TABLE OF BOARD AND IMAGEFigure 1: Principle Factors of Spotify, Apple and Amazon...8Figure 2: Music Streaming Market Share on 2020 and Mergent Online Company Profiles on 2021...9Figure 3
Overview of technology
Concepts
- When it comes to Spotify, there are three most typical algorithms:
→ These algorithms work together to create a personalized music experience for each Spotify user, ensuring that users discover new music they love while enjoying their favorite artists and genres Spotify's commitment to innovation and machine learning ensures that its algorithms continue to evolve and deliver even better recommendations in the future.
● Recommendation System is the algorithm deploying machine learning (ML) algorithms to recommend new titles for all their users Spotify's recommendation algorithm uses a variety of algorithms to process this data and generate recommendations Some of the key algorithms include:
+ Collaborative filtering: This algorithm identifies users with similar listening habits and recommends songs that those users have enjoyed. It's based on the principle that people with similar tastes tend to like the same music
+ Content-based filtering: This algorithm analyzes the musical characteristics of songs to recommend songs that share similar traits with songs the user has already listened to It's based on the idea that users prefer music with similar sonic elements.
+ Hybrid filtering: This algorithm combines collaborative filtering and content-based filtering to create a more comprehensive recommendation system It takes into account both user preferences and song characteristics to provide a more personalized experience.
● Natural language processing (NLP) is an algorithm that provides the ability to understand text and speech Spotify uses NLP to classify their music By searching the web for any text, Spotify can also find out about a specific song.
● Spotify’s NLP then categorized songs based on the language used to describe them Keywords will be picked out and assigned a weight, which can measure how much a song exhibits a particular emotion This helps spotify’s algorithms identify which songs and artists belong in playlists together, which can then be more easily deployed to the recommendation system.
● Reinforcement learning (Rl) is a system produced based on ML methods to understand goals and respond to data through trial and error during interaction. Spotify uses the RL system to feature songs and artists From there they will be accurate and meaningful to the home page of subscribers.
+ New content is first delivered to subscribers through additional filtering or NLP Thereby, subscribers can easily interact with the song in many different ways (listen to the song once, play the song multiple times or listen to more songs from other artists) or stop playing by pressing Skip the song In all cases, subscribers send information to the RL algorithm about their desired level of success.
+ RL offers users to explore other areas of music through knowing their preferences and typical listening history Expanding the range of music subscribers are listening to will expand the range of music consumed within theSpotify app's catalog – benefiting both artists and the Spotify platform.+ New content is first served to subscribers using collaborative filtering or NLP.The subscriber will then engage with the song on varying levels (listen to the song once, on repeat, listen to more songs by the artist) or disengage by skipping the song In either case, the user is sending information to the algorithm about how successful their prediction was.
Characteristics
Spotify's recommendation system relies on analyzing various characteristics to personalize your music experience These characteristics can be broadly categorized into three main types:
- Listening history: This includes the songs you have listened to, skipped, saved, and added to playlists It provides valuable insight into your musical preferences and listening habits.
- Saved songs and artists: These represent songs and artists you explicitly show interest in, indicating a strong preference.
- Followed playlists: This reflects your interest in specific genres, themes, or curators.
- Demographic information: Although not directly used for recommendations, information like age, gender, and location can be incorporated for broader targeting.
- Audio features: This includes musical attributes like tempo, key, energy level, danceability, and acousticness, allowing for categorization and suggestions based on similar sonic qualities.
- Genre and subgenre: Categorization based on genre and subgenre helps identify songs with similar musical styles and themes.
- Lyrics and artist information: Analyzing lyrics can reveal themes, emotions, and concepts, while artist information provides context about their musical style and influences.
- Release date and popularity: Newer releases and trending songs can be factored in for discovering fresh music or exploring popular trends.
- Time of day: Spotify considers the time of day to recommend music suitable for different activities and moods, like upbeat music for mornings or relaxing tracks for evenings.
- Location: This can be used to suggest music popular in your region or relevant to specific locations.
- Device: Depending on the device you use (phone, laptop, speaker), Spotify might suggest music suitable for different listening environments or activities.
- With all the data Spotify has collected, the NLP algorithm is capable of classifying songs based on the type of language used in their descriptions and similarity to other songs used for the same purpose Artists and songs will be classified based on data and each term has a certain weight assigned to them Similar to collaborative filtering, a vector representation of the song is created and used for the purpose of recommending other similar songs to the user.
+ Natural Language Search: Search for music using natural language queries instead of exact keywords, allowing for more flexibility and understanding of user intent Personalized Recommendations: Generate "Discover Weekly" and
"Release Radar" playlists using NLP to analyze listening history, saved songs, and followed artists.
+ Understanding Lyric and Artist Descriptions: Analyze lyrics and artist descriptions to categorize music by genre, mood, and theme, improving search results and music discovery.
+ Voice Search and Assistant: Interact with Spotify using natural language commands to search for music, control playback, and access features.+ Context-Aware Recommendations: Recommend music based on the time of day, location, and activity level using NLP to understand user context.
+ Generating Captions and Transcripts: Generate captions and transcripts for podcasts and other audio content, making them more accessible and engaging.
+ Automatic Song Metadata Generation: Generate song metadata like genre tags, mood labels, and descriptions based on audio features and lyrics analyzed using NLP.
+ Music Categorization and Classification: Categorize music by genre, subgenre, mood, and theme based on lyrics, artist descriptions, and audio features analyzed with NLP.
+ Identifying Named Entities: Identify and classify named entities like artists, instruments, and locations mentioned in lyrics, enhancing search and organization.
- NLP technologies are continually advancing, driven by machine learning and deep learning techniques These advancements enable computers to handle language- related tasks with increasing accuracy and sophistication.
- The limitation of these collaborative filtering methods is that they rely on explicit or implicit feedback signals to know whether a user likes a playlist or not As a result, they will have difficulty considering other important factors (e.g., the coherence of the song's sound, the context of the music listening session, and the optimal presence of musical item sequences) ) This leads to a mismatch between offline metrics and user satisfaction metrics (which we want to optimize).
- For example, collaborative filtering has the ability to recommend playlists with high ratings but does not classify suitability for users as containing a mixture of adult and children's music This leads to user dissatisfaction Therefore creating a suitable and impressive playlist is a difficult task.
- The field of Reinforcement Learning (RL) can be enhanced without explicit feedback signals Instead, RL can learn through interacting with users on Spotify. Therefore, RL agents will interact and learn to increase user satisfaction in creating their own playlists.
Current trends/ application in business
Spotify has applied the power of data to personalize customer experience. Spotify uses data analysis, user profile mining, and customer experience personalization to launch new features for the paid version
Specifically, with current user data and analysis technology, Spotify implements a strategy including the following 4 main activities:
- Reorganize the music store and optimize the user experience of music playlists
- Build campaigns connecting local artists with users.
● Reorganize the music store and optimize the user experience of music playlists:
To suggest the next song to users and play music automatically, Spotify's algorithm is created based on machine learning: This feature analyzes the songs in a certain playlist and tries to predict what music will come next Spotify's AI has studied millions of user-created playlists to understand what a good music playlist is, then provides suggestions that are most similar to user intent
Optimize the UX-UI of the music playback menu: Is an application service on a technology platform (Service as a software - SaaS), as well as other SaaS software Design that ensures optimal UX/UI is one of Spotify's priorities. Therefore, the location of menus, buttons, control bars, tabs, pop-ups, etc are designed to be reasonable and most user-friendly for the user's experience. Creates a limited skip feature and only listens to music in random mode in the free version
The company has created algorithms to optimize the music suggestions that pop up from your home menu to curated playlists like Discover Weekly.While competitors like Apple Music, Amazon Prime Music and GoogleMusic rely on a combination of paid users and community-generated playlists,
Spotify's key differentiator is its degree of customization and expansion of musical knowledge offered to customers.
Spotify just launched a collection of personalized playlists called Spotify Mixes Built around users' listening preferences, Spotify Mixes starts with each person's favorite songs and continually updates with recommended songs Spotify thinks you'll love.
Spotify has playlists tailored to your area To promote this feature, Spotify uses code on playlists and their ads are designed based on the playlists that have been shared in each region.
To create a difference and attract users to use the paid version, Spotify offers a policy: when traveling abroad (not your home country), Spotify free version will no longer be available You will need to purchase a 14-day music package for your trip.
● Build campaigns connecting local artists with users.
They developed a new strategy: supporting and uplifting local artists This strategy has created a connection not only in territory but also in musical thinking between listeners and musicians Local music ads are listed on public transit,airports, or near popular tourist destinations to increase local music awareness as well as Spotify awareness in that location Additionally, advertisements are also posted around local bars or cafes to help promote the artists to their public.
Challenges and impact to different aspect of business
- The music streaming industry is dominated by large, multinational companies that account for the majority of the market share in this field Specifically, Spotify, Apple Music, Amazon Music, YouTube and Pandora are the five big companies They are rivals and there is always great competition in the global music streaming market. According to the survey, they hold more than 74% of the global music streaming market share as of April 2020 The companies do not compete on price because they all offer basic services at mid-range prices average about 9.99 USD/month (see
Picture 1) A framework of strategies is shown for the three companies Apple, Spotify and Amazon to illustrate how the three companies compete with each other (see Picture 2)
Companies generally compete to gain users through several key factors:
1 Size of the music catalog offered by the music streaming service creates value by appealing to a wide range of listeners;
2 Podcasts provided by the music streaming service on the same app as music create value by added convenience;
3 Personalization of listening experience through personalized playlists;
4 Complimentary products offered by the streaming company to work with their music streaming service;
5 Original content and other exclusive offerings to subscribers;
Figure 1: Principle Factors of Spotify, Apple and Amazon
Spotify uses a more widespread, popular differentiation strategy Spotify's goal is to bring a completely different product to its large customer base through personalized playlists and exclusive podcasts only available on Spotify Therefore, Spotify has the market share lead in the industry (see Picture 1)
Figure 2: Music Streaming Market Share on 2020 and Mergent Online Company Profiles on 2021
The biggest potential threat to Spotify's long-term competitive advantage is a legal change to its data collection policies from users Spotify is heavily dependent on collecting data from its customer groups In addition, recently, Spotify has started collecting data on users' speech to analyze guest metadata to improve more value such as emotional state, gender and voice, mood ( according to Hendler, released 2021). The change in laws restricting user data collection and policies allowing data usage will severely limit Spotify's ability to grow in providing new, more personalized listening experiences for users and would take away this key source of Spotify's competitive advantage.
Businesses are using new applications to rethink their business models and — in some cases — disrupting their industries.
Figure 3: Percentage of Digital Share in Music
We have reviewed the potential impact of these technologys on the music streaming app The music industry had rapid changes from the physical market to the digital market in the past decades The consumers download and stream music online and mobile during the digital dominant market These technologys can increase revenue, attract more consumers The positive impact to different aspect on Spotify only be possible if there are detailed consideration of the industry and careful understanding
- Improved music discovery: By combining these technologies, Spotify recommends music that goes beyond user listening history, introducing them to new artists and genres they might enjoy.
- Personalized experience: Recommendations are tailored to individual preferences, taking into account user history, cultural context, and even mood.
- Increased engagement: Users are more likely to discover and enjoy music they hadn't heard before, leading to longer listening sessions and higher retention rates.
- Enhanced music understanding: NLP and RL help Spotify understand how music is perceived and consumed by its users, contributing to a deeper understanding of music preference and culture.
- Content Management: Recommendation: Identifies trending artists and genres, helping curators create relevant playlists and editorial content NLP anh RL : Analyzes music metadata and user feedback to categorize and label music accurately, enhancing optimize, search and discovery features, ensuring maximum visibility and engagement.
- Marketing and Advertising: Recommendation: Personalized recommendations drive advertising campaigns, NLP and RL: Analyzes user data and cultural trends to identify potential target audiences, maximize reach
Overall, in terms of using recommendation algorithms, NLP and RL will have lasting effects on various aspects of the Spotify app Together, these technologies will create a diverse system that will help Spotify deliver personalized, flexible, and high- quality music services to users around the world These technologies create a more personalized, engaging and efficient platform, benefiting users, artists and the music industry as a whole These technologies create a more personalized, engaging and efficient platform, benefiting users, artists and the music industry as a whole.
Technology implementation in business
Business brief overview
On April 23, 2006, Daniel Ek, the former CTO of Stardoll and Martin Lorentzon, a co-founder of Tradedoubler established Spotify as a Swedish audio streaming and music service provider.[1] According to Ek, the name "Spotify" was mispronounced as a name that Lorentzon had yelled out Eventually, they combined the terms "Spot" and "Identify" to form their company's title Spotify was developed to address the issue of music piracy.Before music streaming services gained popularity, a lot of people downloaded music files illegally This was an increasing challenge for the whole music industry and served as the foundation for Spotify's establishment Daniel and Martin founded Spotify after realizing the enormous potential of music streaming
With its principal office in Stockholm, Sweden, Spotify offers a vast collection of over 100 million songs and five million podcasts from a wide range of record labels and media businesses On October 7, 2008, Spotify's services were made available to the general public (by invitation only) in Scandinavia, the UK, France, and Spain In the UK, Spotify began providing free, restricted access to its services in 2009 [2] For the year 2019, Spotify made a profit for the first time ever Currently, Spotify is accessible in the majority of Europe, Africa, the Americas, Asia, and
Oceania It boasts over 574 million users, with 226 million of those being subscribers across 184 regions
“We envision a cultural platform where professional creators can break free of their medium’s constraints and where everyone can enjoy an immersive artistic experience that enables us to empathize with each other and to feel part of a greater whole.”
“Our mission is to unlock the potential of human creativity – by giving a million creative artists the opportunity to live off their art and billions of fans the opportunity to enjoy and be inspired by it.”
Spotify operates on a "freemium" strategy, similar to other music streaming services This model states that it makes use of a freemium business model, providing an unlimited premium service for a membership price and a basic, limited, ad- supported service for free The service includes a number of features, including social sharing, offline playback, and high-quality audio It is accessible on multiple platforms, including desktop, mobile, and web.
1 Draw a sizable user base by offering a free service.
Users of Spotify's free music streaming service can choose from millions of songs in its catalog Users of the free service must listen to advertisements that help to partially fund the service, and it only offers basic functionality.
2 Convert complimentary users to a high value offering
Spotify has had great success turning free users into paying customers In addition to more features, its premium subscription does away with advertisements 2018 saw 46% of Spotify's users become premium subscribers, who account for 90% of the service's earnings
The longer Spotify can keep users, the more money it can get from them over time— the user's lifetime value, or LTV—increases, similar to any other subscription model.
We refer to this as managing customer turnover Spotify's premium customer turnover rate dropped to a record-low 4.6% in the first half of 2019.
4 Equalize the cost of premium and free
Record labels receive almost fifty-two percent of Spotify's revenue from each stream. Sony, Universal, Warner, and Merlin are the four record labels that own more than 85% of the music that is streamed on Spotify In 2018, Spotify paid out €0.5 billion in royalties to free users and €3.5 billion to premium users, or 74% of total expenses.
5 Use your premium revenue stream to finance the entire amount.
The unique aspect of the freemium business model is that you have to be able to pay for both free and paid users In 2019, Spotify's user base reaches over 248 million, for which royalties are required Of those, 54% listen to music for free, albeit in moderation [3]
SWOT/ Market analysis/ Competitor research
Spotify's advanced algorithmic approach, which examines a tonne of user data to comprehend unique preferences and behaviors, is what powers its personalized recommendations The algorithm can provide highly personalized song recommendations and carefully curated playlists based on each user's taste by analyzing listening habits, playlists, and user-generated content.
2 Social media integration that is seamless
Spotify creates a lively environment for music discovery and sharing by enabling users to share their favorite songs, playlists, and recently played tracks with friends and followers This not only makes it easier for users to discover new music through social networks, but it also promotes a sense of musical community as users converse, trade recommendations, and find out new artists together.
Spotify's data-driven insights offer artists a detailed and insightful understanding of their fan base Through demographic analysis, musicians can learn more about the age, gender, and interests of their fan base and adjust their music and marketing tactics accordingly.
The cross-platform synchronization feature of Spotify improves user flexibility and convenience When a user switches between devices during the day or moves from a smartphone to a laptop, Spotify picks up where they left off, saving them the trouble of looking for the last track or playlist they played.
Spotify has demonstrated its dedication to accessibility by offering improved accessibility features By providing support for screen readers, people with vision impairments can use assistive technologies which speak the text on the screen to navigate and interact with the platform.
Additionally, text resizing selections enable users with vision impairments to change the font size for improved legibility
Both artists and industry insiders have criticized Spotify's artist compensation model, raising issues with the comparatively low royalties that are given to artists. Discussions regarding the equitable allocation of earnings in the music industry and the financial viability of musicians have been sparked by this pay gap.
Users may have to rely on other platforms or services in order to stream live concerts and events or listen to live radio broadcasts due to Spotify's lack of live content Some users may find this fragmentation of content inconvenient if they are used to a single, comprehensive music streaming service.
It is a disadvantage if they want more control over how they listen to music Even though Spotify's algorithms are made to offer tailored recommendations, some users might want more precise control—like the capacity to adjust their preferences or bar particular musicians or genres.
Any technical problems can make it difficult for users to stream music or access their music library, such as server outages or app crashes Because of this dependence, users' overall experience may be impacted by outages or service disruptions.Additionally, if Spotify stops supporting particular devices or operating systems, compatibility problems could occur, limiting users' options and requiring them to upgrade or find other music streaming services
1 Application of voice assistants and smart devices
Spotify can take advantage of the expanding market for smart home technology and satisfy users who want voice-activated interactions by integrating with well-known smart devices Through this integration, users can conveniently and intuitively listen to their favorite tracks, playlists, or podcasts by simply using voice commands.
Spotify has the potential to completely change the way people interact with music by utilizing augmented reality technology Users can explore more multimedia content related to the album and gain a deeper understanding of the musicians' artistic vision by utilizing interactive album covers.
Spotify has an exciting opportunity to promote creativity and create original content that connects with users through enhanced artist collaboration As a result, Spotify can present unique and inventive musical collaborations that blur boundaries and combine genres, establishing the company as a platform that actively promotes and fosters artistic cooperation.
4 Sharing music on social media
The social features of Spotify can be strengthened and user engagement increased with the aid of social music sharing As a result, users can collaborate to curate and add to a shared musical experience, encouraging a sense of community and teamwork.
Growing licensing fees pose a serious threat to Spotify's revenue and financial stability The price of obtaining licenses from record labels and other rights holders has gone up in tandem with the growth in popularity of music streaming.
As Spotify grows and enters new markets, it may encounter various data privacy laws and compliance issues To ensure that user data is protected everywhere, it will need to have strong systems and procedures in place Spotify needs to make sure users have control over their private information and continue to be open and honest about its data handling policies in light of the growing scrutiny surrounding the data practices of tech companies.
How do businesses use technology ?
Recommender System (RS) - Suggestion system, is a form of information filtering system (information filtering) with the function to determine the level of evaluation
(rating) or preference (preference) of the user (user) about a certain product (item). Spotify's Recommender system is a complex system, using machine learning algorithms (machine learning) to determine the user's favorite music listening trends through the analysis of music listening history data, genre, artist, sound attributes of the song from which to suggest the best, most relevant songs to the listener, the, to enhance the user experience
2.3.2 Data synthesis and operation process
We can visualize the implementation process of the Recommender algorithm of Spotify through the following steps:
● At the start of account registration: except for personal information (name, age, gender, etc, ) Users will be asked to provide information about their favorite music, artists or songs they are interested in If the user does not make any choice, the system will skip and continue to recommend the songs that are trending to the user Gradually, the system records user data through their listening habits.
● During using process: Spotify collects user data in various ways:
The system will capture and record the user's music listening history, including the songs that the user has heard, listening time and frequency
- User feedback: Spotify collects user feedback on songs, artists, and playlists. This feedback assesses the level of interest and satisfaction of the user to the song or the audience mentioned This response comes from how the user's listening behavior is represented, for example, the music plays that users regularly listen to, or, searches and visits to the artist or playlist Additionally, it can be visualized through custom charts, comments, and playlists Especially the charts, Spotify has the Top 50 or Viral 50 in the Global or Local range.Thereby, the application will easily capture the taste of music in each region,different regions.
- User localization: Spotify also updates user location information to predict the difference in music taste in each region, what the country is like; at the same time try to suggest similar songs to the group of people in that area This localization makes it easier for Spotify to capture user sentiment, because in addition to the songs users love, the, they also tend to listen to articles belonging to their language, ethnicity, homeland.
- Information from third parties: Spotify may also collect data from third parties, such as advertisers and data providers The application will record the level of discussion of users on social networking platforms such as Google, Youtube, Instagram, , or reaction to advertising campaigns, etc, banner introduction, This data is essential for Spotify to analyze and reorganize its marketing strategies more effectively.
- Capturing customer psychology: In 2021, Spotify launched its customer sentiment analysis feature The app collected data by extracting voice and noise to track user psychology In addition, the app also categorizes the emotional state based on the song data users hear such as intonation, stress, rhythm of the song, etc, it will help to draw conclusions about moods such as: Fun, Anger, Fear, Sadness or Neutral, which suggest the right music to the user
2.3.2.2 Using machine learning algorithms for data analysis:
A range of algorithms used by Spotify, including: Collaborative Filtering, Natural Language Processing and Machine Learning In these three models, Collaborative Filtering analyzes data based on patterns in music listening behavior or user manipulation to predict and recommend appropriate music genres Natural Language Processing will focus on analyzing metadata related to a song and its association with other factors such as singers, playlist name, album name, similar song Finally, the Machine Learning algorithm, this is a machine learning algorithm for training models that predict songs that users will enjoy based on their music listening history and music listening history of other users who have similar “gu” music In addition, Spotify also uses Deep Learning to analyze user behavior through
28 activities such as skipping a song, the length of time a user listens to a particular piece of music and the time they listen to it To build a user profile and make decisions about content recommendations [5]
2.3.2.3 Suggest songs, artists, and playlists:
Spotify uses different suggestion methods to suggest new songs that fit your needs, user preference It includes: hints based on the user's music playing history, hints based on the songs in vogue, etc, based on the correlation between the same songs, based on the playlist that the user creates or is interested in,
● Functions of Spotify's recommendation system:
+ Interest-based recommendation: This is the main function of the Recommender system on Spotify The system will recommend songs, albums, and playlists based on songs that users have heard before If a user is listening to classical music, the system will suggest symphonies, sonatas, or other classical compositions.
+ Trend-based suggestions: This function will suggest songs, albums, and playlists that are popular or loved by other users For example, Fifty Fifty Fifty's song “Cupid” became popular on TikTok, Spotify noticed that this song has the potential that many people will like to include the song in the songs, music list suggested to users Thanks to this, the listening of this song increases and becomes more popular.
+ Explore-based suggestions - This function will suggest songs, albums, and playlists that users may not know but might like The app predicts this through the user's listening behavior such as searches, discovery of playlists, related artists Example as a user favorite post “Today I sad” of Phung Khanh Linh, the system will predict and recommend other songs of Phung Khanh Linh to users.
+ Suggesting new songs next for users: This is a prominent feature of Spotify, make it different from other social networking platforms like Apple Music, Youtube Music This feature integrates AI (Artificial Intelligence) and machine learning systems to analyze data on music listening history, genres, etc, the listener's favorite artist from there suggests suggestions for songs that users will enjoy listening to The system goes through two steps: Building a user model and building a song model
+ Personalize the user's listening experience: If Apple Music Amazon Prime Music and Google Music focus on analyzing the willingness of users to pay for the service and playlists, Spotify is focused on personalizing the listening experience, providing a quality experience for users This algorithm is controlled by an AI called BaRT (“Bandits for Recommendations as Treatment”) Its function is to arrange the main screen in its own ways and suitable for each user including items (shelves), which, the list of tracks according to a certain theme and the order of their appearance in each item (shelves), it also helps you find your favorite music based on your previous music history and introduces new ones so you don't get bored.
+ Discovery Weekly playlist Since this feature was launched in 2015, it is rated as an algorithm that provides the best interest-based song suggestion service. Every week, every Monday, Spotify will bring you Discovery Weekly playlists with 30 different songs selected based on user preferences Different individuals will have different playlists This playlists have brought diverse and rich music parties to users [6]
2.3.4 Detailed description of how to operate
Music recommendation system and user experience optimization is a combination of 3 models:
1 Collaborative Filtering models (combination of filters): this model works by analyzing your behavior and that of others.
2 Natural Language Processing (NLP) models (Natural language processing): this model works by analyzing text (text)
3 Audio models: works by analyzing raw audio files.
2.3.4.1 Collaborative Filtering models (kết hợp các bộ lọc):
Assessment on the technological implementation
Overall outcomes
Now we go to next part, that overcome result that Spotify use in particular to enhance their althogrim to improve their target customer’s experiences:
Personalized recommendation is what makes Spotify's brand value - one of the giants of online music streaming in the world with over 100 million users and more than 35 million songs In order to make personalised recommendations, Spotify employs a unique combination of machine and deep learning models include some special technology that Recommendation, natural language processing, and reinforcement learning , coupled with human curation, to process and interpret musical attributes, and behavioural (listening) user data in order to offer recommendations and personalisation.
Spotify uses a hybrid method for recommendation: a combination of techniques such as [8]:
● Collaborative filtering (that is a type of recommendation algorithm that makes predictions about one user’s preferences based on a collection of data from many users Across Spotify’s 406 million subscribers, numerous similarities occur between the types of music certain clusters of people listen to that Spotify was able to use these playlists to recommend titles to subscribers with similar music taste ).
● Contextual and content-based recommendation ( or Reinforcement learning
- Data pertaining to how the music sounds (acoustic), who listens to it (user), when and where they listen to it (contextual), what do they do with it, i.e., playlisting (behavioral) to bring accurate and meaningful songs and artists to their subscribers’ home pages
- (New content is first served to subscribers using collaborative filtering or NLP. The subscriber will then engage with the song on varying levels (listen to the song once, on repeat, listen to more songs by the artist) or disengage by skipping the song In either case, the user is sending information to the algorithm about how successful their prediction was.)
● Moreover, data is combined with natural language processing of what people say about the music on thousands of sites (semantic music knowledge) Taken together this combines to form a listener’s ‘taste profile and is reflected back to users in the form of recommendations.
- In particular, There is relatively little research on the impact of Spotify on users' music consumption and discovery But recent studies have shown that this is related to our own experience and that of the friends we have discussed how Spotify has affected them I have discovered some new artists and albums that I currently "highly rate" through Spotify, especially Discover Weekly And among them, some also tend to buy the actual copy of the album.
- One typical example is Tam 9 by the singer My Tam - a pop album in 2017. I'm not a fan of My Tam’s singer, nor am I interested in her songs before, but Discover Weekly introduced me to some songs in this album around 2020 It might be because of the algorithmic elements described above, especially the fact that I "like" songs with "ballad pop vocals" and "deep powerful female vocals" This album has both, and it quickly became one of my all-time favorite albums And that's probably the most authentic explanation of how Spotify's recommendation system works in reality.
3.1.3 The benefits that algorithms bring to users
This combination of acoustic, user, contextual, behavioral and semantic music knowledge data allows for tailored personalisation and algorithmic
40 curation and recommendation For instance, Spotify offers a number of personalized playlists under their “Made For You” categories, such as with status people like me, Spotify suggest for me some suite playlist that: Pops, Pops Hits 2000s-2023 and more…
● For example, seasonal music we have: Christmas Hits, 100 greatest christmas songs ever… for christmas, or Tet for Tet holidays of Vietnamese…
Figure 10: Playlist of seasonal music
- It created a specialist and signature of spotify that was learned and followed by other music streaming platforms But Spotify always is “the leader” in this field.
- Daily Mixes are six continuous playlists that create a mix of liked artists and genres. algorithmically collates the new music from a users top artists (liked, following, listened to) To achieve this goal thanks to special althogrim that we study above.
- With their powerful analytic and algorithmic capabilities music streaming services have the potential to enable, inform and prescribe the cultural practices and experiences of and with music Taste’ and ‘mood’ are shifted from subjective human traits and experiences to commercial and algorithmic data points for recommendation. Within this infrastructure, previously subjective human processes of curation and discovery are increasingly mediated by algorithmic systems.
- So, both objectively and empirically, I tend to think that the Impact on End-Users with Spotify is a net positive It's been shown to increasingly expose listeners to music they find "highly valued".
- But: Keep in mind these recommendations are coming from an opaque algorithm from an organization with a profit motive While useful, we have little way of knowing what goes into it More on this later.
Pros and cons
- Spotify’s algorithm helps users discover new music that matches their tastes and moods, based on their listening history, preferences, and feedback.
- Spotify’s algorithm also helps artists reach new audiences and potential fans, by exposing their music to users who are likely to enjoy it.
- Spotify’s algorithm is constantly evolving and improving, using various sources of data and methods to refine its recommendations and adapt to changing trends and user behavior from being capable of using data for making more personalized experiences and engaging users into the service.
- In another way, we totally concern about that computational systems such as Spotify cannot capture or emulate the inherently human and subjective experiences of music consumption, discovery and taste
- With a modern diversity althogrim apply in Spotify It is called the “‘big mood machine’”, whereby the company uses emotional surveillance for profit There arose concerns that when “we turn culture into data” it affects content, consumption and taste.
- The increasing dependence on algorithms and automated curation to personalize experiences and discover music on behalf of listeners presents not only a deviation from the music industry and technologies of the past ( that personalism depend because every individual is different)
- Another fears that, the playlist may not be very attractive due to the diversity in the genres because of the way Personalized motivation recommendation.
- While all recommendations are then closely related to the users’ preferences and could receive potentially high evaluations That can lead to althogrim
42 influenced the diversity of the recommended playlist For instance, a person may listen to different genres during varying activities like working and sporting But this can suit people in other fields So this can affect the overall user experience of Spotify's algorithm
- Some fears that through its extensive personalisation and recommendation features, Spotify may trap users in a feedback loop, listening to the same or homogenous music, resulting in what has described as a ‘taste tautology’ that Spotify can navigate the commercially controlled promotional space of the in- house curated Spotify playlist strongly affected by corporate monopolies to establish customer consumption for profit [9]
- Some argue that music recommender systems make that that these ways of seeing are heavily influenced by the consumer categories that are defined and demanded by advertisers Behavioral listening data is aggregated and leveraged for music recommendation (as well as for third party advertising Taken together, the automated curation and recommendation features on music streaming services are programmed with the commercial imperation
- Spotify’s algorithm may create a filter bubble, where users only listen to music that is similar to what they already like, and miss out on other genres, styles, and artists that they might enjoy.
- It’s a big problem in our data century Our research says that there is a positive outcome from the Spotify algorithm However, the algorithm is opaque and proprietary, like most algorithms - including Google Search And considering its massive impact on consumers, it demands criticism and investigation, on two fronts: 1) is Spotify allowing external influence and 2) how is the personal data being used?
- Spotify’s algorithm may also favor popular and mainstream music over niche and independent music, creating a power imbalance and reducing the diversity and quality of music available on the platform.
- Spotify’s algorithm may also influence the way artists create music, by incentivizing them to follow certain formulas, trends, and metrics that appeal to the algorithm, rather than their own artistic vision and expression
Proposal Improvement
Spotify uses machine learning to create value and competitive advantage by providing personalized and relevant recommendations to its users There are many ways to improve its algorithm recommendation, natural language processing and reinforcement learning, depending on the specific goals and challenges some possible ideas of our group to improve this recommendation system includes:
● For algorithm recommendation, one way to improve it is to use more diverse and rich sources of data, such as user feedback, social media, contextual information, and external knowledge bases, to capture the user’s preferences, moods, and intents more accurately Another way is to use more advanced and flexible models, such as deep neural networks, graph neural networks, and generative models, to learn complex and nonlinear patterns and relationships from the data.
+ Improve long-term prediction: User preferences evolve over time The algorithm could benefit from better long-term prediction capabilities to keep recommendations relevant.
● For natural language processing, one way to improve it is to use more pre- trained and fine-tuned language models, such as BERT, GPT-3, and T5, to leverage the large-scale and high-quality linguistic knowledge learned from massive corpora, and to adapt them to specific NLP tasks, such as sentiment analysis, topic modeling, and summarization34 Another way is to use more multilingual and cross-lingual models, such as XML-R, mBERT, and MUSE, to handle the diversity and complexity of natural languages, and to enable cross-language transfer and alignment
+ Integrate with user feedback: Allow users to provide feedback on recommendations, not just by skipping or liking songs, but also through direct feedback on NLP-based interpretations.
+ For reinforcement learning, one way to improve it is to use more efficient and robust exploration strategies, such as intrinsic motivation, curiosity, and meta-learning, to encourage the agent to explore the environment and discover novel and rewarding states and actions56. Another way is to use more hierarchical and modular architectures, such as options, skills, and subtasks, to decompose the problem into simpler and more manageable components, and to enable transfer and reuse of learned policies.
+ Some issues regarding the personalised althogrim of Spotify implementation that could be improved in the future The possibility of adding weights to the song parameters present in the reward system would be one of them Currently, all parameters weigh the same, meaning they all have the same importance Maybe adding different weights to each one of these parameters could result in a better extrapolation of the user’s preferences In turn, the Server-Side Application could return a better set of recommendations This, however, could only be done by testing the different weight applications on test users and waiting for their feedback to know which combination would be best.
+ Federated learning: This technique allows training models on user devices while preserving user privacy It could be used to personalize recommendations without requiring users to share their entire listening history with Spotify Because in nowaday security on the web or platform is a “hot issue” when important information of users can be lost whenever So platforms have to protect customers, avoid getting unsafe on the internet and getting your data hacked.
+ Perform transparent reports on Spotify's personalization algorithm so that specialized agencies can understand, control, manage, and ensure security and user rights:
● In the Google world, there is much research, reporting, sleuthing, and experimentation done against that algorithm From my research, this has happened, albeit on a very limited scale, with Spotify as well And that's completely necessary for businesses (artists) who use the platform, but also for consumers to have confidence that they are not being unduly influenced.
Final conclusion
● Overall, through objective research and empirical observation, I think it is fair say that the net result of Spotify and its Discover Weekly algorithm for listeners has been very positive For musicians (ignoring for a moment the overall music marketplace) it has also been positive, though less conclusively so But ALL of this comes with concerns about transparency.
● Specifically, the Discovery Weekly or daily mix suggestion algorithm appears to do a good job of recommending music users like and tends to promote music discovery and genre expansion in ways that we've never seen before on a massive scale Not everyone has that "one friend" who is a tastemaker, after all.
● However: The overall market share and audience influence that Spotify has coupled with the lack of transparency is deeply concerning and demands constant critique and investigation.
● While not necessarily as important as, say, governance or legislation, music, and the environment needed to create it is a critical part of culture and the human condition Completing ceding that it into the hands of an opaque, monolithic entity is a recipe for disaster.
● And never forget that your data is running it You are a stakeholder, and you have influence.
[2] Chayanika Goswami, Manisha Mishra, (2023) Spotify Success Story: How it Brings
Music for Everyone?, retrived 20/12/2023, from https://startuptalky.com/spotify_success_story/
[3] Zhang, B., Kreitz, G., Isaksson, M., Ubillos, J., Urdaneta, G., Pouwelse, J A., &
Epema, D (2013, April) Understanding user behavior in spotify In 2013
Proceedings IEEE INFOCOM (pp 220-224) IEEE.
[4] Derr, T., Georg, S., & Heiler, C (2021) SWOT-Analyse und TOWS- Normstrategien In Die disruptive Innovation durch Streamingdienste: Eine strategische Analyse der Marktführer Netflix und Spotify (pp 17-24) Wiesbaden:
[5] Ben Wilson (2020) A CRITICAL LOOK AT SPOTIFY'S ALGORITHMS, OR:
HOW I LEARNED TO STOP WORRYING AND LOVE THE ALGORITHM, retrieved 14/12/2023, from https://via.studio/journal/a-critical-look-at-spotify-algorithms
[6] Adam Cash (2019) Is Spotify Legal? Find the Answer Here, 14/12/2023, from https://www.aimersoft.com/spotify/is-spotify-legal.html
[7] Tanca (2022) Bí quyết Spotify ứng dụng dữ liệu khách hàng để cá nhân hóa trải nghiệm, retrived 14/12/2023, from https://tanca.io/blog/bi-quyet-spotify-ung-dung-du-lieu-khach-hang-de-ca-nhan-hoa- trai-nghiem
[8] maingocanh (2020) Spotify's Discover Weekly: Cách máy học tìm thấy những bài hát bạn sẽ yêu thích, retrived 14/12/2023, from https://spiderum.com/bai-dang/Spotifys-Discover-Weekly-Cach-may-hoc-tim-thay- nhung-bai-hat-ban-se-yeu-thich-puh
[9] Millecamp, M., Htun, N N., Jin, Y., & Verbert, K (2018, July) Controlling spotify recommendations: effects of personal characteristics on music recommender user interfaces In Proceedings of the 26th Conference on user modeling, adaptation and personalization (pp 101-109).