Early discovery time: "Song discovery time weighting", which weights the time of the user's collection of a certain song. In the growth cycle when a certain song reaches the highest level, the user who collects the song earlier has a higher score to show whatsapp Database the user's better appreciation ability Large upgrade range: Calculate the difference between the current song level (ground truth) and the level at the time of collection, and the difference in the number of playbacks, and find out the posterior popularity Lift corresponding to the collection behavior whatsapp Database NetEase Cloud Music: How to discover music connoisseurs through data, and then find treasured niche music When defining a music connoisseu.
It should be noted that people with strong appreciation ability in national style music may not be able to appreciate English Rap. Therefore, music connoisseurs also whatsapp Database need to be classified and divided according to music vertical categories, such as rock, R&B, Japanese, electronic, folk country, jazz classical, etc., to score each music connoisseur in this music vertical category. Result: Music connoisseurs successfully discovered many high-quality small songs. After these small songs were supported, the success rate of cold start (the success rate whatsapp Database of playing level transition) exceeded 40%, which was much higher than the 3% of the previous manual scoring . Optimized the platform ecology and the problem of insufficient long-tail content.
NetEase Cloud Music: How to discover music connoisseurs through data, and then find treasured niche music NetEase Cloud Music: How to discover whatsapp Database music connoisseurs through data, and then find treasured niche music The NetEase Cloud team also provided a chestnut that appraisers found niche songs: "Drink more hot water" has been in a state that only a few people have been listening to for a long time since its launch. Collection, the endorsement whatsapp Database of the song by the connoisseur is immediately recognized by the algorithm, and the flow is supported and distributed , and then the support is withdrawn to allow it to grow naturally.