编程知识 cdmana.com

Tiktok, QQ music, NetEase cloud music, the algorithm is about songs recommended, rather than user preferences?

 Tiktok 、QQ music 、 Netease cloud music , Algorithm left and right song recommendation , It's not user preference ?


picture source @ Vision China

writing | melt , author | not attend personally

A normal weekday night , Shtiktok opened the jitter as usual. , I'm going to relax . After a few minutes , Because the background music of the video is popular “ Divine Comedy ”, Xiao Jia, who was more and more annoyed by the brush, soon turned it off APP.

Go to the front if there's a lot of traffic


This is not the first time that Xiaojia has been successfully dissuaded by the repeated brainwashing music . As time goes on , Xiao Jia gradually found out The homogeneity of the tiktok and video music is getting higher and higher. , Even several colleagues and friends who would have carefully selected some minority songs as background music , It has also been used recently “ Platform boom ” As a way to record a movie BGM.

“ For example, I like to watch short videos of movie recommendation , In the past, the soundtrack of these videos was quite diverse , Now it's very consistent , They're all brainwashing melodies with similar rhythms , It's a real headache to hear too much .” Xiao Jia is a little helpless .

From the grain of grain 、 wolves disco To the moon 、 Stepping on mountains and rivers , These repeated soundtracks forced Xiao Jia to repeat his single , The view of the song itself is also from the beginning of no feeling, and then a very tired of hearing the clip , Even affected Xiao Jia's interest in watching the video .

What's troubling Xiaojia “ Tiktok ” Used from the tiktok platform. Information flow funnel algorithm . Tiktok uploads video and checks through the platform. , The system will cold start this video , Assign an inclusion 200-1000 Initial pool of online user traffic .

The platform will be based on this 1000 The completion rate of a single exposure 、 give the thumbs-up 、 Focus on 、 Comment on 、 forward 、 Transfer powder 、 Travel depth and other data , Combined with the user account score for data analysis , Decide whether to give weighting or not . among , give the thumbs-up 、 Comment on 、 forward 、 Click through rate 、 The completion rate and other indicators decide whether to carry out the second round of traffic recommendation and recommendation .

The shaking platform will make the short video with tiktok more weighted. , Different videos will be pushed according to the tags of user groups , Make content distribution more accurate . similar “ Guess you like ” How to mark , According to the tags of the video itself and the user tags , Match between the two 、 Decide whether to push or not .

The whole process of video recommendation from tiktok algorithm is visible. , Video content and tags are very important to the amount of playback . The recommendation algorithm based on traffic intensifies the difficulty of account “ Matthew effect ”, Expand recommendations for popular videos and music , Make them more accessible .

A lot of anchors want to increase fans and spread the video , With “ Rub hot spots ” Direct imitation 、 Use your own flow and label 、 Popular content and background music with wide audience coverage . In this way , Popular music is more popular after being quoted many times ,“ Tiktok ” And then came , Similar rhythm and melody tiktok the whole shaking platform. . It's bothering people like Xiaojia who are sensitive to background music and “ Watch the taste ” More diverse users .

“ It's really annoying , It's always those songs that repeat . Short videos are not like music APP The song name will be displayed , You don't turn on the listening , I don't know it's all the same song ; It's marked by the algorithm as soon as it's opened , And recommended similar videos to me , It's a vicious circle .” Xiao Jia said discontentedly .

Label the same to the front


Similarly, , Ruodun had a similar experience on the music platform . By algorithm “ hear ” After she had the habit of paying attention to ancient songs , The recommended songs on the home page are always popular 、 Ancient style song list with similar singing style , Multiple Click “ Change the batch ” It's not going to go away . There are also many songs she has collected 、 Downloaded tracks , Let the original love of ancient songs ruofun thoroughly tired of listening to almost back .

“ In fact, if the platform recommends different styles of ancient songs, it's OK , After all, the ancient style is also subtly lyrical 、 There are many different types of war songs . But the recommended songs are always of the same type , It's easy to get tired of listening too much .” If you express .

Different from the short video platform, traffic is the king of the recommendation method , music APP Mainly according to user preferences to recommend song list . With QQ Music recommendation system RS(QQ Music Recommendation System) For example ,RS Record each user's listening behavior data , Label it , Conduct statistical analysis , Draw a unique user profile .

In addition to the language that can only rely on the music platform 、 singer 、 Genre and other content preferences ; Ranking List 、 song sheet 、 Preference for local songs and other listening scenes ; And the use of APP Outside the time period of listening to music , Social attribute tag is a mature social ecological chain through Tencent , Get the age of the user 、 Gender 、 occupation 、 Location and other information .

Corresponding , The platform will also tag singer information for the music provided ; Characteristics of audio signal ; On demand 、 Downloads 、 Collection volume 、 Share the same heat data ; Rock and roll 、 Sir 、 Blues and other music schools ; Music emotion and playing instrument label .

After drawing portraits for users and music respectively , According to the respective label , Statistical analysis of big data , Creating tag vector between user and song 、 Establishing preference matrix among different users . By matching music attributes with each other , In the huge song database to find users are likely to like the music .

Such a song list recommendation based on user portrait , Although the original intention is based on user habits and preferences , Recommend more user preferred styles , So as to increase the use experience . But in users like ruozi , But there are some “ Do a seemingly clever thing which turns out to be a foolish one instead ”.

“ Except the recommended songs are so similar , What I don't like the most is , It often recommends songs that I've heard 、 Songs that have been collected or even blackened ! Sometimes it is recommended by Netease cloud There will be two repeated songs in the same song list ! No matter how popular those songs are, they don't have to be pushed like this .”

“ Listening to songs is because you want to hear songs you haven't heard before , Hope to find some new good songs . There are always songs that have been heard, and the meaning of listening to new songs is lost , You have to keep skipping , A little bit of a problem . If you want to listen to my favorite songs , Then I'll play my own song list or collect songs ?” If you tell me “ Melting finance and Economics ”.

“ Now I seldom listen to old songs , Too tired . I prefer to search for different styles of songs , How do I feel about changing the algorithm . In this way, you can touch more types of songs , It's not easy to get tired of listening .”

What do you want to hear , Algorithm has the final say.


Although the specific algorithm rules are different , But through the experience of Xiaojia and ruozhu, we can find that , Even today's Internet companies and multimedia platforms have advanced artificial intelligence big data analysis technology , Be able to record and analyze user habits to recommend content . But the needs of users are not unique “ Personalized recommendation ” So single , be based on “ User portrait ” Song list push of label , Instead, let some users pursue different 、 The demand for more differentiated content is increasingly difficult to meet .

On the platform, the content is more diversified , Today, the cost of music and song transmission is lower and more convenient , In some ways , We may have a narrower range of music . For the busy modern people , Self search on music platform or short video platform 、 There are not many times to select the viewing content . Most of them are still open APP after , Choose the recommended songs or movies on the home page of the platform . These contents are just the parts selected after layer by layer analysis of big data algorithm .

Even if users search for the type of song they want to listen to , When searching for a track style you haven't heard before , The algorithm will still be based on user profiles and tags , Find songs that match well , And then according to the traffic and popularity ranking recommendation . We thought it was ourselves “ choice ” Listen to what you want to see , In fact, it is still restricted by algorithm and traffic . So it looks like , In the age of big data , The user's independent choice of listening to music and watching video can be said to be slowly declining .

“ Now, we can hear the tiktok from time to time. .” Xiao Jia said helplessly .“ I wonder if these platforms are interconnected ? How can it be so difficult to listen to music ?”

“ Maybe pop songs are popular everywhere . Like I used to pull black before very red thief will line , But in QQ The recommended song list still appears occasionally . For algorithms , The popularity of music itself is heavier than the weighting of user preferences ?” Ruo Fen is puzzled about this .

Whether it is the algorithm of short video platform or music platform , Video or song itself “ degree of heat ” Will affect the number of times recommended . from 2018 Year begins , Nearly half of the popular songs come from short video platforms , The algorithm of short video platform is most important for traffic .

Tiktok and Kwai, and other short videos and live broadcast are popular. , In recent years, this proportion has been further enlarged , There are short video music charts on many music platforms . The short video platform has more and more discourse power in the music publicity link , The music industry is also being kidnapped by algorithms , Open the Traffic first The age of madness .

Although the short video platform provides another effective way to promote music , But with the help of traffic blessing and platform algorithm leading , Finally can be heard by the majority of users are often some “ Brainwash ”, Or is the flow of fans continue to brush the list “ Hold red ” The song of saliva . The mass circulation of these songs , It affects the aesthetic orientation of the market imperceptibly , Finally form the user aesthetic simplification .

At present, the public's aesthetic orientation towards mainstream music does have obvious characteristics , It can be seen from the singer characteristics and music selections of previous finals of singing variety show . Such as 《 I am a singer 》 Top three , Lin Zhixuan 、 Yang Zongwei 、 Han Lei 、 Deng Zi chess 、 Han Hong 、 Coco Lee 、 Zhang Xinzhe 、Jessie J、 Hua Chenyu 、 Mr Wang 、 Liu Huan 、 greeny 、 love 、 The voices of Xiao Jingteng and others have the characteristics of high pitched or explosive , Most of them are both . But in earlier singing shows like avenue of fame and super girl , The singers who enter the finals have their own characteristics and styles 、 The difference between them is more significant .

The purpose of the algorithm is to help us in the type of music we like , Find more favorite songs . But at the same time, the algorithm determines the order of song recommendation according to the traffic , It also blocks the types of music that other users want to try to touch , Even if users themselves do not realize that they have such needs .

“ Maybe some people find this kind of recommendation very convenient to find their favorite songs , But for me this kind of music is more miscellaneous , There are no fixed types of people who are also particularly concerned with mood , Netease Yunhe QQ Music List recommendation is not particularly accurate .” Music lover Kaiya told me .“ I still miss the random play of shrimp , At least it feels really random , Unlike Netease QQ There are fixed style types . If it wasn't for shrimps, the songs would be less and less , I will continue to use it .”

For example, Xiaojia 、 If different 、 For users like Kaiya , What they really need may not be “ Personalized recommendation ” Brought about by the “ Recommended song list ”, But choose whether to block the freedom of personalized recommendation and the right to choose songs independently .

( Xiao Jia 、 If different 、 Kaiya is a pseudonym )

Melting finance and Economics : New media for city business , Regional economic links , Where industry trends are found .

版权声明
本文为[Yang Hengchang]所创,转载请带上原文链接,感谢
https://cdmana.com/2020/12/20201225081435201n.html

Scroll to Top