You are on page 1of 1

2019 International Conference on Advances in ICT for Emerging Regions (ICTer)

A Personalized Music Recommendation System


based on User Moods
Champika H.P.D. Wishwanath #1, Supunmali Ahangama #2
#
Faculty of Information Technology, University of Moratuwa
Bandaranayake Mawatha, Moratuwa, Sri Lanka
1 2
champikawishwanath@gmail.com, supunmali@uom.lk

Keywords— Facebook, Lyrics, Mood, Music, Recommendation,


Social Media.

ABSTRACT
Considering millions of songs available in online streaming
services, it is difficult to identify the most suitable song for a
specific user. Most of the music recommendation systems are
based on user ratings and acoustic features of the songs. Those
systems are unable to address the cold start problem and rating
diversity. Furthermore, song preferences will be altering based
on the current mood of the users. If these problems were not
addressed, then these online services will fail to achieve user
satisfaction. To cope with those problems, this paper proposes a
novel music recommendation approach that utilizes social media
content such as posts, comments, interactions, etc. and Fig. 1 High level design of the approach
recommend them with the most relevant songs to relax their
mind considering the current mood (happy, sad, calm and C. Methodology
angry). The proposed recommendation system was created considering
A. Introduction both collaborative filtering and content-based filtering approaches.
Based on the real time current mood updates of the users, songs
The primary motivation to research this area is to provide an were predicted. One of the main challenges in the recommendation
easier way for music listeners to queue up songs based on their model was to address the songs that have been never heard by a user.
current moods. Thus, various alternative approaches to recommend To handle this implicit function with matrix factorization was used,
mood fixing songs that enrich initial music preference of user based where, unseen items were treated as negative with a low confidence.
on artist and songs will be evaluated considering various factors. As implicit model, the inbuilt python ALS function was used
In the proposed recommendation system, there will be a music considering the execution time.
player and a playlist of recommended songs. Moreover, the user’s For any recommendation system, another problem is providing a
listening patterns will be captured to improve the recommendations recommendation when there is a new user. In order to face this
further (learning process). problem, the proposed system used Facebook user profile data
including liked artists. With this option the system can generate a
B. Personalized music recommendation personalized playlist based on mood for a user. Based on the
Social media mining would be the most appropriate and possible evaluation results KD tree with nearest neighbor search algorithm
way to detect the current mood. Among them, Facebook has the was used.
most powerful global network due to the biggest user base with
various types of engagements via words, links, images, and videos. D. Future work
Nowadays, Facebook is a lifestyle of people. Therefore, through As future work to improve the results of the recommendation
Facebook user profiles, data like public posts (image and text posts), system, it is important to develop a context aware recommendation
status, tags, comments, etc. were collected in this study. As current system. For example, need to consider the time of the day as user
moods happy, sad, calm and angry were considered. Songs were preferences changes with the time of the day and based on the
classified into those four (4) moods according to musical features weather user moods will change. Through social media, it is
extracted from lyrics, melody and vocal of the song as illustrated in important to gather more details about the user personalities, thus, it
Fig. 1. Thus, based on the current mood of the user, the most will improve the recommendations further. In addition, the results
appropriate mood fixing song to the situation will be recommended. can be improved through other online streaming services like
Moreover, a list of songs that are related to the condition of the user YouTube history and online playlist logs.
would be listed.
To generate the artist list, a corpus of artists was used. The E. Conclusion
proposed system is accessing Last.fm service through its public API. The proposed recommendation system will be considering various
For each artist, the name, a picture, the title of the most popular factors in developing the personalized recommendation system.
tracks, their play count and a set of tags that describe that artist were Songs were categorized based on the acoustic features of the songs
accessed from in Last.fm. Subsequently, these artists related data to types of user moods and the current mood of the users were
were mapped with the user preferred artist list identified through the captured through the image and textual posts in Facebook.
recommendation system. To identify the user preferences, user’s Subsequently, based on the user preferences, current mood, and
music playing histories or sequences were examined. These modes song types classified based on moods, songs were recommended to
of capturing user preferences will be explored in future. the users.

978-1-7281-5156-4/19/$31.00 ©2019 IEEE

Authorized licensed use limited to: University of Liverpool. Downloaded on July 26,2020 at 16:13:13 UTC from IEEE Xplore. Restrictions apply.

You might also like