improving how different organizations carry out their daily duties
Developments in technology have been essential in improving how different organizations carry out their daily duties. Unlike in the past, most organizations in the present age are embracing the online realm and therefore basing their operations on cloud platforms. This preference can be attributed to the fact that online platforms provide organizations with an easier means of reaching out to a wider community using fewer resources as compared to traditional brick and motor platforms. With an increasing number of clients, there are various implications. Apart from positive implications such as increased profit returns and higher market shares, companies are also faced with the challenge of having significantly large volumes of data containing important insights regarding client characteristics. Organizations rely on such information to make important business decisions which contribute to the competence of the company in the respective field.
Large volumes of data, however, pose challenges in terms of the analytics that has to be conducted on them. To gain insights, such companies must have state of the art analytical systems to effectively handle the generated information and subsequently make proper inferences from them. Traditionally, methods such as custom spreadsheets and other database management systems have been employed to handle such problems. However, the frequency and volume of the data have rendered such approaches ineffective. New techniques such as machine learning have therefore been incorporated into the business models of different investments to aid in processing data and subsequently making important business decisions. This paper, therefore, provides an overview of how machine learning, in general, has been employed by Spotify as a recommendation system.
Spotify is a globally known company that offers music streaming services. Since its inception in 2008, the company has grown to have close to 180 million active users and an additional 83 million subscribers (FINANCIAL TIMES, 2018). The popularity of the company can be attributed to the fact that its online-based services are available to people throughout the world with the only requirement being the availability of stable internet connection. Spotify has however risen into prominence due to its incorporation of predictive methodology whose driver engine is machine learning (HPAC, 2017). The predictive technology used by Spotify is significant in the sense that it has helped the company to gain their customer’s trust and loyalty by presenting content to them in a manner that is appealing and enjoyable (IBM Developer, 2017).
The use of machine learning and subsequently artificial intelligence in Spotify can be attributed to the primary reason that the platform hosts millions of artistic works from different artists globally. Therefore, the constant process of searching through the database, done by the users implies that specific patterns exist for specific individuals. Spotify then leverages on such patterns to come up with recommendations for the users at the end of the week. To better understand the importance of this approach, it is necessary to think about some of the new upcoming artists whom most of the listeners have not listened to yet. While users continuously look for a given type of music, they can’t explore all the available options in their list. Therefore, this implies that new artists will not be easily identified by users. However, based on their search options, Spotify’s machine learning engine tracks down the pattern of the users for an entire week and thereafter presents them with a list of 30 recommended songs which are of the same type as what the user has been listening to. This recommendation is provided with the hope that the Spotify user will enjoy this new suggested songs.
As simple as it sounds, the machine learning model uses a wide variety of techniques. For instance, the first technique used by the organization is called collaborative filtering. This approach works based on suggesting certain artists to new listeners based on some correlation between the listener being targeted and another peer listener. This model is thought of as having being influenced by an algorithm that deduces the likelihood of one listener enjoying songs by an artist A as a result of another listener constantly listening to that artist. Similarly, artists continuously listened to by listener B can be recommended to listener C with a given degree of certainty (“Spotify talks playlists, skip rates and NF’s nordic-fuelled success,” 2017). When looked at from a common perspective, implementing this algorithm may seem non-trivial. However, the complexity of the resulting matrix is what makes the use of artificial intelligence effective in the basic operation of Spotify. The existence of millions of records implies that the engineers at the organization have sufficient data to come up with complex matrices which can be used to generate relationships among the users, subsequently making the collaborative filtering approach successful.
Audio analysis is another common technique used in the recommendation system employed by the company. Using some highly sophisticated patterns, the various songs are broken down into distinct patterns based on aspects such as tempo, beat, the pitch of the notes, instrument types, sounds commonly used and the kind and trend of lyrics used. Using this newly generated features, it is then possible to come up with a model that calculates the probability that an individual is going to enjoy a certain kind of music that will be proposed to them by the recommendation system. The significance of employing this approach is that it helps in providing recommendations to users based on music listened to other users which may also possess characteristics similar to what has been identified from the newly generated features by the audio analysis tool (D’Amico, 2020).
The use of natural language processing is also key to the recommendation system. Spotify uses natural language processing to check out online platforms to also aid in determining how appropriate a given recommendation is to its users. By checking the online platforms and scraping information and insights from aspects such as reviews, Spotify can determine how listeners perceive given songs. This perceptions, based on aspects such as sentimental analysis, can then be used to create a general score of whether or not given users are going to enjoy the recommendations provided to them by their weekly discovery recommendation (D’Amico, 2020).
Finally, Spotify also has a mitigation plan against users who lend their account details to their friends. By taking into consideration the fact that individuals may easily share their login details, Spotify has been able to come up with mechanisms of ensuring that they do not provide recommendations to original account owners based on the songs listened to by people who borrowed their accounts. To actualize this aspect, Spotify’s machine learning algorithms tend to overlook any sharp changes in listening habits that may be witnessed in an individual’s account. However, for such changes to be ignored, their duration must be short, an indication that indeed the individual using the account has borrowed it from someone else (Pasick, 2015).
The combination of these techniques has subsequently led to the recommendation system used by Spotify to be effective. The company owes its success to this technique since clients do not have to search for new music manually. Through the weekly suggestions, Spotify ensures that its customers can pleasantly receive their favourite music. Evidence of the successful implementation of the recommendation system can be seen through the fact that the subscribers of the company have grown by 8 million. This growth subsequently led to the share price of the company increased by a value of approximately 25% in three months. This growth in share price was realized after the company was listed on the New York Stock Exchange in April 2018 (Spotify, 2018).
The resulting success of Spotify’s recommendation system has influenced many more researchers to further experiment with the approaches used by the organization. Pichl et al reported in their paper how they combined Spotify and Twitter data to generate public datasets that are also up to date to aid in music recommendation. The approach followed in the paper included streaming tweets using the Twitter API. These were obtained from the periods between July 2011. By 2014, the total number of tweets collected was 90 million. These were based on keywords such as now playing, listen to and listening to. The data collection method used by the authors was unique from that used by other crawling methods in the sense that the downloading of the tweets were based on URLs that led to Spotify’s music streaming service. From the collected tweets, the authors then went ahead to develop a recommendation system similar to that of Spotify (Pichl, 2014). However, it was noted that with an increasing number of recommendations, the system’s baseline recommendation system became increasingly limited. To overcome this limitation, the authors suggested two solutions. These included modifying their datasets for enrichment purposes using other available information that could provide more information based on the context. Another measure that was introduced to help in overcoming the limitation was incorporating a hybrid recommendation system that put into use any new context obtained from the target information collected from additional insights in the downloaded tweets (Pichl, 2014).
Despite the exemplary performance of the recommendation system brought about by Spotify’s machine learning algorithms, there are various shortcomings which have been noted to exist in the organizational framework of the company. For instance, it has been found that the collaborative filtering approach has drawbacks when it comes to recommending songs. Some types of music are less likely to be recommended as compared to others. In a study, it was realized that indie songs are among some of the music genres which are less likely to be recommended by the system. While various reasons have been proposed to explain this disparity, the most realistic reason for the non-inclusion of these kinds of music is that Spotify is yet to get into a contract with artists producing less popular music such as indie music. Without these binding contracts, it is therefore not possible for such music to be made available in Spotify. Other shortcomings which have not been discussed in this context have also been identified in Spotify’s recommendation system. For instance, in the Radio function of the software, it was found that the music that was recommended by the radio was significantly not congruent to the preference of the user (Ding and Liu, 2015). As a result, it was found that most users tend to avoid using the radio functionality of the software. The shortcoming of the radio function extends to the fact that disliking or liking a given song is not beneficial in resetting the user’s preferences. The incapability of the like and dislike functionalities in the Spotify system can further be thought of to be a shortcoming of the feedback system of the software. As much as the machine learning algorithm has proven to be effective in collecting patterns from the users, this flaw reveals how ineffective the model is in collecting trends from the responses provided by the clients (Ding and Liu, 2015).
Despite the challenges outlined above, the recommendation system of the software has still proven to be effective for the company. It is important to note that the challenges outlined in this discussion are based on empirical studies which nevertheless relied heavily on subjective data. Nevertheless, such information should be treated with concern since it is only through customer feedback that the management of Spotify will be able to identify its shortcomings and subsequently come up with mitigation strategies to address the drawbacks. However, this discussion has shown how machine learning, in general, can be applied in the business environment to significantly boost the performance of an organization. From the number of active users enrolled in Spotify to the increased shares as a result of the listing of the company in the New York Stock Exchange, it is evident that once applied properly, artificial intelligence can be ideal leverage for companies dealing with large data. Adequate research, however, has to be undertaken to ensure that whatever approaches are taken to leverage on the use of machine learning are appropriate to help the company achieve its intended objectives.
References
D’Amico, C. (2020). Spotify’s secret weapon. Music Business Journal – Berklee College of Music. https://www.thembj.org/2014/10/spotifys-secret-weapon/
Ding, Y., & Liu, C. (2015). Exploring drawbacks in music recommender systems: the Spotify case.
FINANCIAL TIMES. (2018, July 26). Spotify gains 8m paid subscribers aided by Latin America growth. Financial Times. https://www.ft.com/content/16c0c91c-90cd-11e8-bb8f-a6a2f7bca546
HPAC, Music Recommendation System Spotify: http://hpac.rwth- aachen.de/teaching/sem-mus- 17/Reports/Madathil.pdf HPAC.
IBM Developer. (2017, October 30). YouTube. https://www.youtube.com/watch? time_continue=166&v=n5gCQWLXJcw
Pasick, A. (2015, December 21). The magic that makes Spotify’s discover weekly playlists so damn good. Quartz. https://qz.com/571007/the-magic-that-makes-spotifys-discover- weekly-playlists-so-damn-good/
Pichl, M., Zangerle, E., & Specht, G. (2014). Combining Spotify and Twitter Data for Generating a Recent and Public Dataset for Music Recommendation. In Grundlagen von Datenbanken (pp. 35-40).
Spotify talks playlists, skip rates and NF’s nordic-fuelled success. (2017, November 29). Music Ally Is A Knowledge Company. https://musically.com/2017/11/29/spotify-playlists-skip- rates-nf/
Spotify. (2020). Spotify — Spotify Technology S.A. Announces financial results for second quarter of 2018. Spotify — IR Home. https://investors.spotify.com/financials/press-release- %20details/2018/Spotify-Technology-SA-Announces-Financial-Results-for-Second- Quarter-2018/default.aspx