MPPP

Music Popularity Prediction

Targets¶

• target — Predict Popularity with acoustic (low-level) feature only(?.

Popularity¶

• Popularity can be described in various perspectives.¹
  

• social factors sometimes play an important role in determining whether a song would be popular. (Experimental study of inequality and unpredictability in an artificial cultural market)

• either use stream or rank as criterion.

  • currently use stream

Data Sources¶

• track's statistics collected from Spotify: kworb.net

• audio source:

  1. Spotfiy API preview (only 30s) (seems enough)
  2. MIDI

Methodology¶

• Train with low-level features only(? > Generic low-level features of songs, like the mel-spetrogram used in [4] and also throughout this work, may suffer from the “semantic gap” [15] and cannot lead to an accurate prediction model for a high-level concept such as hotness. ²

Mixed Region ¶

• Tracks in single region are not sufficient for training. Thus, train with multi-region tracks may be more resonable.
• However, preference and population differ from regions. Region information have to be included.
  1. region cls_token
  2. region embeding

Model¶

• Transformer

• targets

  • predict stream-time curve can yield any metrices described in ¹

MP2 (Music Propularity Prediction)¶

• predict sumation and debut as an regression problem

• fine-tuned MP2 model (Encoder) can be used in MP3 model.

• Encoder (AST)

• loss: MSE

MP3 (Music Propularity Period Prediction)¶

• with decoder
• Predict stream-time curve

Potential Augmentation¶

• use different preview clip of same tracks
  • typically, there are more than one track for a song on Spotify (since they published by different album), and they may not have same preview clip.
• Specaugment
• Rolling
• Random Gain
• pitch
• timbre

Q 2 A¶

• seems no comparsion (people use different datasets)

Q 2 findout¶

• ways to treat null values in streams

• same song different version (live version, studio version, covered version...)

• different TA (country, generation, ...)

  • significant difference

• instrumental? (without singer)

• Spotify youtube popularity relation?

• quality of audio

Misc.¶

• Positional Encoding (sum or concat?)
  • https://github.com/tensorflow/tensor2tensor/issues/1591

Observation¶

• Xmax-related songs get and only get popular near Xmas
  • All I Want for Christmas Is You
  • Last Christmas
  • Underneath the Tree

KKBox Researches¶

Revisiting the problem of audio-based hit song prediction using convolutional neural networks¶

Hit song prediction for pop music by siamese CNN with ranking loss¶

A_Multimodal_End-to-End_Deep_Learning_Architecture_for_Music_Popularity_Prediction¶

• audio and lyrics

Predicting Music Popularity Using Music Charts (Sep. 2019)¶

• target: whether a song is featured in Spotify's "Top 50" charts.

Strategies¶

• uses social network data.

• predict the success of a song by acoustic information of previous successful songs.

• use past data from charts to predict whether a song will be featured in the same chart in the future.

Recycle Bin¶

Track Popularity Dataset (seems unuseful)¶

• Musical track popularity mining dataset: Extension & experimentation,’’ N

http://mir.ilsp.gr/track_popularity.html

Refs¶