Music recommenda tion System - Spotify Collaborative Filtering - - PowerPoint PPT Presentation

Music recommenda tion System - Spotify

Collaborative Filtering and Feedback System

Mithun Madathil 1

Table of contents

 Introduction  Methods of recommendation  Collaborative Filtering in Spotify  Feedback System  Conclusion  References

Mithun Madathil 2

The ideal music recommender

 maximize user‘s satisfaction  Recommend songs to hit top songs of

user‘s favourite list

 Nowadays streaming music provides best

services such as Soundcloud, Deezer, Spotify

Mithun Madathil 3

Spotify

 Uses various ways of recommendation  100 mio. monthly active users with millions

• f songs and playlists

 Three main services for recommendation

and a feedback system

Mithun Madathil 4

Spotify track

Mithun Madathil 5

Spotify track

Mithun Madathil 6

[5]

• 1. Content-based

recommendation

 Without user‘s evaluation or ratings  Uses machine language to acquire

information

 Algorithms: decision trees, neural networks

and vector-based methods

Mithun Madathil 7

• 2. Knowledge-based

recommendation

 Based on demands and preferences of

user

 Predictions decided by functions and

features of objects

Mithun Madathil 8

• 3. Collaborative Filtering - KNN

 Uses K-nearest neighbour (KNN)

technique

 Music taste of users calculates distance

between different users

 Search for neighbour users who share

similar interest in music and recommend content

 Daily life: friend‘s recommendation

Mithun Madathil 9

Categories:

Memory- based Model-based Hybrid Predict items based on previous ratings Uses algorithms and models preferences Combining both models and

• utperforms

them

Mithun Madathil 10

[2]

Collaborative Filtering - Flowchart

Mithun Madathil 11

[1]

Collaborative Filtering - Approach (1)

Mithun Madathil 12

Neighborhood Models: [4] [4] Minimize cost function:

Collaborative Filtering – Approach (2)

Mithun Madathil 13

• 1. Initialize user & item vectors
• 2. Fix item vectors and solve for optimal

user vectors

• 3. Fix user vectors and solve for optimal

item vectors

• 4. Repeat till convergence

[4]

In Spotify: Discover Weekly Playlist

Mithun Madathil 14

[6]

My discover weekly playlist

Mithun Madathil 15

Feedback System

Mithun Madathil 16

Theory of general feedback system [1]

Results in Spotify

Mithun Madathil 17

Frequency of pressing „like“ when users find songs matching their taste [1]

Conclusion – Collaborative Filtering

Mithun Madathil 18

Advantages Disadvantages Evaluates information that is difficult to be analysed Cold-start problem Avoids low accuracy by matching items with neighbourhood users Unusual taste leads to poor recommendations Provides users with not similar recommendations but based

• n taste

Personalization weakened with popular songs recommended Big amount of data needed

Conclusion – feedback system improvements

 Time delay of correcting measures  Requirements, features and development

for every system

 Users moods are not important which

leads into the inaccuracy problem

Mithun Madathil 19

Papers

 [1]:Exploring drawbacks in music

recommendation systems

 [2]:A survey of music recommendation

systems and future perspectives

 [3]:A model-based music

recommendation system for individual users and implicit user groups

 [4]:Collaborative Filtering for implicit

feedbacks

Mithun Madathil 20

Sources

 [5]: https://developer.spotify.com/spotify-

echo-nest-api

 [6]: https://qz.com/571007/the-magic-

that-makes-spotifys-discover-weekly- playlists-so-damn-good

Mithun Madathil 21

Time for your questions!

Mithun Madathil 22