧萧，天地⼀⽚苍茫，⼀剪寒梅傲⽴雪中，只为伊⼈飘⾹，爱我所爱⽆怨⽆悔... (Snow�petals�drifting,�the�north�wind�whistles.�The�world�ever�a�boundless.�A�spray�of�winter�Plum�Blossoms.�Stands�proudly�in�the�snow. Only�for�that�person�its�fragrance�drift.�My�love�is�without�complains�and�regrets...) 已单曲循环雪花飘飘⼀百遍 这⾸歌的声⾳真是扣⼈⼼弦 ⼀剪梅改了这么多版，还是这版耐听 ⽂能⼀剪梅�武能嘿嘿嘿 我喜欢这⾸歌 这是我喜欢的歌 这⾸歌是我最喜欢的⼀⾸ 好喜欢这⾸歌啊 A�single�cycle�of�snowflakes�fluttering�a�hundred�times The�sound�of�this�song�is�really�fascinating This�song�has�changed�many�versions.�It�is�still�this�version He�can�sing�gentle�songs�as�well�as�interesting�songs I�like�this�song This�is�my�favorite�song This�song�is�my�favorite I�really�like�this�song Human�comments Generated�comments Figure 1: Examples of human comments and generated comments. However, there is a gap between human comments and generated comments based on the existing model. The music comment generation task’s existing models tend to generate general but meaningless comments, such as “It is nice”. As an example, an excerpt from human music comments and automat- ically generated comments based on the seq2seq model is given in Figure 1. Obviously, these human comments on the left are more attractive and diverse, benefiting from the creation with multiple musical perspectives: the listener’s behavior, song melody, version of music, and style of lyrics. In human music ∗This work was done when Tengfei Huo was interning at Pattern Recognition Center, WeChat AI, Tencent Inc, China. 1The datasets and code are available at https://github.com/htfhxx/CommentPerspective This work is licensed under a Creative Commons Attribution 4.0 International License. License details: http:// creativecommons.org/licenses/by/4.0/. 2890 comments, there exists excellent distinction and diverse perspectives for the same music content. There goes a saying that there are a thousand Hamlets in a thousand people’s eyes. As a result, the music comment generation is typically regarded as a one-to-many generation task. Nowadays, researchers have noticed such problems and tried to solve them via multiple methods in dialogue generation. Some of them have …

Content truncated to 500 words