Song Cold Start Recommendation Based on Tags

This article introduces a tag-based song cold start recommendation method. The method is simple and effective, and is an important part of the NetEase Cloud Music song recommendation system. It is an important cornerstone for building a healthy music distribution ecosystem, providing the first step of assistance for the growth of many cold start new songs.

The article first explains the concept of cold start songs, which refers to songs that have not been recommended to users and have not been evaluated by users. These songs are in a cold state and need to be recommended to users to generate initial data. The article then describes the challenges of cold start songs, including the lack of user data, the difficulty of evaluating song quality, and the need to balance the distribution of cold start songs with the user experience of the existing recommendation system.

The article then introduces the tag-based cold start recommendation method. The method uses song tags to calculate the similarity between songs and recommends similar songs to users. The article describes the process of generating song tags, calculating song similarity, and recommending songs to users. It also discusses the challenges of tag-based recommendation, such as the need for high-quality tags and the difficulty of handling new tags.

The article then describes the business effect of the cold start recommendation system. It measures the success of the system by the exposure coverage rate of cold start songs, which is the percentage of users who have been exposed to cold start songs among all users who have been exposed to songs. The article also discusses the need to balance the distribution of cold start songs with the user experience of the existing recommendation system.

The article concludes by summarizing the tag-based cold start recommendation method and its importance in the NetEase Cloud Music song recommendation system. It also mentions that the article is part of a series on cold start solutions and that more methods and practices will be shared in the future.