Proceedings of the Second International Conference on Electronics and Renewable Systems (ICEARS-2023).

DVD Part Number: CFP23AV8-DVD; ISBN: 979-8-3503-4663-3

Music recommendation system using hybrid

approach
Nakka Venkata Durga Malleswari Kuchipudi Gayatri Katepalli Yaswanth Sai Kumar
Computer Science and Engineering Computer Science and Engineering Computer Science and Engineering
Koneru Lakshmaiah Educational Koneru Lakshmaiah Educational Koneru Lakshmaiah Educational
Foundation Foundation Foundation
Guntur, India Guntur, India Guntur, India
malleswarinakka2345@gmail.com gayatrichowdary216@gmail.com yaswanth000002@gmail.com
Nekkanti Likhita Padmanaban K Debnath Bhattacharyya
Computer Science and Engineering Computer Science and Engineering Computer Science and Engineering
Koneru Lakshmaiah Educational Koneru Lakshmaiah Educational Koneru Lakshmaiah Educational
Foundation Foundation Foundation
Guntur, India Guntur, India Guntur, India
likhitasalur@gmail.com padmanaban.k@yahoo.com debnathb@gmail.com

Abstract— Due to the expansion of the World Wide Web a lot of songs with ease [9]. It is integral for these companies to
music data is now accessible online. It becomes a requirement to maintain their customer base and generate revenue by
create a suggestion service in addition to conducting searches for
offering subscription services [9]. Nonetheless, the task of
user’s anticipated musical things. A Music Recommendation
System using a hybrid approach is presented in this paper, creating playlists from a long list of music can be challenging
which provides a personalized music recommendation service for many users, thus leading them to simply select songs
by using a combination of several factors. A hybrid approach is randomly or use the recommendation feature [15]. As such,
a powerful tool in the development of music recognition, as it the development of a successful personalized music
can classify music into different genres based on the audio signal recommendation technique has become a crucial part of
beats. This allows for easier categorization and understanding sustaining customer loyalty. To prevent members from
of the similarities between various music genres. Moreover, this
approach could be further developed to improve the accuracy of
switching over to another service, approaches to
automatic music classification. By using the hybrid approach, recommending music have been developed in the last ten
music recognition research can become more efficient and years.
accurate, resulting in an advancement of the field. In a Many young people seek digital music as mainstream
personalized music recommendation System, to integrate the consumer content. Music recommendation technology has
results of the hybrid algorithm with the log files, we used the been paid close attention to by domestic and foreign scholars
recommended method of collective filtering (CF). To make in recent years. Song Amazon Prime Music and Spotify are
music recommendations to the user, this algorithm was created,
two well-known personalized music providers.
based on their individual preferences. The CF recommendation
algorithm takes as input the user's favorite songs and the log
Personalization, which is often not able to satisfy users'
files Every user who uses a system for tailored music demands, is often missing [12].
recommendations has their history recorded in the log file. The Traditional search engines are not as efficient as desired,
system that provides tailored music recommendations takes the making it hard to satisfy users. This led to the emergence of
user's history from the log file and suggests songs for each genre. recommendation systems, which can provide tailored
Giving the user with positive human-computer interaction and services for each user [14]. Such systems, for instance, the
top-notch recommendations is the primary goal of music music recommendation system that is based on a user's
recommendation. And to achieve this, the user interface allows preferences, can boost the user's confidence in the website by
users to play songs and simultaneously recommends songs based offering the most relevant music [8],[13].
on the current song being played. The audio features and lyrics
scores of each pair of songs are fed into the algorithm for
Locating the flaws of the single-mode recommendation
analysis. The suggestions are stored using firebase storage, system is not a challenging task; content-based and
which ensures that the user gets reliable recommendations. collaborative filtering technologies come with a scarcity of
data and the cold start dilemma, while association rules
recommendation technology is hard to implement due to
Keywords—Personalized music recommendation, personalization-related low-level difficulties, which all lead
Collaborative filtering, Root Mean Squared Error (RMSE), Mean to a decrease in the efficiency of recommendations.
Absolute Error (MAE). Collaborative filtering recommendations have a more
comprehensive range of use than other methods, and their
I. INTRODUCTION recommendation accuracy is relatively high. Sparsity and
cold start issues [1],[4],[11],[7]. However, require user
In today's era of cloud computing, the usage of online recommendations that rely on content evaluation information
music streaming services such as Spotify, Pandora, and and not collaborative filtering; the range of application is
YouTube have become increasingly popular [9]. Big data significantly smaller.
technology has enabled music listeners to access limitless

979-8-3503-4664-0/23/$31.00 ©2023 IEEE 975

 Proceedings of the Second International Conference on Electronics and Renewable Systems (ICEARS-2023).
DVD Part Number: CFP23AV8-DVD; ISBN: 979-8-3503-4663-3

The standard procedure is to employ the music sound carried

by the base gene for the music suggestion system. Then a
collection of songs was used to produce a featured record, and
this feature record was established to obtain the song's feature
vector and the final analysis of each vector's degree of
similarity. Finally, the degree of resemblance is assessed, and
songs that are wider than the predetermined width are
included in the final list of recommendations.

B. Key Technologies
1.Collaborative Filtering Recommendation
Users may be grouped according to their interests through
collaborative filtering technology, which allows for user
sorting into several groups. since each user group's members
share the same interests, we may target people by referring to
other users' choices for items or information [2]. The
Fig 1: Types of recommender system
collaborative filtering algorithm analyses target user interest
based on a comparison of the target user's habits and
The hybrid mode advice addresses the drawbacks of these
preference information with that of other users' preference
two recommendations and can be employed in a different
information, to determine the similarity and the target user's
weight approach. When the user's assessment information is
recent set, based on a recent set of user interests or
relatively little, the weight based on the music gene's
preferences, and to predict the target user's interest or
recommendation is applied. When the amount of assessment
preference [3].
information provided by the user is high, the weight based on
The following are the fundamental beginning points for
collaborative filtering recommendation is applied [6]. This
collaborative filtering:
system takes extensive use of the benefits of collaborative
filtering and music gene suggestion. Finally, utilizing the
 All users' interests are determined by their scores
enhanced recommendation mode, a customized music
for various things.
recommendation system based on hybrid mode is created,
and the system's function requirements, performance  Different user groups are split according to user
requirements, and data needs are detailed. The outcomes of interests.
the experiments reveal that the system successfully improves  All users' evaluations of various things will be
[5]. consistent with those of the users in the user group
that shares their interests.
II. RELATED WORK
The collaborative filtering system's fundamental
A. Music Personalized Recommendation System components are listed above. From these beginning
The standard three elements of the music-specific principles, it is clear that the collaborative filtering
recommendation system are the user preference model, algorithm's most important considerations are how to
description of the music resource, and recommendation identify user similarity and how to choose a reference user
algorithm, as shown in the figure. list based on it.

2 Music Genes
Some of the qualities of a piece of music include the presence
of fixed genes, such as melody, pace, and so on. These might
be thought of as the internal DNA of music since they are
symbols that people can use to comprehend words and
feelings. Various types of music have different social
characteristics because of the emergence of society, and this
is again a function of social genes. It contains information
such as artist names, song lyrics, composer names, performer
Fig 2: Music Personalized Recommendation System genders, language, area, kind, etc. The only stable
The traditional approach of music suggestion. One of them is characteristic and feel are appropriate for the ambiguous
the description of the arrangement and administration of the characteristics established by circumstances. The following
musical resources. We may create a database of musical overall structure of the music gene may be derived as a figure
characteristics and provide useful input data for music if all the attributes of music are combined.
recommendation models by specifying the complexity and
abstraction level of various levels. As a result, the foundation
and essential component of music suggestion is music
resources. The standard procedure is to employ the music
sound carried by the base gene for the music suggestion
system.

979-8-3503-4664-0/23/$31.00 ©2023 IEEE 976

 Proceedings of the Second International Conference on Electronics and Renewable Systems (ICEARS-2023).
DVD Part Number: CFP23AV8-DVD; ISBN: 979-8-3503-4663-3

however, are unknown and dependent on the user's

interpretation. Diverse users have different opinions
even about the same song. Users who are unfamiliar with the
system may also use various labels to represent them for the
same understanding. As a result, the user's annotation
information may be utilized to determine their preferences for
various musical qualities over the whole free gene. Such a
preference may be a reflection of the user's propensity for
placing greater focus on music when looking for music. The
user may create a preference table for various Music Free
Gene features.
Fig 3: Types of Music Genes

Each gene's categorization information and associated gene

preference information must be entered into the database for
the system's music gene. Both the database's model and the
amount of data it contains will keep growing. Because they
do not need to understand how to construct interfaces, system
developers for hierarchical systems may directly call the
interfaces of all levels to service themselves, which is also Different users frequently have similar interests. The
practical for the system's ongoing upgrading. The following interested users are "neighbors" with one another. The
layers make up the music recommendation system's three- attention of the user's neighbors can aid the system in
layer architecture: recommending them when they are exposed to music, we
 Presentation layer: This layer is shown using the utilize the function sin (x, y) to indicate how closely users x
boundary layer for user observation, which also and y have similar interests. the user. It is assumed that a
accepts less "interactive operational behavior data." piece of music is user-interest-conforming if it appears in the
The system has functions for both users and album the user has created or collected. The calculation of
administrators, such as landing, registering, playing, user interest similarity is then:
folding and stopping, cutting songs, prior music,
singer marking, and adding or removing music.
 Business logic layer: It oversees the system's
primary business's data transport and logical
processing. The system's business data mostly
consists of details on its users and administrators,
music libraries, users who are now listening to
songs, and so forth. In order to provide consumers
with recommendation services, logical processing
adopts the above business data and makes use of
a collaborative filtering algorithm and hybrid III. HYBRID RECOMMENDATION ALGORITHM DESIGN
recommendation algorithm based on the music
gene recommendation algorithm. Most of the time, the music recommendation system uses a
 Data access layer: the process that is primarily in hybrid recommendation algorithm that draws on both
charge of database access. The system must carry collaborative filtering and music gene content [7],[10].
out database activities such as addition, deletion, Before generating a list of recommendations using a
querying music information, user listening collaborative filtering algorithm, the system first retrieves the
preferences, recording accounts, and other user self-rating data created by user action or automatically fills
information automatically, as well as updating data based on user preferences from the database. While
data automatically while in operation. doing so, the system also calculates and proposes a different
User Behavior Analysis-The foundation and heart of list of recommendations based on a gene-based
customized suggestions is the user model. Customized advice recommendation algorithm after extracting the gene
can only be better fulfilled by effectively capturing and information of users' preferred songs from the database. The
comprehending the user's interests. Based on the data two suggested lists each take a different tack in a different
obtained from the Pandora Radio website, we extracted users' situation before delivering the suggested outcomes. The
information on music annotations by crawling their own parallel hybrid algorithms are combined based on weight and
album production and collecting. The user's choice for each can be changed as needed. The system's general algorithm
characteristic is examined in light of the external gene data flow diagram.
from the music genome, and a user preference vector is
established. Music has a fixed gene that is practically
exclusive to music. As a result, the user's preference for
various types of music may be calculated using their level of
attentiveness. The qualities of the free gene of music,

979-8-3503-4664-0/23/$31.00 ©2023 IEEE 977

 Proceedings of the Second International Conference on Electronics and Renewable Systems (ICEARS-2023).
DVD Part Number: CFP23AV8-DVD; ISBN: 979-8-3503-4663-3

comment set is q1, q2...., qn. The calculation for RMSE,

which is used to assess the volatility of prediction outcomes,
is:

Users chose between 20 and 200 songs for each of the three
algorithms mentioned above in the experiment, and each song
was broken into 20 portions. The experimental findings are
provided in table 1 together with the computed MAE and
RMSE for each interval. The hybrid filter-based
recommendation system has the minimum MAE and RMSE,
according to the experimental results. As a result, the hybrid
mode-based recommendation algorithm will be used to create
the music recommendation system in this work.

Fig 4: Hybrid Recommendation Algorithm Design

VI.CURRENT CHALLENGES AND VISION
Detailed hybrid suggestion flowchart Two
recommendation algorithms are presently used by the music Although music recommendation systems have advanced
recommendation system. The system only needs to be greatly in recent years, there are still many problems and
assigned to the list of recommendations produced by the new areas for research. Following are some of the current issues
algorithm according to the weight, and this hybrid structure and future goals in the study of music recommendation
is advantageous to the advancement of the system's systems:
algorithm. Subsequent new recommendation algorithms that  Personalization is one of the biggest obstacles facing
may be added include association rule-based researchers studying music recommendation
recommendation algorithms and knowledge-based systems. It is difficult to develop a system that can
recommendation algorithms. precisely predict a user's musical choices because
each person has different musical tastes. The
IV. DATABASE development of techniques that can accurately
On the Pandora Radio network, you may download the test record user preferences and behaviour is currently
data set for free. 40,000 MSD songs and 101000 independent the subject of research.
users are included in the data collection, making a total of  Context-awareness: A user's musical preferences
50000000 people who recorded the number of music plays. can alter based on the time of day, where they are,
Ten participants were chosen at random for the experiment, or what they are doing. Research is concentrating on
and each participant received two sets of the whole song. The developing more efficient ways to gather and exploit
user's attention-getting songs make up the first collection of context information. Context-aware recommender
U’s, while the user's non-attention-getting songs make up the systems that can adapt to various situations are being
second collection of V. Then, 7 songs are chosen at random created.
from set U to serve as the user interest analysis's data. After  Diversity: The "filter bubble" effect, when
deleting the user's annotation data from the first piece of recommender systems only suggest music that is
music chosen, it is added to group V, as well as the collection well-known and popular, is a common problem with
serves as the test corpus. recommender systems. Users may as a result miss
out on hearing fresh, varied music. Research focuses
V. PERFORMANCE E VALUATION on creating better strategies for diversifying
We choose the top 20 music scores as the suggested results in suggestions and introducing people to fresh and
order to adapt to the smaller display displays of mobile distinctive music.
devices and mobile devices, and we assess the system based  Explainability: It can be difficult to comprehend
on how the extracted music is positioned in the final results. why a certain song or artist was recommended when
The proper results are marked after the first N results in the using the state-of-the-art techniques used in music
returning result have been examined. Based on the user's recommender systems, such as deep learning
annotation, the assessment algorithm decides which of the N models. Researchers are looking into how to
outcomes is valid before making each suggestion. The improve the openness and comprehension of music
assessment criteria we employ are RMSE (Root Mean recommendation systems.
Squared Error) and MAE (Mean Absolute Error). The quality  Integration with other domains: To produce a
of recommendation outcomes is assessed using MAE, while more individualized and varied user experience,
the variation of suggested results is assessed using RMSE. music recommender systems can be combined with
The computation of MAE is as follows: Assuming the user other domains like movies, books, or podcasts.
comment set is p1, p2…., pn and its corresponding real Multi-domain recommender systems, which can
offer more thorough recommendations, are being

979-8-3503-4664-0/23/$31.00 ©2023 IEEE 978

 Proceedings of the Second International Conference on Electronics and Renewable Systems (ICEARS-2023).
DVD Part Number: CFP23AV8-DVD; ISBN: 979-8-3503-4663-3

researched. [5] Singh, J., 2020, December. Collaborative filtering-based hybrid music
recommendation system. In 2020 3rd International Conference on
All in all, the vision of music recommender system research Intelligent Sustainable Systems (ICISS) (pp. 186-190). IEEE.
is to build systems that are further personalized, versatile, [6] Shah, K., Salunke, A., Dongare, S. and Antala, K., 2017, March.
and adaptable to various contexts, while also remaining Recommender systems: An overview of different approaches to
transparent and simple to understand. recommendations. In 2017 International Conference on Innovations in
Information, Embedded and Communication Systems (ICIIECS) (pp.
VII.CONCLUSION 1-4). IEEE.
[7] Çano, E. and Morisio, M., 2017. Hybrid recommender systems: A
In this research, we describe two well-known music genre systematic literature review. Intelligent Data Analysis, 21(6), pp.1487-
recommender approaches—collaborative modifying and 1524.
content-based model—as well as a fundamental metadata- [8] Shalini, S.K., Jaichandran, R., Leelavathy, S., Raviraghul, R., Ranjitha,
based model. Even though they have achieved remarkable J. and Saravanakumar, N., 2021. Facial Emotion Based Music
Recommendation System using computer vision and machine learning
progress, a music recommendation system is established techiniques. Turkish journal of computer and mathematics
using an algorithm based on two hybrid models after an education, 12(2), pp.912-917.
experimental comparison of collaborative filtering [9] Smt. Namitha S J , “Music Recommendation System” (IJERT), 07,
suggestions. It combines the advantages of both approaches. July-2019.
Due to the subjective character of the music and the problems [10] Garipelly, Vyshnavi, Padma Teja Adusumalli, and Priyanka Singh.
with the prior techniques, its complexity has not yet been "Travel Recommendation System Using Content and Collaborative
Filtering-A Hybrid Approach." In 2021 12th International Conference
completely investigated. Two human-centered approaches on Computing Communication and Networking Technologies
are thus suggested. The standard of exhortation was typically (ICCCNT), pp. 1-4. IEEE, 2021.
raised by the emotion-based model and the context-based [11] Soleymani, Mohammad, Anna Aljanaki, Frans Wiering, and Remco C.
model by considering emotive and social variables. This Veltkamp. "Content-based music recommendation using underlying
research is still in its early stages, though. music preference structure." In 2015 IEEE international conference on
multimedia and expo (ICME), pp. 1-6. IEEE, 2015.
VIII.REFERENCES [12] Kim, Kunsu, Donghoon Lee, Tae-Bok Yoon, and Jee-Hyong Lee. "A
music recommendation system based on personal preference analysis."
[1] P. Prabhu, N. Anbazhagan “Fuzzy Filter Based Clustering Approach In 2008 First International Conference on the Applications of Digital
for Recommender System” Vol. 41 No 4 February 2014, pp.477-486. Information and Web Technologies (ICADIWT), pp. 102-106. IEEE,
[2] He, Ran, Guo Hui Song, and Shu Tao Sun. "Network Literature 2008.
Recommendation Technology Based on User-Based Collaborative [13] Cho, Soo-Hyun, Young-Hak Kim, and Jae-Bum Park. "Music
Filter." Advanced Materials Research. Vol. 926. Trans Tech recommendation system for public places based on sensor
Publications Ltd, 2014. network." IJCSNS 7, no. 8 (2007): 172.
[3] Ziming ZE. A context-aware personalized commodity [14] Schedl, Markus. "Towards personalizing classical music
recommendation for ubiquitous commerce based on click streams and recommendations." In 2015 IEEE International Conference on Data
collaborative filtering. Journal of Computational Information Systems. Mining Workshop (ICDMW), pp. 1366-1367. IEEE, 2015.
2012;8(8):3489-96.
[15] Mogale, Hope, and M. B. Esiefarienrhe. "Optimizing Recommendation
[4] Chen, R., Hua, Q., Chang, Y.S., Wang, B., Zhang, L. and Kong, X., Algorithms Using Self-Similarity Matrices for Music Streaming
2018. “A survey of collaborative filtering-based recommender Services." In 2021 International Conference on Artificial Intelligence,
systems: From traditional methods to hybrid methods based on social Big Data, Computing and Data Communication Systems (icABCD), pp.
networks”. IEEE Access, 6, pp.64301-64320. 1-4. IEEE, 2021.

979-8-3503-4664-0/23/$31.00 ©2023 IEEE 979